Cohomology of the Free Loop Spae
of a Projetive Spae
Ph.d. Thesis
Mia Hauge Dollerup
Thesis advisor: Marel Bökstedt
July 1, 2009
Department of Mathematis
Aarhus University
1 Projetive spae and geodesis 8
1.1 The quaternions . . . 8
1.2 Spaesof geodesis . . . 8
1.3 Fibrationsinvolvingspaes of geodesis . . . 12
1.4 Homotopy orbits of spaes of geodesis . . . 14
2 Cohomology of spaes of geodesis in
HP r
17 2.1 The parametrizedgeodesis . . . 172.2 The unparametrized geodesis . . . 21
2.3 Equivariantohomologyof spaes of geodesis . . . 30
3
K
-theory of spaes of geodesis inCP r
35 3.1 The unparametrized geodesis . . . 353.2 Equivariant
K
-theory of spaes of geodesis . . . . . . . . . . 384 The free loop spae and Morse theory 47 4.1 The negative bundle . . . 49
4.2 The power map . . . 51
4.3 The Morse theory spetral sequene . . . 52
5
S 1
-equivariant ohomology ofLHP r
58 5.1 The Morse spetral sequenes . . . 585.2 The MainTheorem . . . 65
6
S 1
-equivariantK
-theory ofLCP r
75 6.1 The rst dierential. . . 766.2 The MainTheorem for
r > 1
. . . . . . . . . . . . . . . . . . . 796.3 The MainTheorem for
r = 1
. . . . . . . . . . . . . . . . . . . 84Notation 91
Referenes 92
Introdution
The free loopspae
LX
of a spaeX
is the spae of ontinuous maps fromS 1
toX
. The irle groupS 1
ats onLX
by rotation, and we study thespaeofhomotopy orbits,
LX hS 1 = ES 1 × S 1 LX
,sometimesalled theBorelonstrution. The main method for understanding this spae will be Morse
theory on the energy funtional, whih to a losed urve assoiates its en-
ergy. This version of Morse theory has been studied by W. Klingenberg in
[Klilngenberg1℄. Asonewouldexpet,theritialpointsofthisfuntionalare
the losed geodesis of
X
, soknowing those willbe animportantingredientinunderstanding
LX hS 1
viaMorse theory.Inthispaperwestudy
LX hS 1
forapartiularspae,namelytheprojetivespae
X = FP r
, whereF = C
orF = H
. The goal is to determine theohomology of
LHP r hS 1
and the omplexK
-theory ofLCP r hS 1
. This isalled
S 1
-equivariantohomology(orK
-theory)ofLFP r
. In general,wegeta map
ES 1 × S 1 LX −→ BS 1
byprojetionontherstfator. Foraohomologytheory
h ∗
,wethereforegeta map
h ∗ (BS 1 ) −→ h ∗ (LX hS 1 )
, soh ∗ (LX hS 1 )
beomes ah ∗ (BS 1 )
-module.The methods of Morse theory require the use of Thom isomorphism, whih
destroystheprodutstruture,soweannothopetoalulate
h ∗ (LFP r hS 1 )
asa ring. But the
h ∗ (BS 1 )
-module struture is preserved by the Morse theorymahinery, so the aim is to alulate
h ∗ (LFP r hS 1 )
as anh ∗ (BS 1 )
-module,where
h ∗
iseither singularohomologyH ∗
or omplexK
-theoryK ∗
.We will now outline our main results. For
X = HP r
, we study theohomology with
F p
-oeients ofLX hS 1
, whereF p = Z/pZ
, and obtain aomplete desription as an
H ∗ (BS 1 ; F p ) = F p [u]
-module:Theorem 1. As a graded
H ∗ (BS 1 ; F p ) = F p [u]
-module,H ∗ (LHP r hS 1 ; F p )
isisomorphi to
F p [u] ⊕ M
2k ∈IF
F p [u]f 2k ⊕ M
2k ∈IF
F p [u]f 2k − 1 ⊕ M
2k ∈IT
(F p [u]/ h u i ) t 2k − 1 .
Here the lower index denotes the degree of the generator, and the index sets
IF
andIT
are knowndisjointsubsets of{ (4r + 2)i + 4j | 0 ≤ j ≤ r, i ≥ 0 }
.In partiular, there is at most one generator in eah degree.
For
X = CP r
, we study the omplexK
-theory ofLCP r hS 1
, and obtainTheorem 2. As a
K ∗ (BS 1 ) = Z[[t]]
-module,K 0 (LCP r hS 1 ) = K 0 (BS 1 ) = Z[[t]] .
As an abelian group,
K 1 (LCP r hS 1 )
is torsion-free.This isone of the rst alulationsof
K (LM hS 1 )
for anon-trivial mani- foldM
. TheresultisquitesurprisingwhenomparedtoH ∗ (LCP r hS 1 )
,whihhas a lotof torsion aording to[Bökstedt-Ottosen ℄.
Unfortunately, we have not been able to determine
K 1 (LCP r hS 1 )
as aK ∗ (BS 1 )
-module. As a partial resultin this diretion, we haveTheorem 3. There is a spetralsequene of
K ∗ (BS 1 ) = Z[[t]]
-modules on-verging strongly to
K ∗ (LCP r hS 1 )
, whih hasE 1
page,E 1 0,j =
Z[[t]] ⊗ Z Z[h]/ h h r i ,
j even;0,
j odd.E 1 n,j =
Z[[t]] (n) ⊗ R(S 1 ) Z[x, y ]/ h Q r , Q r+1 i ,
j odd;0,
j even.The rst dierential
d 1
is given byd 1 (p(t) ⊗ h j ) = p(t) ⊗ (x j − y j )
, wherep(t) ∈ Z[[t]]
.Theorem2statesthat
K 0 (LCP r hS 1 )
is(almost)trivial,whileK 1 (LCP r hS 1 )
is free abelian. This is rather similar to the well-known ase of
K 0 (BG)
asthe ompletion of the representation ring
R(G)
for a ompat Lie groupG
,while
K 1 (BG) = 0
. This is a lassial result of M. Atiyah. One an alsoompare to e.g. [Freed-Hopkins-Teleman℄, who nd
K τ ∗ (LBG)
as the om-pletion of ertain representations of the loopgroup
LG
, although it shouldbe remarked that they onsider
K
-theory twisted by a ohomologylassτ
,and not
S 1
-equivariantK
-theory as we do. Still,this promptsthe followingConjeture. The exists a representation theory type group, suh that
K 1 (LCP r hS 1 )
isa ompletion of this group.The outline of this paper is as follows: The paper onsists of two main
parts,eahdividedinthreesetions. Therst setionofeahparttreatsthe
general theory needed and investigates the relevant spaes and strutures,
while the next two setions are more omputational and deal, respetively,
with the ohomologyfor
F = H
, and theK
-theory forF = C
.Setion 1 investigates
FP r
and its geodesis, obtaining some useful -brations. We onsider both the spae of parametrized and unparametrized
geodesis;the latter beingthe quotient of the former under the ation of
S 1
by rotation.
Setion 2 alulates the ohomology of the above spaes using Serre's
spetral sequene for the brations found in setion 1. We then turn to
S 1
-equivariant ohomology of the spae of parametrized geodesis, via two brationsandthe non-equivariantohomologyresultsfromthepreviousse-Setion 3 obtains similar results for
K
-theory. We use the Atiyah-Hirzebruh spetral sequene along with the known ohomology results for
CP r
to determine theK
-theory of the spae of unparametrized geodesis.The
S 1
-equivariantK
-theory is determined using the same brations as forohomology,butthemethodisdierent,employingtheresultofAtiyahabout
K
-theory of lassifying spaes.Setion 4 studies of the free loop spae,
LFP r hS 1
. First we explain theworkings of Morse theory in this setting, then we apply this to
LFP r
andLFP r hS 1
toget the so-alledMorse spetral sequene.Setion 5 is dediated to proving Theorem 1. The method is losely
baseduponasimilaralulationbyM.BökstedtandI.Ottosenintheirpaper
StringCohomologyGroupsofComplexProjetiveSpaes,[Bökstedt-Ottosen℄.
We extrat a lotof informationabout the Morse spetral sequene, its size,
itsdierentials,andtherelationbetweentheequivariantandnon-equivariant
ase. All this information is brought together to prove the Main Theorem
for ohomology, Theorem 1 above. But even then, it is neessary to turn
to other soures of information to omplete the proof. One is loalization,
the other is omparison with the Serre spetral sequene alsoonverging to
H ∗ (LHP r hS 1 )
.Setion6isdediatedtoprovingTheorem2. Themethodsherearequite
dierent, relying on the fat that the Morse spetral sequene in Theorem
3 has a rather speial onguration, whih implies that all its non-trivial
dierentials start from the zeroth olumn. A very important point is the
alulation of the rst dierential
d 1
. The entral idea is then to twist therotationationof
S 1
withapositiveinteger,whihgivesnew Morse spetralsequenes related to the standard one. This gives enough information to
prove Theorem
2
.For the reader's onveniene, we have assembled a table of notation at
the end of this doument.
Aknowledgements. Finally,it is apleasure to thank my advisor, Marel
Bökstedt, forhishelp throughinnumerable fruitfuldisussions,whihadded
many new insights and ideas to this projet. Also, I would like to thank
Jørgen Tornehave for his time and valuable input when standing in as my
advisor for one year.
1 Projetive spae and geodesis
1.1 The quaternions
I start by introduing the quaternions,
H
, as an assoiative algebra of realdimension
4
,generated by1, i, j, k
with the followingmultipliation rules:i 2 = j 2 = k 2 = − 1, ij = − ji = k, jk = − kj = i, ki = − ik = j.
Itshouldbestressed,eventhoughitisobviousfromthe aboverelations,that
H
is not ommutative. If one wants to be onrete, one an realizeH
as asubalgebra of
M 2 (C)
generated overR
by (inthe matrix entries,i = √
− 1
):i =
i 0 0 − i
, j =
0 1
− 1 0
, k =
0 i i 0
.
Itisstraightforward tohekthe abovemultipliationrules. Similartoom-
plexonjugation, there is an
R
-linear map, alsoalled onjugation,H −→ ∗ H
z = x 0 + x 1 i + x 2 j + x 3 k 7→ z ∗ = x 0 − x 1 i − x 2 j − x 3 k,
satisfying the usual rule
(zw) ∗ = w ∗ z ∗
. In the matrix desription, this is preisely the usual∗
-operation of taking the onjugate transpose. This anbeusedtodeneaninnerprodut
h z, w i H = w ∗ z
,whoserealpartistheusualinner produt on
R 4
. Noting thath z, z i H ∈ R
we an then dene a norm| z | = p
h z, z i H
. This satises| zw | = | z | | w |
and| z ∗ | = | z |
. The unit spherein
H
is usually denotedSp(1) = { z ∈ H | | z | = 1 }
, and this is anoniallyidentied with
S 3
. Finally we note that ifz 6 = 0
thenz
is invertible thisis most easily seen by using the matrix desription, whih gives an expliit
inverse, and heking thatthis belongs to
H
.Weantakethediretprodutof
H
withitselftoformH r
. Theoperationsh· , ·i H
and|·|
fromH
areextendedtoH r
intheusualway: Forz = (z 1 , . . . , z r )
and
w = (w 1 , . . . , w r )
, we seth z, w i H =
X r j=1
h z j , w j i H , | z | = q
| z 1 | 2 + . . . + | z r | 2 .
1.2 Spaes of geodesis
Let
F
denoteeitherC
orH
. ToeasethenotationwedenotetheunitsphereinF
byS(F)
. WedenetheprojetivespaeFP r
astheset ofall1
-dimensionalF
-subspaeszF
ofF r+1
, forz ∈ F r+1
. We dene the projetion mapπ : F r+1 \ { 0 } −→ FP r
(1)z = (z 0 , . . . , z r ) 7→ [z 0 , . . . , z r ] = zF,
so
π(z) = zF
is the subspae spanned byz
. Note that forF = H
it is im-portantthatwespeify whihside wemultiplyon; Ihavehosen tomultiply
from the right. We give
FP r
the quotient topology fromπ
. To show thatFP r
is a smooth manifold of real dimension2r
(resp.4r
) forF = C
(resp.F = H
), we display the expliit hartsh j : U j = { [z 0 , . . . , z r ] ∈ FP r | z j 6 = 0 } −→ F r , h j ([z 0 , . . . , z r ]) = (z 0 z j − 1 , . . . , z \ j z j − 1 , . . . , z r z j − 1 ),
where the hat denotes omission;the harts have inverses
h − j 1 (w 1 , . . . , w r ) = [w 1 , . . . , 1, . . . , w r ],
with the
1
atthej
thplae.Example 1.1. We willshow
HP 1
is dieomorphi toS 4
. This an be seenby stereographi projetion. Think of
S 4 ⊆ R 5 = R × H
with north polep + = (1, 0)
and south polep − = ( − 1, 0)
. Stereographi projetion are the mapsψ ± : S 4 \ { p ± } −→ H,
whih takes a point
(t, z )
inS 4
to the intersetion of the line through(t, z)
and
p ±
with0 × H
. This iseasily omputed:ψ + (t, z) = z
1 − t , ψ − (t, z) = z 1 + t ,
and arelearly smoothmaps. Now wewanttoompose
ψ +
andψ −
withtheh − j 1
toget two maps toHP 1
. When we dothis, wewould like the two mapstoagree when
t ∈ ] − 1, 1[
. To ahieve this, wereplaeψ +
with itsonjugateψ + ∗ (t, z) = 1 z − ∗ t
. Doing this, we get maps,S 4 \ { p + } ψ
∗
−→ + H h
− 1
−→ 0 HP 1 , S 4 \ { p − } −→ ψ − H h
− 1
−→ 1 HP 1 ,
given by
(t, z) 7→
1, z ∗
1 + t
, (t, z) 7→
z 1 − t , 1
.
By multiplying the rst expression from the right by
z
1 − t
and using that1 = | (t, z) | = t 2 + | z | 2 = t 2 + z ∗ z
, we see that these two maps agree whent ∈ ] − 1, 1[
, so they ombine toa dieomorphismS 4 −→ HP 1
.We an modify the projetion map
π
in (1)toa mapπ : S(F r+1 ) −→ FP r
where
S(F r+1 ) ⊆ F r+1
is the unit sphere. This an be used to desribe thetangent bundle of
FP r
. Speially forz ∈ S(F r+1 )
there is anF
-linearisometry,
t z : (zF) ⊥ ⊆ T z S(F r+1 ) −→ π ∗ T π(z) FP r ,
where
(zF) ⊥ = { w ∈ F r+1 | h w, z i F = 0 }
. This map satisest zλ (wλ) = t z (w)
forλ ∈ S(F).
(2)The above properties of
FP r
are rather elementary, and the reader ansee e.g. [Madsen-Tornehave℄ Chapter 14for proofs of the results in the ase
of
CP r
.Consider the Riemannian metri on
FP r
given by the real part of theinner produt on
F r+1
. This is the standard metri onFP r
, and we willuse a metri
g
whih is a salar multiple of this metri. Take the uniqueonnetion on
T (FP r )
ompatible with this metri, alled the Levi-Civita onnetion. We now deneG(r) = G(FP r )
as the spae of parametrized, simple, losed geodesisf : [0, 1] −→ FP r
with respet to this onnetion.The salar determining
g
is speied by requiring that suh a geodesi haslength 1 with respet to
g
. Note that every geodesi inFP r
is losed: Thegroup of
F
-orthogonal matries (U (r + 1)
orSp(r + 1)
, respetively) ats transitively onHP r
, so it is onlyneessary to hek it for one geodesi, e.g.on
FP 1 ⊆ FP r
,andsineCP 1 ∼ = S 2
andHP 1 ∼ = S 4
, allgeodesis onFP 1
areknown to be losed.
We also onsider the set of
n
times iterated geodesisG n (r)
for everyinteger
n ≥ 1
, whose elementsγ : [0, 1] −→ FP r
are given byγ(t) = f(nt)
forsome
f ∈ G(r)
, wherewe makethe obviousidentiationof the intervals[j − 1, j]
with[0, 1]
forj = 2, . . . , n
. There isanationonG n (r)
byS 1
givenby rotation;expliitly,
S 1 × G n (r) −→ G n (r) (e 2πiθ , f (t)) 7→ f (t − θ).
Wean twistthe rotationationon
G(r)
byanintegern ≥ 1
,and wedenotethe resulting
S 1
-spaeG(r) (n)
:S 1 × G(r) (n) −→ G(r) (n)
(3)(e 2πiθ , f(t)) 7→ f (t − nθ).
This ation is the rotation ation preomposed with the
n
th power mapP n : S 1 −→ S 1
,P n (z) = z n
in omplex notation. ThenG n (r)
andG(r) (n)
are isomorphias
S 1
-spaesvia theobviousmapG(r) (n) −→ G n (r)
given byf (t) 7→ f(nt)
, so from now on, we will hiey useG(r) (n)
instead ofG n (r)
.We also onsider the quotient
∆(r) = S 1 \ G(r)
under the rotation ation,whih is the spae of oriented, unparametrized, simple, losed geodesis on
FP r
.We now want to get a more onrete desription of
G(r)
and∆(r)
, fol-lowing [Bökstedt-Ottosen ℄, 2. Let
V 2 = V 2 (F r+1 )
be the Stiefel manifold ofF
-orthonormal2-frames inF r+1
,soV 2 =
(v, w) ∈ F r+1 × F r+1 | k v k = k w k = 1, h v, w i F = 0 ,
and let
P V 2
be the quotient manifold by the right diagonalS(F)
ation,(v, w) ∗ z = (vz, wz)
. OnV 2
we have a left ation ofS 1
by rotation by anangle
θ
: Forθ ∈ R
, the ation isw v
7→ R(θ) w v
, where
R(θ) =
cos(θ) − sin(θ) sin(θ) cos(θ)
For eah
n ∈ N
, we an dene an ation ofS 1
onP V 2
, and we denote theresulting
S 1
-spae byP V 2 (n)
:S 1 × P V 2 (n) −→ P V 2 (n) ; e 2πiθ ∗ [x, y] = [R(nπθ)(x, y)].
This gives awell-dened
S 1
-ationonP V 2
, beause we multiplythe matrixR
onthe left,whileP V 2 = V 2 /
diagS(F)
, wherewemultiplyontheright. Wean now make an
S 1
-equivariant dieomorphismϕ 1 : P V 2 (n) −→ G(r) (n)
(4)[x, y] 7→ π ◦ c(x, y)
where
π : S(F r+1 ) −→ FP r
is the projetion, andc(x, y )
isthe simple losedgeodesi startingat
x
in diretiony
; expliitly,c(x, y)(t) = cos(πt)x + sin(πt)y,
fort ∈ [0, 1].
This iswell-dened,and abijetionbeauseeverygeodesion
FP r
islosed.Clearly,
ϕ 1
isa dieomorphism,and it isstraightforward to hek that it isS 1
-equivariant, using the trigonometriformulas.Another very useful model for
G(r)
isS(τ ) = S(T (HP r ))
, the spherebundle of the tangent bundle
τ
ofFP r
. There isa dieomorphismψ : P V 2 −→ S(τ )
[x, y] 7→ t x (y) ∈ T π(x) FP r
This is well-dened beause of (2), and we an give an expliit inverse:
Given
y ∈ T π(x) FP r
,ψ − 1 (y) = [x, t − x 1 (y)]
. Thus we an giveS(T (FP r ))
a rotation ation of
S 1
, namely the ation that makes this dieomorphismS 1
-equivariant. Combining this with (4), we have anS 1
-equivariant dieo- morphismψ − 1 ◦ ϕ 1 : S(τ ) −→ G(r).
(5)The last desription only works for
CP r
. Going bak toP V 2 (C r+1 )
, werst hange oordinates as follows
ϕ 2 : P V 2 (C r+1 ) −→ g P V 2 (C r+1 ), [x, v] 7→
x + iv
√ 2 , x − iv
√ 2
.
Here
P V g 2
isP V 2
equipped theS 1
-ationindued fromthis hangeofoordi-nates. Itis easilyomputed that the ation of
θ ∈ [0, 1]
isθ ∗ [a, b] = [za, zb]
where
z = e πiθ ∈ S 1
.We are interested in
∆(CP r )
, i.e. we divide out the rotation ation.Therefore we now onsider the following spae: Let
γ 2
be the standard 2-dimensional bundle over the Grassmannian Gr
2 (C r+1 )
of2
-planes inC r+1
,and let
p : P(γ 2 ) −→
Gr2 (C r+1 )
be the assoiated projetive bundle. ThenP(γ 2 ) = { V 1 ⊆ V 2 ⊆ C r+1 | dim C (V j ) = j }
. Wean make adieomorphism,ϕ 3 : S 1 \ P V g 2 (C r+1 ) −→ P(γ 2 ), [a, b] 7→
spanC { a } ⊆
spanC { a, b } .
This is well-dened, but only for
F = C
. In onlusion we get a ompositeS 1
-equivariant dieomorphismϕ : ∆(CP r ) ϕ
− 1
1 // S 1 \ P V 2 (C r+1 ) ϕ 2 // S 1 \ P V g 2 (C r+1 ) ϕ 3 // P(γ 2 ).
(6)1.3 Fibrations involving spaes of geodesis
We are going to ompute the ohomology and
K
-theory of the spaesG(r)
and
∆(r)
. In ohomology, our most important tool will be Serre's spe-tralsequene. I will write down the most important part; for the omplete
formulationand proof, see e.g [Hather2 ℄ Thm 1.14 pp.
Theorem 1.2 (Serre's Spetral Sequene). Let
F −→ X −→ B
be a -bration, with
B
a path-onneted CW omplex, andπ 1 (B )
ating triviallyon
H ∗ (F ; G)
. Then there is a spetral sequene{ E r p,q , d r }
onverging toH ∗ (X; G)
withE 2 p,q ∼ = H p (B; H q (F ; G)).
If
G = R
is a ring, then there is a produtE r p,q × E r s,t −→ E r p+s,r+t
, and thedierentials are derivations, i.e.
d(xy) = (dx)y + ( − 1) p+q x(dy)
. Forr = 2
the produt is
( − 1) qs
times the standard up produt. The produt strutureon
E ∞
oinide with that indued by the up produt onH ∗ (X; R)
.Forthe denitionofabration,and theuseful fatthatberbundlesare
brations, see [Hather1 ℄, p. 375 and Prop. 4.48.
Thereisasimilarresultforabrationin
K
-theory,butIamhiey goingtousetheimportantspeialasewherethebrationis
∗ −→ X −→ X
,alledthe Atiyah-Hirzebruhspetralsequene, see [Atiyah-Hirzebruh ℄:
Theorem1.3(Atiyah-HirzebruhSpetralSequene). Let
X
be aniteCWomplex. Then there is a spetral sequene
{ E r p,q , d r }
onverging toK ∗ (X)
with
E 2 p,q ∼ = H p (X; K q ( ∗ )).
We will need a way tobuild brations fromother brations, and this is
provided by the following theorem.
Theorem1.4. Let
F −→ X −→ B
bea bration,andassumethatthegroupG
ats freely onX
. Then,(i)
IftheG
-ationpreservesthebres,F/G −→ X/G −→ B
isabration.(ii)
IfG
ats freely onB
, thenF −→ X/G −→ B/G
is a bration.Proof. This follows from the fat that
G −→ X −→ X/G
is a bration,whihis aonsequene of the slie theorem, [Bredon ℄ Thm. 5.4.
To applythe spetral sequenes, wemust knowsome brationsinvolving
the spaes of geodesis. Firstby denition we have the bration
S 1 −→ G(r) −→ ∆(r).
(7)For the appliation of Serre's spetral sequene, note that the base is 1-
onneted. This an be seen from the long exat sequene of homotopy
groups, usingthat
G(r) ∼ = S(τ )
is 1-onneted.Then there is the map
P V 2 (F r+1 ) −→
Gr2 (F r+1 )
indued by the map
V 2 (F r+1 ) −→
Gr2 (F r+1 )
,(x, y) 7→ { xλ + yµ | λ, µ ∈ F }
,whih is well-dened on
P V 2
. The bre isP V 2 (F 2 )
. Bythe dieomorphism (4), this means we havethe brationG(1) −→ G(r) −→
Gr2 (F r+1 ).
Sinethe left
S
ationonthe totalspae isfreeand preservesthe bres,wean divide by it in the total spae and bre, by Theorem 1.4
(i)
obtainingthe bration
∆(1) −→ ∆(r) −→
Gr2 (F r+1 ).
(8)Againwe note that the base is 1-onneted.
1.4 Homotopy orbits of spaes of geodesis
In this setion we are going to study the so-alled homotopy orbits of the
spaes of geodesis we have studied so far. For this denition we need the
following onepts: Let
G
be a group, and suppose we have a ontratiblespae with a free
G
ation. It turns out that all suh spaes are homotopyequivalent, so we an dene
EG
to be any suh spae. We an then deneBG = EG/G
to bethe lassifying spae ofG
. Notethat this isa workingdenition; atually
BG
is dened for a ategory, but this is all I willneed.For
G = S 1
we ndES 1 ≃ S ∞
, sine this is ontratible. Thus we getBS 1 ≃ S ∞ /S 1 = CP ∞
.Denition1.5. Let
X
beatopologialspaewith a(left)ationofS 1
. Wedenethe spae of homotopy orbits of
X
byX hS 1 = ES 1 × S 1 X = ES 1 × X/
(e, tx) ∼ (et, x), t ∈ S 1 .
Projetion onthe rst fator givesamap
X hS 1 −→ BS 1
, and fora oho-mologytheory
h ∗
(weonsiderohomologyandK
-theory),wegetaninduedmap
h ∗ (BS 1 ) −→ h ∗ (X hS 1 ).
As explained in the introdution, this gives
h ∗ (X hS 1 )
the struture of anh ∗ (BS 1 )
-module.Reall that
G(r)
is the spae of simple parametrized geodesis with the free left ation ofS 1
given by rotation. The spae ofn
-times iteratedgeodesis,
G n (r)
, we have identied as anS 1
-spae withG(r) (n)
, whih isG(r)
with therotationationtwistedby then
thpowermapP n : S 1 −→ S 1
,see (3).
horizontal maps are brations with 1-onneted base spaes:
G(r)
BC n // ES 1 × S 1 G(r) (n) //
∆(r)
BS 1 B P n // BS 1
Here
C n ⊆ S 1
denotes the group ofn
th roots of unity.Proof. To see that the vertial map is a bration, use the produt bundle
G(r) (n) −→ ES 1 × G(r) (n) −→
pr1 ES 1
, and divide out by the free ation ofS 1
onboth total spae and base,aording toTheorem 1.4
(ii)
. Usingthe longexat homotopy sequene for the bration
S 1 −→ ES 1 −→ BS 1
shows thatthe base
BS 1
is 1-onneted.The horizontal bration is built up in steps: We start with the produt
bre bundle,
ES 1 −→ ES 1 × G(r) (n) −→
pr2 G(r) (n) .
Clearly,
C n ⊆ S 1
ats freely onES 1 × G(r) (n)
, preserving the bres. So byTheorem 1.4
(i)
, dividing out byC n
in the total spae and bre yields thebration:
BC n −→ ES 1 × C n G(r) (n) −→ G(r) (n) .
Weget
ES 1 /C n = BC n
beauseES 1
is a ontratiblespae uponwhihC n
ats freely, and so
ES 1 ≃ EC n
. Now onsider the quotient groupS 1 /C n
,whih isisomorphi to
S 1
by then
'th powermap. SineC n
ats triviallyonG(r) (n)
, we have an ation ofS 1 /C n
onG(r) (n)
. By denition, this ats onG(r) (n)
exatly asS 1
ats onG(r)
, so(S 1 /C n ) \ G(r) (n) ∼ = S 1 \ G(r)
. ByTheorem1.4
(ii)
,dividingoutbythis freeationinthe totaland basespaesgives usthe bration
BC n −→ ES 1 × C n G(r) (n)
/(S 1 /C n ) −→ S 1 \ G(r).
Now
ES 1 × C n G(r) (n)
/(S 1 /C n ) ∼ = ES 1 × S 1 G(r) (n)
,bythe denitionoftheations, sowe get the desired bration. As noted inSetion 1.3, the base is
1
-onneted.To get the ommutative square, note that we have the homotopy equiv-
alene pr
2 : ES 1 × G(r) −→ G(r)
, sineES 1
is ontratible. Sine this is anS 1
map andS 1
ats freely on both spaes, we an use [tomDiek℄Prop. 2.7 to onlude that
ES 1 × S 1 G(r) −→ S 1 \ G(r) = ∆(r)
is also ahomotopy equivalene. The upper vertial map in the square is dened as
pr
2 : ES 1 × S 1 G(r) −→ ∆(r)
usingthis homotopyequivalene. Fortheiden- tiationS 1 /C n
withS 1
above,we usedthen
thpowermapP n : S 1 −→ S 1
,so for the diagram to ommutate, the lower horizontal map
BS 1 −→ BS 1
must also be the one indued by
P n
. Note: This is well-dened onBS 1
beause
S 1
is ommutative.Remark 1.7. If we let
n = 1
, the vertial bration beomesG(r) −→
ES 1 × S 1 G(r) −→ BS 1
. Asnotedintheproof,ES 1 × S 1 G(r) −→ S 1 \ G(r)
isahomotopyequivalene. So, uptohomotopy,wehaveinpratieabration
G(r) −→ ∆(r) −→ BS 1 .
(9)2 Cohomology of spaes of geodesis in
HP r
2.1 The parametrized geodesis
Inthissetionwendthe ohomologyofthespaeofparametrizedgeodesis
on
HP r
,G(r) = G(HP r )
, followed by some Lemmasneessary to determinethespaeoforiented,unparametrizedgeodesis,
∆(r) = ∆(HP r ) = S 1 \ G(r)
.Theorem 2.1. As a graded ring,
H ∗ (G(HP r ); Z) ∼ = Z[y, τ ]/
(r + 1)y r , y r+1 , τ 2 ,
where
y ∈ H 4 (G(HP r ); Z)
andτ ∈ H 4r+3 (G(HP r ); Z)
.Let
p
be a prime number. ThenH ∗ (G(HP r ); F p ) ∼ =
F p [y, σ]/ h y r+1 = 0, σ 2 = 0 i , p | r + 1;
F p [y, τ ]/ h y r = 0, τ 2 = 0 i , p ∤ r + 1.
where
y ∈ H 4 (G(HP r ); F p )
,σ ∈ H 4r − 1 (G(HP r ); F p )
,τ ∈ H 4r+3 (G(HP r ); F p )
.Proof. We use the dieomorphismfrom(5),
G(r) ∼ = S(τ )
, whereS(τ)
isthesphere bundleof the tangentbundle,
S 4r − 1 −→ S(τ ) −→ HP r .
Sine
HP r
is 1-onneted, we an use Serre'sspetral sequene,H p (HP r ; H q (S 4r − 1 )) ⇒ H p+q (S(τ ))
(10)(here the oeients will be
Z
atrst, andF p
toprove the lastpart) whihhas the following
E 2
page:4r − 1 σ yσ y 2 σ y r σ
0 1 y y 2 y r
0 4 8 . . . 4r
We an see for dimensional reasons that there an only be one non-trivial
dierential,namely
d 4r (σ)
. Forthespherebundle,itisageneraltheoremthatthis dierentialismultipliation by the Euler harateristi of the manifold,
here
HP r
, sod 4r (σ) = (r + 1)y r
. This is proved in [Milnor-Stashe℄, Cor.11.12and Thm. 12.2. This is an injetivemap
Z −→ Z
, sowhen passing tothe
E 4r+1
page, the result is4r − 1 0 yσ y · yσ y r − 2 · yσ y r − 1 · yσ
0 1 y y 2 y r − 1 y r
0 4 8 . . . 4r − 4 4r
Asmentioned,therearenoothernon-trivialdierentials,sothisis
E ∞
. Also,therearenoextensionproblemssinethereisatmostonenon-trivialgroupon
eahdiagonal
p + q = n
,soyσ
denesalassinH 4r+3 (S(τ); Z)
whihweallτ
. We an then read o the lassesy ∈ H 4 (S(τ ); Z)
andτ ∈ H 4r+3 (S(τ ); Z)
with the relations
y r+1 = 0
,(r + 1)y r = 0
,andτ 2 = 0
.Toprovetheresultwith
F p
oeients,weusethesamespetralsequene(10), now with
F p
-oeients. In asep | r + 1
,d 2 (σ) = 0
, so there are nonon-trivial dierentials, and
E ∞ = E 2
. As above, there are no extensionproblems, and
σ
denes an element inH 4r − 1 (S(τ); F p )
. So we an read othedesiredresult. Inase
p ∤ r + 1
,r + 1
isaunit inF p
,sod 2 : F p σ −→ F p y r
is an isomorphism. So when passing to the
E 4r+1
page, these two groupsdisappear. The result follows.
Now we an deal with the smallest ase,
HP 1
, whih we have shown inExample1.1isdieomorphito
S 4
. This isgoingtobeuseful, sine wehavethe bration
∆(HP 1 ) −→ ∆(HP r ) −→
Gr2 (H r+1 )
from(8).Lemma 2.2.
H ∗ (∆(HP 1 ); Z) ∼ = Z[x, t]/
2t − x 2 , t 2 ,
where
x ∈ H 2 (∆(HP 1 ); Z)
andt ∈ H 4 (∆(HP 1 ); Z)
.Proof. Weusethebration
S 1 −→ G(HP 1 ) −→ ∆(HP 1 )
fromtheS 1
ation.Hereweknowtheohomologyofthebreandthetotalspae,thelatterfrom
Theorem 2.1,
H n (G(HP 1 )) =
Z, n = 0, 7
;Z/2Z, n = 4
;0,
else.We an use the Serre'sspetralsequene,
H p (∆(HP 1 ); H q (S 1 ; Z)) ⇒ H p+q (G(HP 1 ); Z),
to nd the ohomology of the base. Let
σ ∈ H 1 (S 1 )
denote a generator.The
E 2
page has only two non-zero rows. We see that the only possiblenon-trivial dierentials are
d 2
, soE 3 = E ∞
. We know the total spae hasnothing in degree 1, so there must be zero at
(1, 0)
sine this annot bekilledby anything. So
H 1 (∆(HP 1 )) = 0
, whih meansthereiszero at(1, 1)
,too. Also,
σ
must be killed by an outgoing dierential, sod 0,1 2
is injetive.Atually it must be an isomorphism, otherwise something would survive in
degree2,and thereisnothing. Sowehave a
H 2 (∆(HP 1 ); Z) ∼ = Z
generated,say, by
x = d 2 (σ)
. Letus take a look attheE 2
page as we knowit now:1 σ 0 σx ? ? ? ? ? ? · · · 0 1 0 x ? ? ? ? ? ? · · · 0 1 2 3 4 5 6 7 8 · · ·
Continuinginthisfashionweseethereiszeroat
(3, 0)
sineH 3 (G(HP 1 ); Z) = 0
, and so also at(3, 1)
. Likewise, there are zeroes at(5, 0)
and(5, 1)
. Nowonsider
d 2,1 2
. This must be injetive, sine it starts in degree 3, where thetotal spae has nothing. Also,
d 2,1 2
ends at(4, 0)
, and must besuh that weget
H 4 (G(HP 1 ); Z) = Z/2Z
when taking the okernel of it. This means itmust be multipliation by
± 2
; we might as well say 2 for onreteness. SoH 4 (∆(HP 1 ); Z) ∼ = Z
generated by somet
, whih we an hoose suh thatd 2 (σx) = 2t
. A quik summary:1 σ 0 σx 0 σt 0 ? ? ? · · · 0 1 0 x 0 t 0 ? ? ? · · · 0 1 2 3 4 5 6 7 8 · · ·
Nowwehavegottensomethingat
(4, 1)
,butthetotalspaehaszeroindegree5,so
σt
must bekilledby the outgoingdierentiald 4,1 2
. Againitmust beanisomorphism. Note that by the derivation property of
d 2
,d(σt) = d(σ)t − σd(t) = d(σ)t = xt
so
xt
is agenerator ofH 6 (∆(HP 1 ); Z)
. This givesusaZ
at(6, 1)
generatedby
σxt
. Now to see what further happens, we note that∆(HP 1 )
is at most7
-dimensional, sineG(HP 1 ) = S(T (HP 1 ))
is a7
-manifold. So we knowthat
H ∗ (∆(HP 1 ); Z)
is zero above degree7
. This means thatσxt
annotbe killed, soit survives to
E ∞
, meaningthere an be nothing else in degree7
. So from olumn 7 and onwards there are zeroes in theE 2
page. Now weknow the full story:
1 σ 0 σx 0 σt 0 σxt 0 0 · · ·
0 1 0 x 0 t 0 xt 0 0 · · ·
0 1 2 3 4 5 6 7 8 · · ·
Toget tothe bottomof the multipliativestruture wealulate:
2t = d(σx) = d(σ)x − σd(x) = d(σ)x = x 2 .
Fordimensional reasons
t 2 = 0
, and all other relationsome from these two(e.g.
x 3 = x 2 · x = 2xt
). This proves the result.Wenowturntothegeneralaseof
∆(r)
. Wehavethe brationfrom(8),∆(HP 1 ) −→ ∆(HP r ) −→
Gr2 (H r+1 ).
So in order to apply Serre's spetral sequene, we need to know the oho-
mology of Gr
2 (H r+1 )
. This is taken are of by the following Lemma, whihisthe quaternion version of [Bökstedt-Ottosen ℄ Thm. 3.1:
Lemma 2.3. For
r ≥ 1
,H ∗ (
Gr2 (H r+1 ); Z) ∼ = Z[p 1 , p 2 ]/ h ϕ r , ϕ r+1 i ,
where
p 1 , p 2
arethePontryaginlassesofthestandardbundleγ 2 ց
Gr2 (H r+1 )
,and
ϕ i = ϕ i (p 1 , p 2 )
is the polynomial given indutively byϕ 0 = 1, ϕ 1 = p 1 , ϕ i = − p 1 ϕ i − 1 − p 2 ϕ i − 2 ,
fori ≥ 2.
Proof. We use a result of Borel, [Borel℄ Prop. 31.1. Let
γ 2 ց
Gr2 (H r+1 )
denote the standard 2-dimensional bundle, i.e. the bre over
V ⊆ H r+1
is
V
. Letp i
,i ≥ 0
be the Pontryagin lasses,p i ∈ H 4i (
Gr2 (H r+1 )
, whihsatisfy
p i = 0
fori > 2
, sineγ 2
is 2-dimensional. Letγ ¯ r − 1
denote its(r − 1)
-dimensional orthogonal omplement, i.e. the bre overV ⊆ H r+1
isV ⊥ ⊆ H r+1
. Denote the Pontryagin lasses of this bundlebyp ¯ j
,j ≥ 0
,p ¯ j ∈ H 4j (
Gr2 (H r+1 ))
,and note thatp ¯ j = 0
forj > r − 1
. Thenγ 2 ⊕ γ ¯ r − 1 ∼ = ε r+1
,thetrivialbundleofdimension
r + 1
. ThesumformulaforPontryagin lassesgives the relations
X
i+j=k
p i p ¯ j = ¯ p k + ¯ p k − 1 p 1 + ¯ p k − 2 p 2 = 0,
fork > 0
(11)Borel'stheoremstatesthat
H ∗ (
Gr2 (H r+1 ); Z)
isgeneratedbythePontryaginlasses of
γ 2
and¯ γ r − 1
, subjet tothe relations mentioned above:H ∗
Gr2 (H r+1 ); Z ∼ = Z[p i , p ¯ j | i, j > 0]/ h{ p i } i>2 , { p ¯ j } j>r − 1 , X
i+j=k
p i p ¯ j
k>0 i .
By(11) we see thatwean indutivelyexpress
p ¯ k
asa polynomialinp 1
andp 2
. Callthat polynomialϕ k
, sop ¯ k = ϕ k (p 1 , p 2 )
, and we get from (11)ϕ 0 = 1, ϕ 1 = p 1 , ϕ i = − p 1 ϕ i − 1 − p 2 ϕ i − 1 , i ≥ 2.
Then we get
H ∗ (
Gr2 (H r+1 ); Z) ∼ = Z[p 1 , p 2 , p ¯ j | j > 0]/
*
{ p ¯ j } j>r − 1 , X
i+j=k
p i p ¯ j
k>0
+
∼ = Z[p 1 , p 2 , p ¯ 1 , p ¯ 2 , . . .]/ D
{ p ¯ j } j>r − 1 , { p ¯ k − ϕ k (p 1 , p 2 ) } k>0
E
∼ = Z[p 1 , p 2 ]/ h ϕ k | k ≥ r i .
From the indutive formula for
ϕ k
it is seen thath ϕ k | k ≥ r i = h ϕ r , ϕ r+1 i
,and this proves the lemma.
2.2 The unparametrized geodesis
Reall
H ∗ (BS 1 ) ∼ = H ∗ (CP ∞ ) ∼ = Z[u]
whereu
has degree 2; a fat that anbe dedued from
H ∗ (CP n ) ∼ = Z[u]/ h u n+1 i
.Theorem 2.4. The spae of unparametrized oriented geodesis,
∆(HP r )
,has the following ohomology:
H ∗ (∆(HP r ); Z) ∼ = Z[x, t]/ h Q r , Q r+1 i ,
where
x ∈ H 2 (∆(HP r ); Z)
istheimageofthegeneratoru ∈ H ∗ (BS 1 ) ∼ = Z[u]
and
t ∈ H 4 (∆(HP r ); Z)
.Q k
fork ∈ N
isapolynomialinx
andt
indutivelygiven by
Q 0 = 1, Q 1 = 2t − x 2 , Q s = (2t − x 2 )Q s − 1 − t 2 Q s − 2 ,
fors ≥ 2.
Note that Lemma 2.2 is a speial ase of this with
r = 1
:Q 1 = 2t − x 2
,and
Q 2 = (2t − x 2 )Q 1 − t 2 ≡ t 2 (mod Q 1 )
. TheproofofTheorem2.4forHP r
isnot atalllikethe
CP r
ase, sine∆(HP r )
isnot isomorphitoP(γ 2 )
, andthe proof willtakequite sometime. First we show that the ohomologyisa
polynomial algebragenerated by lasses
x
andt
asinthe Theorem, moduloertain relations. It will follow from Lemma 2.3 that the polynomials
Q r
,Q r+1
are among these relations. Then we use a purely algebrai ountingargumentto showthat there an beno further relations.
Proposition 2.5 (Theorem 2.4, Part1). There is a surjetive map
Z[x, t]/ h Q r , Q r+1 i ։ H ∗ (∆(HP r ); Z).
Proof of Theorem 2.4,Part 1. We write down the
E 2
page of the Serre'sspetral sequene for the bration