since pev0Ψqpp, zq p meaning thatev0Ψis a submersion.
We clearly have
Dl rpSnSn1,Ψ, Dp1qqs rpSnSn1,Ψ,0qs 0 . It remains to check that aa0 and that alla.
Recall that for de Rham chains we have
rpU, φ, π!ωqs rpU1, φπ, ωqs,
where π!: ΩrcpU1q ÑΩrcdimU1 dim UpUq is the integration along the ber dened for C8-submersionsπ :U1ÑU.
We use the concatenationc:c0,0 dened in the previous section. This yields
aacpaSnaq c rpDn,Φ, ωqs Sn rpDn,Φ, ωqs c rpDnSnDn,ΦΦ, πω^πωqs
rpDn,Φ, ω^ωqs 0 for π :DnSnDnÑDn since ω2 PΩ2nc pDnq vanishes.
We further have allasince alc aSnl
c rpDn,Φ, ωqs SnrpSnSn1,Ψ,1qs c rpDnSnpSnSn1q,ΦΨ, πω^π1qs c rppSnSn1q SnDn,ΨΦ, π1^πωqs la
for the dieomorphismDnSnpSnSn1q Ñ pSnSn1q SnDnthat in particular is a C8-submersion. FurthercpaSnlq cplSnaq sinceais a family of constant Moore loops.
For even dimensional spheres this construction does not work since we do not nd a nowhere vanishing vector eld on these spheres. The existence of poles complicates the denition of l as a regular chain.
can be realized by choosing a representative γα :S1 ÑN and considering
Γα :S1 R{ZÑLN
t Ñγαp 1{ltq . forl being the winding number of γα. For α0we have
Γ0 :N ÑLN pÑγp
whereγpptq p for all tP S1. We set imΓα :Sα1 and imΓ0 :N. ForX :M N the free loop space LX thus topologically looks like
LpM Nq
LM LN LM §
αPπ0pLNq
LαN Cor. p2.15q
LM pN \ §
0α
P rπ1pNq rπ1pMNq
Sα1q pLMNq \ pLM §
α0
Sα1q .
Our goal is to transfer the structure dened above, namely the dg algebra structure and the dg Lie structure of
pCpXq, D,,t,uq to anA8-algebra and an L8-algebra on homology
HpCpXq, Dq HpLXq.
The basic idea of the following construction is that we want the subspaces Sα1 to be disjoint implying that theA8{L8-algebra operations on homology are essentially zero.
The construction further yields an A8-algebra morphism f and an L8-algebra mor-phismφ that are 8-quasi-isomorphism
HpLXq,tmnun¥1,tλnun¥1 ftfnu // CpLXq,mr1,mr2 . and
HpLXq,tmnun¥1,tλnun¥1
φtφnu // CpLXq,λr1,rλ2 . where
r
m1 :D, mr2pa, bq: p1q|a|ab , rλ1 :D, rλ2pa, bq: p1q|a|ta, bu .
We work with the disk of radius r dened as Dnprq : tx P Rn| |x| ¤ ru and set Dn:Dnp1q in the following.
Proposition 4.11
Let A rπ1pNq be the set of primitive nontrivial homotopy classes of loops in N. Then there exist curves γa in N indexed by a P A and closed tubular neighbour-hoods Oa γa with the following properties:
piq The curveγa represents the homotopy classa.
piiq Oa is a smooth submanifold of N with boundary and is dieomorphic to S1Dn1 via a dieomorphism φa:S1Dn1 ÑOa.
piiiq For ab the submanifolds Oa and Ob are disjoint.
Remark that in particular Oa and Oa are disjoint.
Proof : For a0 the curveγaN is chosen as a representative ofa.
The manifold N is compact and thus π0pLNq rπ1pNq is countable. We choose a counting
A ta1, a2, ...u .
Fix γa1 and the closed tubular neighbourhood Oa1 in N, which is possible due to corollary 2.3 of [23] for example. We have a dieomorphism φa1 :S1Dn1 ÑOa1. We recursively isotope γai for i1 and use the same notation for the perturbedγai. Since we use isotopies the perturbed γai is still a representative of ai.
For the inductive step assume that we have modied γa1, ..., γak and constructed disjoint closed neighbourhoods Oa1, ...,Oak satisfying piq - piiiq of the proposition.
Isotopeγak 1 such that
γak 1&γa1 , γak 1&γa2, ..., γak 1&γak .
Such isotopies exist due Corollary IV.2.4 of [23] for example and the fact that the γai's are smooth compact submanifolds. Since the curvesγai are one dimensional and we assume N to be of dimensionm¥3this implies
γaiXγaj H for0 i, j¤k 1withij .
By radially moving out we can achieve that γak 1 intersectsOai (1¤i¤k) only in BOai :Bai S1Sn2 and that
γak 1 XOai H .
These submanifoldsBai, Baj N are disjoint, closed and compact, thus have positive pairwise distances di,j ¡0. We x disjoint open neighbourhoodsUai ofBai inM for 1¤i¤k. TheBai's are dieomorphic to S1Sn2 and in particular hypersurfaces.
We can achieve that
γak 1 XOai H for 1¤i¤k
by perturbing γak 1 in Uai. After these perturbations for all 1 ¤i¤k the subman-ifolds γak 1 and Oai M are disjoint and have a distance di ¡ 0. We thus can
constructOak 1 as a closed tubular neighbourhood of γak 1, and in particular we can arrange
OaiXOaj H for0 i, j¤k 1withij . This concludes the inductive step and thus proves the proposition.
Remark 4.12. We x a smooth homotopy
H :S1Dn1 r0,1s ÑS1Dn1 pτ, x, tq ÞÑHtpτ, xq where Htpτ, xq is the ow of the vector eld
Vpτ, xq:ρp|x|q BxBn1
on S1 Dn1 at time t. Here x px1, ..., xn1q P Dn1 Rn1 and ρ smooth is a cut-o function of the form
The homotopy H satises
piq Ht id near BpS1 Dn1q for all t . piiq H0 id .
piiiq H1 S1Dn1p1{4q
X S1Dn1p1{4q
H .
Due to the work of Irie in [20] we know that the homology of the complex CpLpM Nqq
is isomorphic toHpLpMNqq. Further the de Rham loop product and the de Rham loop bracket descend to homology and there they coincide with the loop product and the loop bracket respectively dened by Chas and Sullivan in [5].
We have
CpLpM Nqq C0pLpM Nqq `à
aPA
CapLpM Nqq
where CapLpM Nqq contains chains in homotopy classes which are positive iterates of a and C0pLpM Nqq contains chains of contractible loops.
Remark the subcomplex C1 CpLpMNqq that splits as
C1 : pC1q0`à
aPA
pC1qa
where pC1qa CapLpM Nqq contains all the chains whose loops are in M φapS1Dn1q
and pC1q0 C0pLpM Nqq contains all the chains whose loops are contractible in M N and further constant in N.
Lemma 4.13
The inclusion of the chain complex
C1 ãÑCpLpM Nqq induces an isomorphism on homology. In particular
HpC1q HpLpM Nqq.
FurtherC1 is closed under the de Rham loop product and the de Rham loop bracket dened in [20].
Proof : By proposition 4.11 for aPA andαkawe have homotopy equivalences
LM §
k¥1
LkpS1Dn1qidÝÑLφaLM §
k¥1
LkaN
and clearly
LMN ÝÑLML0N .
The two complexes C1 and CpLpM Nqq are the complexes of de Rham chains on the loop spaces on the left and the right respectively. By corollary 4.6 we thus get that the homology of these spaces is isomorphic.
By denition the homotopy equivalences are compatible with the de Rham loop prod-uct and the de Rham loop bracket.
Remark 4.14. The lemma in general holds for Dn1prq with 0 r ¤1. For reasons of clarity in the upcoming proofs we highlight the radius as C1,r if r 1.
The homotopy of remark 4.12 yields chain maps h: à
aPA
pC1qa Ñà
aPA
pC1qa of degree 1 induced by H and
T : à
aPA
pC1qaÑ à
aPA
pC1qa
of degree 0 induced by H1. Further H0 induces the identity on C1. These relate to
Dh hD id T (4.23) by proposition 2.5. of [20] which guarantees that smoothly homotopic maps induce chain homotopic ones onC1.
The topological rewriting and simplication of the set-up will imply that xy0 and tx, yu 0
forx, y P HpLXq being homology classes of loops in non-trivial conjugacy class com-ponents of rπ1pM Nq since the classes x, y can be either represented as families of loops that are disjoint inN due to (iii) of remark 4.12 and (4.23).
The following theorems state the generalization of this fact to the higher A8{L8 -algebra operationsmk¥3 and λk¥3 on homology HpLXq.
Remark that in the following we work with
HpLXq HpL0Xq ` à
aPA
HpLaXq whereHpLaXq À
αka fork¥1
HpLαXq when setting
LaX : §
αka fork¥1
LαX
forα P rπ1pXq rπ1pNq.
In the following theorems we assume X M N and M, N to be smooth, closed and oriented Riemannian manifolds of nite dimension dimM m ¥ 0 respectively dimN n ¥3. Further M is simply connected and N has negative sectional curva-ture.
Theorem 4.15
The homotopy transfer construction for
HpLXq ÝÑCpLXq
equips HpLXq with an A8-algebra structure pHpLXq,tmkuk¥1q and yields an A8-algebra morphism
f tfkuk¥1 :HpLXq ÝÑCpLXq such that:
(i) m1 0,
(ii) f1 is a cycle choosing homomorphism and in particular a quasi-isomorphism, (iii) m2 corresponds to the loop product, and
(iv) mkpx1, ..., xkq 0
for k ¥ 2 whenever the inputs xi are classes of families of loops that are non-contractible, that is xi P À
aPAHpLaXq.
Theorem 4.16
The homotopy transfer construction for
HpLXq ÝÑCpLXq
equips HpLXq with an L8-algebra structure pHpLXq,tλkuk¥1q and yields an L8-algebra morphism
φ tφkuk¥1 :HpLXq ÝÑCpLXq such that:
(i) λ1 0,
(ii) φ1 is a cycle choosing homomorphism and in particular a quasi-isomorphism, (iii) λ2 corresponds to the loop bracket, and
(iv) λkpy1, ..., ykq 0
for k ¥ 2 whenever the inputs are elements yi P HpLaiXq for primitive classes ai PA which are not all equal.
For proving the theorems we apply the homotopy transfer construction presented by Kadeishvili in [21] by recursively constructing the higher operations and morphisms.
Proof of theorem 4.15 : We use the notation from section 3.
For the rst operations we set
m1 0U1 and f1 ι
whereι:HpLaXq Ñ pC1,1{4qafor aPAand ι:HpL0Xq Ñ pC1q0 are cycle choosing homomorphism. Thus equation (3.3) is satised, namely
U1f1m10Dι rm1f1 .
The operation m2 r rm2ib2s onHpLXq is the loop product up to sign due to how r
m2 : de Rham loop product are constructed by Irie.
It remains to prove pivq. For generalk¥2 we have
Ukpx1, ..., xkq
k¸1 s1
r
m2pfspx1, ..., xsq, fkspxs 1, ..., xkqq
k¸2 i0
k¸1 j2
p1qi 1 |x1| ... |xi|fkj 1px1, ..., xi, mjpxi 1, ..., xi jq, xi j 1, ..., xkq .
We will show that there exist maps
fk:HpLXqbk ÝÑC1 such that
Dfk rm1fkUkf1mkUk
when acting on inputsxithat are classes of families of loops that are non-contractible.
This then yields
mkpx1, ..., xkq rUkpx1, ..., xkqs 0 for such inputs.
For the induction assume the stated assertion holds up to degree k. We perform the inductive step for kÑk 1.
Assume all operations and morphisms are constructed up to degreek. In the induction hypothesis we assume that the image of fkpx1, ..., xkq is contained in the support of f1pxk 1q when acting onx1, ..., xk 1as in the condition for pivq. In particular we thus have
r
m2pfkpx1, ..., xkq, T f1pxk 1qq 0. (4.24) According to the denition of the loop product by Irie we know that for chainsxPC1 and xi as in the condition for piiiq we have that the supports of
r
m2pf1pxiq, xq and rm2pT f1pxiq, xq
are contained in the support of f1pxiq and T f1pxiq respectively. By piiiq of remark 4.12 we thus have
r
m2pf1pxiq, T f1pxiqq 0 . (4.25) We dene
fk 1 : p1qd1 ... dk k1mr2 pfkbhf1q (4.26) when acting on Hd1pLai1Xq b...bHdk 1pLaik 1Xq fora1, ..., aik 1 PA. Remark that for x1, ..., xk 2 as in the condition for pivq we get that the image offk 1px1, ..., xk 1q is contained in the support of f1pxk 2q
Due to he work of Kadeishvili in [21] we can dene fk 1 for the remaining cases if at least one input is ofHpL0Xq such thatDfk 1Uk 1f1mk 1 implyingmk 1 not necessarily zero. We do not want to prove something about these operations here.
It remains to show that
Dfk 1 Uk 1
when acting onx1, ..., xk 1 as in the condition for piiiq. Recall the A8-operations in the case rmk0 for k¥3, namely
r
m1 rm10, r
m1p rm2px, yqq mr2p rm1pxq, yq p1q|x| 1mr2px,mr1pyqq 0, r
m2p rm2px, yq, zq p1q|x| 1mr2px,mr2py, zqq 0 . Since |fk| k1 we get
p1q|x1| |xk| k1pDfk 1qpx1, ..., xk 1q (4.27) D rm2 pfkbhf1q
px1, ..., xk 1q
rm2 ppDfkq bhf1q p1q|x1| |xk| kmr2 pfkb pDhf1qq
px1, ..., xk 1q. For the rst summand we use the induction hypothesis
Dfj Uj implying mj 0 for 1¤j¤k. In particular we thus get
DfkUk
k¸1 s1
r
m2pfsbfksq
k¸2 i0
k¸1 j2
fkj 1p1ibmj b1kijq
k¸1 s1
r
m2pfsbfksq .
For the second summand we use (4.23) and Df1 0, that is r
m2 pfkb pDhf1qq rm2 pfkb ppid Tqf1qq rm2 pfkbf1q by (4.24). For (4.27) we deduce
p1q|x1| |xk| k1pDfk 1qpx1, ..., xk 1q
k¸1 s1
r
m2 p rm2pfsbfksq bhf1q p1q|x1| |xk| kmr2 pfkbf1q
px1, ..., xk 1q . Using that the de Rham loop product and thus rm2 is associative implies
p1q|x1| |xk| k1pDfk 1qpx1, ..., xk 1q
k¸1 s1
p1q|x1| |xs| smr2 pfsb rm2pfksbhf1qq p1q|x1| |xk| kmr2 pfkbf1q
px1, ..., xk 1q
By denition (4.26) we have r
m2 pfksbhf1q p1q|x11| ... |x1ks| ks1fks 1
for x1i xs i, that is we get
p1q|x1| |xk| k1pDfk 1qpx1, ..., xk 1q
k¸1 s1
p1q|x1| |xk| k1mr2 pfsbfks 1q p1q|x1| |xk| k1mr2 pfkbf1q
px1, ..., xk 1q p1q|x1| |xk| k1
k¸1 s1
r
m2 pfsbfks 1q mr2 pfkbf1q
px1, ..., xk 1q p1q|x1| |xk| k1
¸k s1
r
m2 pfsbfks 1q
px1, ..., xk 1q
which is
p1q|x1| |xk| k1Uk 1px1, ..., xk 1q since mj|pÀ
aPA
HpLaXqqbk 0 for1¤j ¤kby the induction hypothesis.
Proof of theorem 4.16 : We use the notation from section 3.3.
For the rst operations we set
λ1 0V1 and φ1 ι ,
where ι:HpLaXq Ñ pC1qa for a PA and ι: HpL0Xq Ñ pC1q0 are cycle choosing homomorphisms. Thus equation (3.11) is satised, namely
φ1λ1V10Dι rλ1φ1 .
The operationλ2 rrλ2ib2s on HpLXq is the loop bracket up to sign due to how λr2 : de Rham loop bracket
are constructed by Irie.
The recursive construction of Kadeishvili yields
V2px, yq rλ2pφ1pxq, φ1pyqq
for x, yPHpLXq. Therefore by construction forabPA and xPHpLaXq we get
V2px, yq
"
0 , for yPHpLbXq
P pC1qa , for yPHpLaXq ory PHpL0Xq
since pC1qa and HpLaXq contains families/classes of loops that are positive iterates of the primitive nontrivial homotopy classa.
Forλ2: rV2s we get
λ2px, yq
"
0 , for yPHpLbXq
PHpLaXq , for yPHpLaXq or yPHpL0Xq which allows to dene φ2 such that
φ2px, yq
"
0 , fory PHpLbXq
P pC1qa , fory PHpLaXq or yPHpL0Xq Analogously we get
λ2px, yq PHpL0Xq and φ2px, yq P pC1q0 for x, yPHpL0Xq.
We end up with (3.11), namely
φ1λ2V2 rλ1φ2 .
It remains to prove piiiq. We perform the inductive step forkÑk 1.
Assume all operations and morphisms are constructed up to degreek and that
λlpc1, ..., clq
$&
%
0 , pIq
PHpLaXq , pIIq
PHpL0Xq , pIIIq (4.28) and
φlpc1, ..., clq
$&
%
0 , pIq P pC1qa , pIIq
P pC1q0 , pIIIq (4.29) for all 1¤l¤k. Here condition pIq means
pIq p Di, jP t1, ..., lu such that ci PHpLaXq and cj PHpLbXq for abPA , pIIq means
pIIq p @iP t1, ..., lu we either have ciPHpL0Xq orci PHpLaXq foraPA and there exists at least onei0P t1, ..., lu such that ci0 PHpLaXq and pIIIq means
pIIIq p @iP t1, ..., lu we haveci PHpL0Xq .
We prove that (4.28) and (4.29) hold forλk 1 andφk 1 which then proves piiiq of the theorem, namely that
λlpc1, ..., clq 0 if Di, j such thatci PHpLaXq andcj PHpLbXq forabPA and for alll¥1.
The recursive construction of Kadeishvili yields
Vk 1pc1, ..., ck 1q ¸
σ, p qk 1
cσrλ2pφppcσp1q, ..., cσppqq, φqpcσpp 1q, ..., cσpk 1qqq
¸
τ, 1 l k 1
cτφkl 2pλlpcτp1q, ..., cτplqq, cτpl 1q, ..., cτpk 1qq .
The multiplicities, the signs and in particular the question which σ and τ are used, shues or permutations, is an important issue in general. We may bypass these ques-tions since the statements above will hold independently for each summand.
Since only morphisms and operations of degree ¤kare involved we apply the induction hypothesis and get
Vk 1pc1, ..., ck 1q
$&
%
0 , pIq P pC1qa , pIIq P pC1q0 , pIIIq Sinceλk 1: rVk 1s we get
λk 1pc1, ..., ck 1q
$&
%
0 , pIq PHpLaXq , pIIq PHpL0Xq , pIIIq
.
According to the denition of Kadeishvili φk 1 is dened such that
Dφk 1pc1, ..., ck 1q:Vk 1pc1, ..., ck 1q φ1pλk 1pc1, ..., ck 1qq . Sinceφ1 satises (4.29) we can chooseφk 1 such that
φk 1pc1, ..., ck 1q
$&
%
0 , pIq P pC1qa , pIIq P pC1q0 , pIIIq
.
This nishes the inductive step and proves piiiq namely that for pIq we have λk 1pc1, ..., ck 1q 0 .