Associahedral categories, particles and Morse functor

Jean-Yves Welschinger
June 25, 2009
Abstract:
Every smooth manifold contains particles which propagate. These form objects and
morphisms of a category equipped with a functor to the category of Abelian groups, turning
this into a 0 + 1 topological eld theory. We investigate the algebraic structure of this
category, intimately related to the structure of Stashes polytops, introducing the notion
of associahedral categories. An associahedral category is preadditive and close to being
strict monoidal. Finally, we interpret Morse-Witten theory as a contravariant functor, the
Morse functor, to the homotopy category of bounded chain complexes of particles.
Introduction
In a recent paper [12] dealing with Lagrangian Floer theory, a category of open strings was
associated to every symplectic manifold of dimension at least four, which comes equipped
with a functor to the category of Abelian groups, turning this into a 1 + 1 topological
eld theory. Lagrangian Floer theory was then interpreted as a contravariant functor,
the Floer functor, to the homotopy category of bounded chain complexes of open strings.
The purpose of the present paper is to detail the Morse theoretic counterpart of this
formalism. To every smooth manifold is associated a category of particles whose morphisms
are trajectories of such particles. It comes with a functor, the functor Coecients, whose
target is the category of Abelian groups and which satises the axioms of a 0+1 topological
eld theory. The category of particles has basically the same structure as the category of
open strings, a structure which is intimately related to Stashes polytopes, namely the
associahedra and multiplihedra. We propose an algebraic presentation of this structure and
call such categories associahedral. These are small, preadditive and almost strict monoidal
categories. Actually, these categories come with morphisms of cardinality and indices to
(Z, +) which we use to twist the functorial property of strict monoidal categories, see
Denition 1.1. Chain complexes in such categories satisfy some Leibnitz rule, chain maps
are twisted morphisms with respect to the monoidal structure and homotopies satisfy some
twisted Leibnitz rule. In the same way as in [12], we interpret the Morse data, given by a
Keywords: Morse theory, categories.
AMS Classication : 53D40.
1
a
r
X
i
v
:
0
9
0
6
.
4
7
1
2
v
1


[
m
a
t
h
.
S
G
]


2
5

J
u
n

2
0
0
9
Morse function, or by a nite collection of Morse functions in Fukayas version, together
with a generic Riemannian metric, as objects of a category of conductors. Then, we
interpret Morse-Witten theory as a contravariant functor from the category of conductors
to the homotopy category of bounded chain complexes of particles. Composition with the
functor Coecients then provides the usual chain complexes of free Abelian groups. As in
the work [12], this is possible thanks to the following observation: a Morse function gives
more than nitely many Morse trajectories between critical points with index dierence
one. It gives nitely many such trajectories together with canonically oriented Fredholm
operators. These Fredholm operators are given by the rst variation of Morse equation
and are thus just connections along the trajectories.
We rst give an algebraic treatment of associahedral categories, then introduce the
category of particles and nally dene the Morse functor. The latter is only dened for
closed smooth manifolds.
Acknowledgements:
I am grateful to the French Agence nationale de la recherche for its support.
Contents
1 Associahedral categories 3
1.1 Denition and the functor Coecients . . . . . . . . . . . . . . . . . . . . 3
1.2 Chain complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Chain maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Homotopies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Particles 8
2.1 Metric ribbon trees and Stashes associahedron . . . . . . . . . . . . . . . 8
2.2 The category of particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Coparticles and functor Coecients . . . . . . . . . . . . . . . . . . . . . . 14
3 Morse functor 14
3.1 Conductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.3 Gradient ow trajectories and Morse complexes . . . . . . . . . . . . . . . 18
3.4 Morse continuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.4.1 Stashes multiplihedron . . . . . . . . . . . . . . . . . . . . . . . . 20
3.4.2 Morse continuations . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.5 Morse functor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Bibliography 25
2
1 Associahedral categories
1.1 Denition and the functor Coecients
Recall that a category T =
_
Ob(T), Hom(T)
_
is said to be small when Ob(T) and Hom(T)
are sets and preadditive whenever the set Hom
P
+
,P
(T) Hom(T) of morphisms P
+

has the structure of an Abelian group for every objects P

+
, P

Ob(T). We denote
by s, t : Hom(T) Ob(T) the source and target maps and by
P
+
,P
Hom
P
+
,P
(T) the
unit element. A category T is said to be strict monoidal i it is equipped with a functor
: T T T which is associative and has a unit element. We denote by (P, P

)
Ob(T)Ob(T) PP

Ob(T) and (,

) Hom
P
+
,P
(T)Hom
P
+
,P
(T)

Hom
P
+
P
+
,P

P
(T) these associative products and by Ob(T), id

Hom
,
(T) their
unit elements. Functoriality means that id
P
id
P
= id
PP
for every P, P

Ob(T) and
(
2

2
) (
1

1
) = (
2

1
) (

1
) whenever s(
2
) = t(
1
) and s(

2
) = t(

1
). We
are going to twist the latter property in Denition 1.1.
Denition 1.1 An associahedral category T is a small preadditive category equipped
with an associative product : T T T having a unit element Ob(T) and distribu-
tive with respect to the preadditive structure such that:
1) The set (Ob(T), ) is a free monoid equipped with morphisms of cardinality q :
(Ob(T), ) (N, +) and index
N
: (Ob(T), ) (Z/NZ, +), where N N is given. The
unit element is the only one whose cardinality vanishes whereas elements of cardinality
one, called elementary, generate Ob(T).
2) The set (Hom(T), ) is a free monoid generated by morphisms with elementary
targets, whose unit element is id

Hom
,
(T). It is equipped with a morphism of index
: (Hom(T), ) (Z, +) additive with respect to composition and such that
N
t
N

s = mod (N), where the source and target maps are morphisms s, t : (Hom(T), )
(Ob(T), ).
3) For every P, P

Ob(T), id
P
id
P
= id
PP
. Moreover, for every
1
,
2
,

1
,

2

Hom(T) such that s(
2
) = t(
1
) and s(

2
) = t(

1
), we have the relation (
2

2
)(
1

1
) =
(1)
(
1
)(

2
)
(
2

1
) (

1
).
Remark 1.2 The third property of Denition 1.1 could have been replaced by the property
3) For every P, P

Ob(T), id
P
id
P
= id
PP
. Moreover, for every
1
,
2
,

1
,

2

Hom(T) such that s(
2
) = t(
1
) and s(

2
) = t(

1
), we have the relation (1)
(

1
)(
2
)
(
2

2
) (
1

1
) = (
2

1
) (

1
).
If T is an associahedral category, its opposite category T
op
, that is the category having
same objects and morphisms but with source and target maps exchanged, satises property
3. Likewise, if T and T

are equipped with a bijection : T T

such that (P
1

P
q
)

= P

q
P

1
, where T is associahedral, T

satises properties 1 and 2 of Denition

1.1 and preserves . Then, T

satises property 3

.
We dene the cardinality of a morphism Hom
P
+
,P
(T) of an associahedral category
T to be the quantity q() = q(P
+
) q(P

). Morphisms with elementary target -which

3
generate (Hom(T), ) - are called elementary. The category of open strings of a symplectic
manifold, introduced in [12], is associahedral. The aim of this paper is to introduce likewise
an associahedral category of particles associated to any smooth manifold.
Denote by /b the Abelian category of Abelian groups.
Denition 1.3 Let T be an associahedral category. The functor ( : T /b dened
by P Ob(T) Hom
,P
(T) Ob(/b) and Hom
P
+
,P
(T) :

Hom
,P
+(T)

Hom
,P
(T) is called the functor Coecients.
In our examples, all Abelian groups are free.
1.2 Chain complexes
Denition 1.4 Let T be an associahedral category. A bounded chain complex (, )
of elements of T is a nite set of objects of T together with a matrix : of
morphisms of T having the following three properties.
A
1
) For every P, P

Ob(T), P P

= P, P

.
A
2
) For every member of , () + q() = 1 and / .
A
3
) For every P
1
P
2
, the matrix of opposite morphisms
op
satises the Leibniz
rule

op
P
1
P
2
= (1)
q(P
2
)(
op
P
1
)

op
P
1
id
P
2
+ (1)
q(P
1
)
id
P
1

op
P
2
.
Axiom A
1
implies that contains all the elementary components of its objects. Axioms
A
2
implies that the function
N
q denes a graduation modulo N on for which is of
degree one. Axioms A
1
and A
3
imply that the dierential is determined by its elementary
components, so that for every P
1
P
q
, where P
i
is elementary for every 1 i q,

op
P
1
P
q
=
q

i=1
(1)
(qi)(
op
P
i
)+i1
id
P
1
P
i1

op
P
i
id
P
i+1
P
q
.
It is convenient to consider the pair (,
,
) as a chain complex as well, it will be called
the empty chain complex.
Lemma 1.5 Let
1
: P
+
1
P

1
and
2
: P
+
2
P

2
be two morphisms of the asso-
ciahedral category T and (, ) be a bounded chain complex of elements of T such that
P
+
1
P
+
2
, P

1
P
+
2
, P
+
1
P

2
and P

1
P

2
are in and such that
1
,
2
are members
of . Then, the contribution of
1
,
2
to the restriction of : P
+
1
P
+
2
P

1
P

2
vanishes.
Proof:
From Leibniz rule A
3
, the contribution of
1
,
2
to
P
+
1
P
+
2
equals (1)
q(P
+
2
)(
1
)

1

id
P
+
2
(1)
q(P
+
1
)
id
P
+
1

2
whereas their contribution to
P

1
P
+
2
,
P
+
1
P

2
equals (1)
q(P

1
)
id
P

2
and (1)
q(P

2
)(
1
)

1
id
P

2
respectively. From the third property of Denition 1.1, their
contribution to
P
+
1
P
+
2
thus equals (1)
q(P

2
)(
1
)+q(P
+
1
)

2
+(1)
(
1
)(
2
)+q(P
+
2
)(
1
)+q(P

1
)

2
. Axiom A
2
then provides the result.
4
Lemma 1.6 Let
1
: P
+
2
P
0
2
and
2
: P
0
2
P

2
be two morphisms of the associahe-
dral category T such that
2

1
= . Let (, ) be a bounded chain complex of elements
of T such that P
1
P
+
2
P
3
, P
1
P
0
2
P
3
and P
1
P

2
P
3
are in and such that
1
,
2
are members of , where P
1
, P
2
are objects of T. Then, the contribution of
1
,
2
to
the restriction of : P
1
P
+
2
P
3
P
1
P

2
P
3
vanishes.
Proof:
From Leibniz rule A
3
, the contribution of
1
to
P
1
P
+
2
P
3
equals (1)
q(P
1
)+q(P
3
)(
1
)
id
P
1

1
id
P
3
while the contribution of
2
to
P
1
P
0
2
P
3
equals (1)
q(P
1
)+q(P
3
)(
2
)
id
P
1

2
id
P
3
.
Hence, their contribution to the composition
P
1
P
+
2
P
3
equals (1)
q(P
3
)(
2

1
)
id
P
1

(
2

1
) id
P
3
= .
1.3 Chain maps
Denition 1.7 Let T be an associahedral category. A chain map (or morphism) H :
(, ) (

) between the bounded chain complexes (, ) and (

) of T is a map
satisfying the following two axioms:
B
1
) For every member of H having non empty source, () + q() = 0. Moreover,

H = H .
B
2
) For every P
1
P
2

, the opposite morphism H

op
satises the relation H
op
P
1
P
2
=
(1)
q(P
2
)(H
op
P
1
)
H
op
P
1
H
op
P
2
.
Axiom B
1
implies that H is of degree 0 for the modulo N graduation dened by the
function
N
q. Axioms A
1
and B
2
imply that the morphism H is determined by its
elementary components, so that for every P
1
P
q

, where P
i
is elementary for
every 1 i q,
H
op
P
1
P
q
= (1)
P
q
i=1
(qi)(H
op
P
i
)
H
op
P
1
H
op
P
q
.
Lemma 1.8 Let T be an associahedral category and H : (, ) (

), H

:
(

) (

) be two chain maps given by Denition 1.7. Then, the composition

H

H : (, ) (

) is a chain map.
Proof:
This composition satises Axiom B
1
. Moreover,
(H

H)
op
P
1
P
2
= H
op
H
op
P
1
P
2
= (1)
q(P
2
)(H
op
P
1
)
(1)
(q(P
2
)+q(H
op
P
2
))(H
op
tH
op
P
1
)
(H
op
tH
op
P
1
H
op
tH
op
P
2
) (H
op
P
1
H
op
P
2
)
from B
2
= (1)
q(P
2
)((H

H)
op
P
1
)
(H

H)
op
P
1
(H

H)
op
P
2
,
from B
1
and the third property of Denition 1.1, so that this composition satises Axiom
B
2
.
5
Lemma 1.9 Let H : (, ) (

) be a degree 0 map satisfying Axiom B

2
between
the bounded chain complexes (, ) and (

) of the associahedral category T. Then, for

every P
1
P
2

, the morphisms K =

H or H satisfy the following twisted Leibniz

rule: K
op
P
1
P
2
= (1)
q(P
2
)(K
op
P
1
)
K
op
P
1
H
op
P
2
+ (1)
q(P
1
)+(q(P
2
)+1)(H
op
P
1
)
H
op
P
1
K
op
P
2
.
Proof:
Let P
1
P
2

, we have
(H )
op
P
1
P
2
=
op
H
op
P
1
P
2
= (1)
q(P
2
)(H
op
P
1
)

op
(H
op
P
1
H
op
P
2
) from B
2
= (1)
q(P
2
)(H
op
P
1
)
(1)
(q(P
2
)+q(H
op
P
2
))(
op
tH
op
P
1
)
(
op
tH
op
P
1
id
tH
op
P
2
) (H
op
P
1
H
op
P
2
)
+(1)
q(P
2
)(H
op
P
1
)
(1)
q(P
1
)+q(H
op
P
1
)
(id
tH
op
P
1

op
tH
op
P
2
) (H
op
P
1
H
op
P
2
) from A
3
= (1)
q(P
2
)((H)
op
P
1
)
(H )
op
P
1
H
op
P
2
+ (1)
q(P
1
)+(q(P
2
)+1)(H
op
P
1
)
H
op
P
1
(H )
op
P
2
,
from A
2
, B
1
and the commutation relation (H
Q
1
H
Q
2
) ( id
Q
2
) = (1)
()(H
Q
2
)
(H
) H
Q
2
deduced from the third property of Denition 1.1, see Remark 1.2. Likewise,
(

H)
op
P
1
P
2
= H
op

op
P
1
P
2
= (1)
q(P
2
)(
op
P
1
)
H
op
(
op
P
1
id
P
2
) + (1)
q(P
1
)
H
op
(id
P
1

op
P
2
) from A
3
= (1)
q(P
2
)(
op
P
1
)
(1)
q(P
2
)(H
op
t
op
P
1
)
(H
op
t
op
P
1
H
op
P
2
) (
op
P
1
id
P
2
) +
(1)
q(P
1
)
(1)
(q(P
2
)+q(
op
P
2
))(H
op
P
1
)
(H
op
P
1
H
op
t
op
P
2
) (id
P
1

op
P
2
) from B
2
= (1)
q(P
2
)((

H)
op
P
1
)
(

H)
op
P
1
H
op
P
2
+ (1)
q(P
1
)+(q(P
2
)+1)(H
op
P
1
)
H
op
P
1
(

H)
op
P
2
,
from A
2
, B
1
and the same commutation relation as before.
Note that the twisted Leibniz rule given by Lemma 1.9 coincides with A
3
when H = id.
From this rule follows that a map H given by Lemma 1.9 is a chain map provided the
equality

H = H holds for its elementary members.

1.4 Homotopies
Denition 1.10 Let T be an associahedral category. A primitive homotopy K : (, )
(

) between the bounded chain complexes (, ) and (

) of T is a map satisfying the

following three axioms:
C
1
) For every P
1
P
q

, the elementary morphism (

K + K )
op
P
i
is non
trivial for at most one i 1, . . . , q.
C
2
) For every member of K having non empty source, () + q() = 1.
6
C
3
) For every P
1
P
2

, the opposite morphism K

op
satises the following twisted
Leibniz rule:
K
op
P
1
P
2
= (1)
q(P
2
)(K
op
P
1
)
K
op
P
1
H
op
P
2
+ (1)
q(P
1
)+(q(P
2
)+1)(H
op
P
1
)
H
op
P
1
K
op
P
2
,
where H : (, ) (

) is a chain map given by Denition 1.7

A primitive homotopy is determined by its elementary components, so that for every
P
1
P
q

, where P
i
is elementary for every 1 i q,
K
op
P
1
P
q
=
q

i=1
(1)
i1+
P
q
j=1
(qj)(L
op
P
j
)+
P
i1
j=1
(H
op
P
j
)
H
op
P
1
H
op
P
i1
K
op
P
i
H
op
P
i+1
H
op
P
q
,
where L
op
P
j
= H
op
P
j
if j ,= i and L
op
P
i
= K
op
P
i
.
Lemma 1.11 Let K : (, ) (

) be a primitive homotopy between the bounded

chain complexes (, ) and (

) of the associahedral category T and H : (, ) (

)
be the associated chain map. Then, H +

K + K is a chain map.
Proof:
Let P
1
P
2

, we have
(K )
op
P
1
P
2
=
op
K
op
P
1
P
2
= (1)
q(P
2
)(K
op
P
1
)

op
(K
op
P
1
H
op
P
2
) + (1)
q(P
1
)+(q(P
2
)+1)(H
op
P
1
)

op
(H
op
P
1
K
op
P
2
)
= (1)
q(P
2
)(K
op
P
1
)
_
(1)
(q(P
2
)+q(H
op
P
2
))(
op
tK
op
P
1
)
(
op
tK
op
P
1
id
tH
op
P
2
) (K
op
P
1
H
op
P
2
)
+(1)
q(P
1
)+q(K
op
P
1
)
(id
tK
op
P
1

op
tH
op
P
2
) (K
op
P
1
H
op
P
2
)
_
+
(1)
q(P
1
)+(q(P
2
)+1)(H
op
P
1
)
_
(1)
(q(P
2
)+q(K
op
P
2
))(
op
tH
op
P
1
)
(
op
tH
op
P
1
id
tK
op
P
2
) (H
op
P
1
K
op
P
2
)
+(1)
q(P
1
)+q(H
op
P
1
)
(id
tH
op
P
1

op
tK
op
P
2
) (H
op
P
1
K
op
P
2
)
_
from A
3
= (1)
q(P
2
)((K)
op
P
1
)
(K )
op
P
1
H
op
P
2
+ (1)
q(P
2
)(H
op
P
1
)
H
op
P
1
(K )
op
P
2
+
(1)
q(P
1
)+(q(P
2
)+1)((H)
op
P
1
)
(H )
op
P
1
K
op
P
2
+
(1)
q(P
1
)+1+(q(P
2
)+1)(K
op
P
1
)
K
op
P
1
(H )
op
P
2
,
7
from the third property of Denition 1.1, see Remark 1.2. Likewise,
(

K)
op
P
1
P
2
= K
op

op
P
1
P
2
= (1)
q(P
2
)(
op
P
1
)
K
op
(
op
P
1
id
P
2
) + (1)
q(P
1
)
K
op
(id
P
1

op
P
2
) from A
3
= (1)
q(P
2
)(
op
P
1
)
_
(1)
q(P
2
)(K
op
t
op
P
1
)
(K
op
t
op
P
1
H
op
P
2
) (
op
P
1
id
P
2
) +
(1)
q(P
1
)+q(
op
P
1
)+(q(P
2
)+1)(H
op
t
op
P
1
)
(H
op
t
op
P
1
K
op
P
2
) (
op
P
1
id
P
2
)
_
+
(1)
q(P
1
)
_
(1)
(q(P
2
)+q(
op
P
2
))(K
op
P
1
)
(K
op
P
1
H
op
t
op
P
2
) (id
P
1

op
P
2
) +
(1)
q(P
1
)+(q(P
2
)+q(
op
P
2
)+1)(H
op
P
1
)
(H
op
P
1
K
op
t
op
P
2
) (id
P
1

op
P
2
)
_
from C
3
= (1)
q(P
2
)((

K)
op
P
1
)
(

K)
op
P
1
H
op
P
2
+ (1)
q(P
2
)(H
op
P
1
)
H
op
P
1
(

K)
op
P
2
+
(1)
q(P
1
)+1+(q(P
2
)+1)((

H)
op
P
1
)
(

H)
op
P
1
K
op
P
2
+
(1)
q(P
1
)+(q(P
2
)+1)(K
op
P
1
)
K
op
P
1
(

H)
op
P
2
,
from the third property of Denition 1.1. Summing up, we deduce from B
1
the relation
(

K+K)
op
P
1
P
2
= (1)
q(P
2
)((

K+K)
op
P
1
)
(

K+K)
op
P
1
H
op
P
2
+(1)
q(P
2
)(H
op
P
1
)
H
op
P
1

K+K)
op
P
2
. From C
1
, one of the two terms in the sum vanishes, so that H+

K+K
satises B
2
since H does. Likewise, H +

K + K satises B
1
.
Denition 1.12 Two chain maps H, H

: (, ) (

) given by Denition 1.7 are

said to be primitively homotopic i there exists a primitive homotopy K : (, ) (

)
associated to H such that H

= H +

K +K . They are said to be homotopic i there

are chain maps H
0
, . . . , H
k
: (, ) (

) such that H
0
= H, H
k
= H

and for every

1 i k, H
i1
and H
i
are primitively homotopic.
Let T be an associahedral category. We denote by Ob(K
b
(T)) the set of chain com-
plexes given by Denition 1.4 and by Hom(K
b
(T)) the set of chain maps given by Denition
1.7 modulo homotopies given by Denition 1.12. The category K
b
(T) =
_
Ob(K
b
(T)), Hom(K
b
(T))
_
is the homotopic category of bounded chain complexes of T. We denote with slight abuse
by Ob(K
b
(T)) the empty chain complex.
Denition 1.13 Let (, ) Ob(K
b
(T)). An augmentation of (, ) is a chain map
(, ). A complex equipped with an augmentation is said to be an augmented complex.
2 Particles
2.1 Metric ribbon trees and Stashes associahedron
Following [1], for every integer l 2, we denote by K
l
the space of (connected) metric
trees T
l
satisfying the following three properties.
8
1) Each edge of T
l
has a lengh in [0, 2] and a tree with an edge of length 0 is identied
with the tree obtained by contraction of this edge.
2) The tree T
l
has l +1 monovalent vertices, labeled v
0
, . . . , v
l
where v
0
is the root, and
has no bivalent vertex. The edges adjacent to the l + 1 monovalent vertices have length
one. The tree is oriented from the root to the leaves.
3) For every edge e of T
l
, the partition of v
0
, . . . , v
l

= 0, . . . , l induced by the
connected components of T
l
e only contains cyclically convex intervals of the form [i, j],
i j, or [j, l] [0, i], i j.
This space K
l
has the structure of a (l2)-dimensional convex polytope of the Euclidian
space isomorphic to the associahedron of Stashe, see 1.4 of [1].
Denition 2.1 The elements of K
l
, l 2, are called metric ribbon trees.
We agree that K
1
is a singleton and that the corresponding tree T
1
is a compact interval
of length one. Likewise, we agree that K
0
is a singleton and that the corresponding tree
T
0
is an open-closed interval of length one, isometric to [0, 1), that is with only one vertex,
the root, and only one edge. These trees T
0
and T
1
will also be called metric ribbon trees.
The dihedral group D
l+1
acts on K
l
by cyclic permutation of the labeling. We denote
by
l
: K
l
K
l
the ordered l + 1 automorphism induced by the cyclic permutation
(0, . . . , l) (1, . . . , l, 0). We denote by
l
: K
l
K
l
the ordered two automorphism
induced by the reection which xes zero. It decomposes as the product of transpositions
(1, l)(2, l 1) . . . (
1
2
l,
1
2
l + 1) when l is even and as (1, l)(2, l 1) . . . (
1
2
(l 1),
1
2
(l + 3))
when l is odd, so that
1
2
(l + 1) is xed as well in this case. Finally, when l is odd,
we denote by
l
: K
l
K
l
the ordered two automorphism induced by the reection
(0, l)(1, l 1) . . . (
1
2
(l 1),
1
2
(l + 1)) which has no xed point.
Lemma 2.2 For every l 2,
l
acts as (1)
l
on the pair of orientations of K
l
whereas
the involution
l
acts as (1)

1
2
(l1)
and, when l is odd,
l
acts as (1)
1
2
(l+1)
.
Proof:
When l is odd,
l
xes the comb P represented by Figure 1 where every edge has length
one.
. . . . . .
l 0
1
2
1/2(l-1)
l-1
l-2
1/2(l+1)
e e e e e
1 2 l-2 1/2(l-1) l-3
Figure 1:
9
The tangent space T
P
K
l
of K
l
at the comb P is equipped with the basis (e
1
, . . . , e
l2
),
where for 1 i l 2, e
i
represents the stretch of the edge represented by e
i
on Figure 1,
xing lengths of the other edges. The dierential map of
l
at P xes the vector e1
2
(l1)
and
exchanges the pairs (e
1
, e
l2
), . . . , (e1
2
(l3)
, e1
2
(l+1)
). Its action on the pair of orientations of
T
P
K
l
is thus the same as the determinant of the antidiagonal matrix of order l 2, that
is (1)
1
2
(l+1)
. Still when l is odd,
l
xes the tree T represented by Figure 2. The tangent
.
.
.
.
.
.
0
1
2
l
l-1
1/2(l+1)
1/2(l-1) 1/2(l+3)
e
e
1
1/2(l-3)
f
f1
1/2(l-1)
Figure 2:
space T
T
K
l
of K
l
at this tree T is equipped with the basis (e
1
, . . . , e1
2
(l3)
, f
1
, . . . , f1
2
(l1)
),
where for 1 i
1
2
(l 3), e
i
represents the stretch of the edge represented by e
i
on Figure
2, xing lengths of the other edges, and for 1 j
1
2
(l 1), f
j
represents the stretch of
the length zero edge represented by the vertex of valence four denoted by f
j
on Figure 2.
The dierential map of
l
at T xes all the vectors e
i
and reverses all the vectors f
j
, so
that its action on the pair of orientations of T
T
K
l
is (1)
1
2
(l1)
. Finally, when l is odd,

l
=
l

l
, which nishes the proof of Lemma 2.2 in this case. Likewise, when l is even,

l
acts on the the pair of orientations of K
l
as (1)
1
2
l1
and
l
decomposes as the product
of two elements conjugated to
l
so that it preserves orientations of K
l
.
From now on, for every l 2, we denote by P
l
K
l
the comb represented by Figure 3.
. . .
0
1
2 3 l l-1
e e
1 l-2
Figure 3:
The tangent space T
P
l
K
l
of K
l
at the comb P
l
is equipped with the basis (e
1
, . . . , e
l2
),
10
where for 1 i l 2, e
i
represents the stretch of the edge represented by e
i
on Figure
3, xing lengths of the other edges. From now on, we equip K
l
with the orientation
which turns this basis into a direct one. Every face of K
l
canonically decomposes as a
product of lower dimensional associahedra. In particular, codimension one faces of K
l
decompose as products K
l
1
K
l
2
, where l
1
+l
2
= l +1. More precisely, such a face writes
F = T = T

K
l
[ T

K
l
1
, T

K
l
2
and v

i
= v

0
, where i 1, . . . , l
1
. The only
0
v''
v"
v'
v'
v'
1 i-1
l
1
l
1 2
edge of length two of the tree T = T

F
is the concatenation of the edge adjacent
to v

i
in T

with the one adjacent to the root v

0
in T

. The labeled vertices w

0
, . . . , w
l
of
T satisfy w
j
= v

j
if 0 j i 1, w
j
= v

j+1i
if i j i + l
2
1 and w
j
= v

j+1l
2
if
i + l
2
j l. These facets inherit two orientations, one induced by K
l
and one induced
by the product structure K
l
1
K
l
2
. Lemma 2.3, analogous to Lemma 2.9 of [12], compare
these two orientations.
Lemma 2.3 Let F = T = T

K
l
[ T

K
l
1
, T

K
l
2
and v

i
= v

0
be a
codimension one face of K
l
, where l
1
, l
2
2, l
1
+ l
2
= l + 1 and 1 i l
1
. Then, the
orientations of F induced by K
l
and by K
l
1
K
l
2
coincide if and only if l
1
l
2
+ i(l
2
1)
is odd.
Proof:
Assume rst that i = 1. Then, the concatenation of the comb P
l
1
K
l
1
with the
comb P
l
2
K
l
2
provides the comb P
l
K
l
. The innitesimal contraction of the length
two edge f of P
l
which joins P
l
1
and P
l
2
is an inward vector of K
l
at P
l
F. The
product orientation of K
l
1
K
l
2
= F induces on K
l
the orientation which turns the basis
(f, e

1
, . . . , e

l
1
2
, e

1
, . . . , e

l
2
2
) of T
P
l
K
l
into a direct one. The result follows in this case
from the relation fe

1
e

l
1
2
e

1
e

l
2
2
= (1)
l
1
l
2
+l
2
e

1
e

l
2
2
fe

1
e

l
1
2
.
Now the automorphism
i1
l
sends a facet for which i > 1 to a facet for which i = 1.
This automorphism preserves the decomposition K
l
1
K
l
2
of these facets, induces the
automorphism
i1
l
1
on the rst factor K
l
1
and induces the identity on the second one. The
result thus follows from the rst part and Lemma 2.2.
Denition 2.4 A forest F
l
, l 0, is an ordered nite union T
l
1
T
l
q
of metric
ribbon trees T
l
i
K
l
i
, where q 1, l
1
+ + l
q
= l and l
i
0 for every 1 i q.
For every forest F
l
= T
l
1
T
l
q
given by Denition 2.4, we denote by v

1
, . . . , v

q
the
roots of T
l
1
, . . . , T
l
q
respectively. Likewise, we denote by v
+
1
, . . . , v
+
l
the leaves of T
l
1
T
l
q
11
in such a way that for 1 i q, (v

i
, v
+
l
1
++l
i1
+1
, . . . , v
+
l
1
++l
i
) provides the ordered set of
labeled vertices of T
l
i
.
Let T
q
K
q
, q 2, be a metric ribbon tree and F
l
be a forest made of q trees. Let
T
q
F
l
be the union of T
q
and F
l
where the leaves v
+
1
, . . . , v
+
q
of T
q
are identied with the
roots v

1
, . . . , v

q
of F
l
respectively. If F
l
does not contain the trees T
0
and T
1
, this union
T
q
F
l
is a metric ribbon tree denoted by T
q
F
l
. For every copy of the tree T
1
in F
l
,
the union T
q
F
l
contains an edge of length two adjacent to a leaf. We then divide the
metric of this edge by two to get the metric ribbon tree T
q
F
l
. Likewise, for every copy of
the tree T
0
in F
l
, the union T
q
F
l
contains a non compact edge of length two. We then
remove (or contract) this free edge. In case F
l
only contains trees without leaf, we contract
all such edges except the rst one for which we divide the metric by two. The tree T
q
F
l
coincides with T
0
in this case. The composition T
q
F
l
is dened in the same way when
q = 1 or when T
q
is a forest instead of a tree.
Denition 2.5 Let F
q
and F
l
be forests such that F
l
is made of q trees. The forest
F
q
F
l
just dened above is called the composition of F
q
and F
l
.
2.2 The category of particles
Let M be a smooth manifold.
Denition 2.6 An elementary particle P of M is a pair (x, P
x
), where x is a point
in M and P
x
a linear subspace of T
x
M. The point x is called the based point of P and
denoted by b(P). The dimension (P) of P
x
is called the index of the elementary particle
P.
Denition 2.7 A particle P = P
1
P
q
of M is an ordered nite collection of
elementary particles P
i
of M, 1 i q. The integer q is called the cardinality of the
particle P and denoted by q(P). Its index (P) is the sum of the indices (P
i
), 1 i q.
The empty set is a particle whose cardinality and index vanish.
We denote by Ob(T(M)) the set of particles of M. It is a free monoid for the product
, where we agree that is a unit element. The cardinality and index functions dene
morphisms onto the monoid (N, +).
Denition 2.8 A primitive trajectory from the particle P
+
= P
+
1
P
+
l
to the
particle P

= P

1
P

q
of M, l 0, q 1, is a homotopy class with xed ends of
triples = (u, F
l
, ), such that:
1) F
l
= T
l
1
T
l
q
is a forest with q trees given by Denition 2.4.
2) u : F
l
M is a continuous map which sends the roots v

1
, . . . , v

q
of T
l
1
, . . . , T
l
q
to
the base points b(P

1
), . . . , b(P

q
) respectively and their leaves v
+
1
, . . . , v
+
l
to the base points
b(P
+
1
), . . . , b(P
+
l
) respectively.
3) is an orientation of the real line
_

q
i=1

max
P

i
_

l
j=1

max
P
+
j
_

.
12
In particular, when l = 0, such a primitive trajectory P

reduces to an orientation
of the linear space
q
i=1
P

i
. If = (u, F
l
, ) is a primitive trajectory, we denote by
the primitive trajectory (u, F
l
, ) obtained by reversing the orientation . We agree that
there are no primitive trajectory with target the empty particle except the ones with
empty source. There are two primitive trajectories denoted by id

and id

. For
every particle P of cardinality l 1, we denote by id
P
the primitive trajectory (u, F
l
, ) :
P P for which F
l
is the union of l copies of T
1
, u is constant on every such tree and
is the canonical orientation of the line
max
P (
max
P)

.
Denition 2.9 A trajectory =
iI

i
: P
+
P

from the particle P

+
= P
+
1

P
+
l
to the particle P

= P

1
P

q
of M is a nite collection of primitive trajectories

i
: P
+
P

, i I, modulo the relation () =

P
+
,P
, where
P
+
,P
is the empty
trajectory. The index () of such a trajectory is the dierence (P

) (P
+
) whereas its
cardinality q() equals q(P
+
) q(P

). The trajectory is said to be elementary whenever

its target P

is an elementary particle.
We denote by Hom(T(M)) the set of trajectories given by Denition 2.9. Trajectories
between two particles come equipped with the operation which turns this set into a free
Abelian group with unit the empty trajectory. If

= (u

, F

l
,

) : P
+
1
P

1
and

=
(u

, F

l
,

) : P
+
2
P

2
are two primitive trajectories, we denote by

: P
+
1
P
+
2

P

1
P

2
the primitive trajectory (u

, F

l
F

l
,

). Here,

is the orientation
induced on
_

max
P

1

max
P

2
_

max
P
+
1

max
P
+
2
_

from
_

max
P

1
(
max
P
+
1
)

max
P

2
(
max
P
+
2
)

_
via the isomorphism (
max
P
+
1
)

max
P

=
max
P

2
(
max
P
+
1
)

.
This product is extended to trajectories in such a way that it is distributive with respect to
the addition . Likewise, if

= (u

, F

l
,

) : P
+
P
0
and

= (u

, F

q
,

) : P
0
P

are two primitive trajectories and P

is elementary, we denote by

: P
+
P

the primitive trajectory (u

, F

q
F

l
,

), where the composition of forests is the

one given by Denition 2.5. This operation is extended to primitive trajectories using
the rule (

2
) (

2
) = (1)
(

1
)(

2
)
(

1
) (

2
). It is then extended to
trajectories in such a way that it is distributive with respect to the addition .
Proposition 2.10 Let M be a smooth manifold. The category of particles T(M) =
_
Ob(T(M)), Hom(T(M))
_
is associahedral.
Proof:
First of all T(M) is indeed a category. The composition of morphisms is given by the
operation , it is indeed associative and has unit id
P
for every object P. This category is
small, the preadditive structure is given by the operation and the product : T(M)
T(M) is associative and distributive. Properties one and two of Denition 1.1 are obviously
satised with N = 0. Property three also follows from our denitions.
13
2.3 Coparticles and functor Coecients
The dual coparticle of an elementary particle P = (x, P
x
) is the coparticle P

= (x, T
x
M/P
x
).
Its index satises the relation (P) + (P

) = dimM, so that (

) = dimM. The dual

coparticle of a particle P = P
1
P
l
is P

= P

l
P

1
. We denote by Ob(T

(M))
the set of coparticles of M.
Denition 2.11 A primitive cotrajectory from the coparticle P

= P

1
P

q
to
the coparticle P

= P

1
P

l
of M, l 0, q 1, is a homotopy class with xed ends
of triples

= (u

, F
l
,

), such that:
1) F
l
= T
l
1
T
l
q
is a forest with q trees given by Denition 2.4.
2) u

: F
l
M is a continuous map which sends the roots v

1
, . . . , v

q
of T
l
1
, . . . , T
l
q
to
the base points b(P

1
), . . . , b(P

q
) respectively and their leaves v

1
, . . . , v

l
to the base points
b(P

1
), . . . , b(P

l
) respectively.
3)

is an orientation of the real line

_

l
j=1

max
P

j
_

q
i=1

max
P

i
_

.
A primitive cotrajectory from the coparticle

to the coparticle P

= P

1
P

l
is
an orientation of the real line
_

l
j=1

max
P

j
_
.
Note that when M is oriented, an orientation of the real line
max
P

is the same
as an orientation of (
max
P)

, so that a cotrajectory

: P

is the same as a
trajectory : P

P. A cotrajectory is a nite collection of primitive cotrajectories

i
: P

, i I, modulo the relation

) =

P
+
,P

, compare Denition
2.9. We denote by Hom(T

(M)) the set of cotrajectories of M. The pair T

(M) =
_
Ob(T

(M)), Hom(T

(M))
_
is the dual category of T(M).
Let /od
Z
be the category of free Abelian groups of nite type and /od
1
Z
/od
Z
be
the subcategory of rank one such groups. The functor Coecients provided by Denition
1.3 give functors ( : T(M) /od
1
Z
and (

: T

(M) /od
1
Z
. The two generators of the
rank one free Abelian group Z
P
(resp. Z
P
) associated to a particle P (resp. coparticle
P

) are the two orientations of the real line

max
P (resp
max
P

). When M is oriented,
the restrictions of ( and (

to elementary (co)particles give dual functors.

3 Morse functor
3.1 Conductors
Denition 3.1 Let M be a smooth manifold. A conductor of M is a collection F =
(f
0
, . . . , f
l
; g), where f
j
f
i
: M R are Morse functions with distinct critical points,
0 i < j l, and g is a generic Riemannian metric which is standard near every critical
point of these functions.
Denition 3.2 Let F = (f
0
, . . . , f
l
; g) be a conductor of M. A subconductor of F is a
conductor of the form (f
i
0
, . . . , f
i
q
; g), where 0 i
0
< < i
q
l. We also say that F is
a renement of (f
0
, . . . , f
q
; g) by (f
q+1
, . . . , f
l
; g), where 0 q l.
14
We denote by Ob(((M)) the set of conductors of M.
Denition 3.3 An eective continuation H from the conductor F = (f
0
, . . . , f
q
; g) to
the conductor

F = (

f
0
, . . . ,

f
q
; g) of M is a homotopy (F
t
)
t[0,1]
between the subconductor
F
0
of F associated to I
H
and the subconductor F
1
of F associated to
H
(I
H
). Here, I
H
is
a subset of 0, . . . , q and
H
: I
H
0, . . . , l an increasing injective map.
Denition 3.4 A continuation H : F

F from a conductor F to a conductor

F
of the smooth manifold M is a homotopy class with xed ends of eective continuations
between these conductors.
We denote by Hom(((M)) the set of continuations given by Denition 3.4. These
notions of conductors and continuations are analogous to the one given in [12]. Let H :
F

F be a continuation given by Denition 3.4. The subconductor of

F associated to

H
(I
H
) is called the image of H whereas the subconductor of F associated to I
H
is called
the cokernel. A sequence F
H

F
K

F is called exact when the intersection Im
H
I
K
contains at most one element.
Proposition 3.5 Let M be a smooth manifold. The pair ((M) =
_
Ob(((M)), Hom(((M))
_
has the structure of a small category equipped with the properties of sub-objects, renements,
exact sequences, cokernel and image.
Proof:
The composition of morphisms

HH satises, with the notations of Denition 3.3, the
relation I
e
HH
= I
H

1
H
(I
e
H
). It is obviously associative with units.
3.2 Connections
Let us denote by I the closed interval [a, b], where a < b +. We denote by
H
1,2
(I, R
n
) the Hilbert space of functions of class L
2
with one derivative in L
2
in the sense of
distributions. Let A : I M
n
(R) be a bounded continuous map, we denote by
A
=

t
A
the associated operator H
1,2
(I, R
n
) L
2
(I, R
n
). We denote by H
1,2
0
(I, R
n
) H
1,2
(I, R
n
)
the subspace of functions which vanish at the ends of I (they are of class C
1
2
loc
from Sobolevs
Theorem). This is a closed subspace of vanishing index when I = R, index n when one
of the extremities a or b is innite and index 2n when I is compact. For every invertible
diagonalisable matrix B GL
n
(R), we denote by
+
(B) (resp.

(B)) the number of its

positive (resp. negative) eigenvalues, so that
+
(B) +

(B) = n.
Lemma 3.6 Let A : R M
n
(R) be a constant path with value an invertible diagonal-
isable matrix. Then, the operator
A
: H
1,2
(R, R
n
) L
2
(R, R
n
) is an isomorphism.
Proof:
Without loss of generality, we may assume that n = 1. The operator then writes

A
=

t
a, a R

, and is injective. The inverse operator writes w L

2
(R, R
n
)
15
_
+
0
w(t s) exp(sa)ds H
1,2
(R, R
n
) when a < 0 and w L
2
(R, R
n
)
_
0

w(t
s) exp(sa)ds H
1,2
(R, R
n
) when a > 0.
Lemma 3.7 1) Let A : R
+
M
n
(R) be a continuous map which converges to an in-
vertible diagonalisable matrix A

at innity. Then, the associated operator

A
: H
1,2
(R
+
, R
n
)
L
2
(R
+
, R
n
) is Fredholm of index

(A

).
2) Let A : [a, b] M
n
(R) be a continuous map, where < a < b < +. Then,
A
is Fredholm of index n.
Proof:
In order to prove the second part, it suces to prove that the restriction of
A
to
the subspace H
1,2
0
(I, R
n
) is Fredholm of index n. This restriction is injective. More-
over, every v H
1,2
(I, R
n
) satises the estimate |v|
H
1,2 = |

t
v|
L
2 +|v|
L
2 |
A
(v)|
L
2 +
(sup
[a,b]
|A|+1)|v|
L
2. From this elliptic estimate follows that the image of
A
is closed. In-
deed, let w = lim
n+

A
(v
n
) L
2
([a, b], R
n
) be a point in the closure of Im(
A
). If (v
n
)
nN
is a bounded sequence in H
1,2
0
(I, R
n
), then it lies in a compact subset of L
2
([a, b], R
n
) from
Rellichs theorem. From the above elliptic estimate follows that a subsequence of (v
n
)
nN
is of Cauchy type and thus converges to v H
1,2
0
(I, R
n
). Hence, w =
A
(v) lies in the
image of
A
. If (v
n
)
nN
is not bounded, we divide it by its norm and get in the same
way a sequence converging to v H
1,2
0
(I, R
n
) of norm one and in the kernel of
A
, which
is impossible. It remains to prove that the orthogonal complement of the image of
A
restricted to H
1,2
0
(I, R
n
) is of dimension n. Let w be a point in this complement. Then, for
every v H
1,2
(I, R
n
),
_
b
a

v
t
A(v), w
_
dt = 0, so that
_
b
a

v
t
, w
_
dt =
_
b
a
v, A

(w)) dt.
As a consequence, the derivative of w in the sense of distributions is of class L
2
, so that
w H
1,2
([a, b], R
n
). Moreover, an integration by parts shows that
w
t
+ A

(w) = 0, since
v vanishes at a and b. The kernel of this adjoint operator

A
is n-dimensional made of the
solutions of the linear system
w
t
= A

(w).
Let us now prove the rst part of Lemma 3.7. The restriction of
A
to H
1,2
0
(R
+
, R
n
)
is injective. It suces to prove that this restriction has closed image. Indeed, the ker-
nel of the adjoint operator is made of solutions of the linear system
w
t
= A

(w), so
that it is of nite dimension bounded by n. Hence,
A
is Fredholm. There exists a con-
tinuous path (A
s
)
s[0,1]
of bounded continuous maps R
+
M
n
(R) converging to A

at
innity such that A
0
= A and A
1
A

is constant. The kernel of

A
1
is of dimen-
sion
+
(A

) so that ind(
A
) = ind(
A
1
) = n
+
(A

) =

(A

). To prove that
the restriction of
A
to H
1,2
0
(R
+
, R
n
) has closed image we follow the scheme of Floer
[2], compare [10]. Let M > 1 and
+
M
: R
+
R be a strictly increasing smooth
function such that
+
M
() = 0 if M and
+
M
() = 1 if M + 1. We set

0
M
= 1
+
M
. Let B
M
: R M
n
(R) be a continuous map such that B
M
() = A()
if M and sup
R
|B
M
() A

| sup
M
|A() A

|. From Lemma 3.6, there

exists M 0 such that
B
M
: H
1,2
(R, R
n
) L
2
(R, R
n
) is an isomorphism and we
choose such an M. There exist then constants C
+
1
, C
+
2
> 0 such that for every v
H
1,2
(R
+
, R
n
), |
+
M
v|
H
1,2 C
+
1
|
B
M
(
+
M
v)|
L
2 = C
+
1
|
A
(
+
M
v)|
L
2 C
+
1
|
A
(v)|
L
2
(R
+
,R
n
)
+
C
+
2
|v|
L
2
([0,M+1],R
n
)
. Likewise, as before, there exist constants C
0
1
, C
0
2
> 0 such that
16
for every v H
1,2
(R
+
, R
n
), |
0
M
v|
H
1,2 |
A
(
0
M
v)|
L
2
(R
+
,R
n
)
+ C
0
1
|
0
M
v|
L
2
([0,M+1],R
n
)

|
A
(v)|
L
2
(R
+
,R
n
)
+C
0
2
|v|
L
2
([0,M+1],R
n
)
. Summing up, we deduce the elliptic estimate |v|
H
1,2
(C
+
1
+1)|
A
(v)|
L
2
(R
+
,R
n
)
+(C
+
2
+C
0
2
)|v|
L
2
([0,M+1],R
n
)
, which implies the result as before.
Now, let T
l
K
l
be a metric ribbon tree, where l 2. We denote by c
T
l
its set of
edges and by 1
T
l
its set of vertices. Every bounded edge of T
l
is by denition isometric
to exactly one level of the function (x, y) [0, 1]
2
xy [0, 1], the level zero if it is
of length two and one if it is of length zero for example. We equip these edges with the
measure d induced by the gradient of this function, so that the measure is innite when
they are of length two. Likewise, the edge adjacent to the root v
0
(resp. leaves v
1
, . . . , v
l
)
is isometric to the x-axis (resp. y-axis) of [0, 1]
2
and equipped with the induced innite
measure. This is called a gluing prole in [7]. We denote by H
1,2
(T
l
, R
n
) the Hilbert
space of functions of class L
2
with one derivative in L
2
in the sense of distributions for the
measure d and by H
1,2
0
(T
l
, R
n
) H
1,2
(T
l
, R
n
) the subspace of functions which vanish at
the vertices of T
l
. The latter is canonically isomorphic to the product
eE
T
l
H
1,2
0
(e, R
n
).
For every continuous map A : T
l
M
n
(R), we denote by
A
=

t
A the associated
operator H
1,2
(T
l
, R
n
) L
2
(T
l
, R
n
).
Lemma 3.8 Let T
l
K
l
be a metric ribbon tree and A : T
l
M
n
(R) be a continuous
map, where l 0. Assume that for 0 i l, the value A(v
i
) of A at the leaf v
i
is
an invertible diagonalisable matrix A
i
and that the same holds at the middle of the length
two edges of T
l
. Then, the operator
A
: H
1,2
(T
l
, R
n
) L
2
(T
l
, R
n
) is Fredholm of index

+
(A
0
)

l
i=1

+
(A
i
).
Proof:
When l = 0, the map x R
+
x R

conjugates the operator

A
to the oper-
ator
A
: H
1,2
(R
+
, R
n
) L
2
(R
+
, R
n
) which is Fredholm of index

(A
0
) =
+
(A
0
)
from Lemma 3.7. When l 2, the restriction of
A
to H
1,2
0
(T
l
, R
n
) is Fredholm of in-
dex
+
(A
0
) +

l
i=1

(A
i
) n#c
T
l
from Lemma 3.7. Thus,
A
is Fredholm of index

+
(A
0
) +

l
i=1

(A
i
) n(#1
T
l
#c
T
l
l 1) =
+
(A
0
) +

l
i=1

(A
i
) nl. The result
follows along the same lines when l = 1.
Let
A
: H
1,2
(T
l
, R
n
) L
2
(T
l
, R
n
) be an operator given by Lemma 3.8. Following the
appendix of [3], we denote by det(
A
) the real line (
max
ker
A
) (
max
coker
A
)

. For
every 0 i l, we denote by P
i
R
n
the maximal linear subspace invariant under A
i
on
which A
i
has positive eigenvalues, so that dim(P
i
) =
+
(A
i
).
Lemma 3.9 Let
A
: H
1,2
(T
l
, R
n
) L
2
(T
l
, R
n
) be an operator given by Lemma 3.8 and
for every 0 i l, P
i
R
n
be the maximal linear subspace invariant under A
i
on which
A
i
has positive eigenvalues. Then, the real lines det(
A
) and (
max
P
0
) (
l
i=1

max
P
i
)

are canonically isomorphic.

Proof:
17
Let us rst assume that l 2 and that T
l
does not contain any edge of length two. For
every 0 i l, we denote by e
i
the length one edge adjacent to the leaf v
i
. We deduce
a decomposition T
l
= T

l
i=0
e
i
, where T

is a tree of nite measure. We then deduce a

short exact sequence of complexes from the sequence 0 H
1,2
(T
l
, R
n
) H
1,2
(T

, R
n
)

l
i=0
H
1,2
(e
i
, R
n
)
l
i=0
R
n
i
0, where the injective map is the restriction map to T

and the closure e

i
of e
i
and the surjective map is the evaluation map at the intersections
point T

e
i
of the dierence between the two functions. The isomorphism det(
A
[
T
)

l
i=0
det(
A
[
e
i
)

= det(
A
) det(R
n
i
) follows, see the appendix of [3]. Now the evaluation
map at T

e
0
provides an isomorphism between det(
A
[
T
) and det(R
n
0
) since from Lemma
3.7,
A
[
T
is of index n and it has an n-dimensional kernel determined by the initial condition
at T

e
0
. Likewise, evaluation at T

e
0
provides an isomorphism between det(
A
[
e
0
)
and
max
P
0
whereas evaluation at T

e
i
, 1 i l, provides an isomorphism between
det(
A
[
e
i
) det(R
n
i
)

and (
max
P
i
)

. Hence the result in this case. The result follows

likewise from Lemma 3.7 when l = 0 or 1 and by concatenation when T
l
contains edges of
length two.
3.3 Gradient ow trajectories and Morse complexes
Let M be a closed smooth manifold and F = (f
0
, . . . , f
l
; g) Ob(((M)) be a conductor
of M. For every 0 i < j l and every critical point x of f
j
f
i
, we denote by P
x
the elementary particle based at the point x whose linear space is the +1 eigenspace of
the Hessian bilinear form Hess
f
j
f
i
of f
j
f
i
for the metric g. We set CM(f
i
, f
j
) =

xCrit(f
i
,f
j
)
P
x
and then for every 1 q l, CM
q
(F) =
0i
0
<<i
q
l
CM(f
i
0
, f
i
1
)
CM(f
i
q1
, f
i
q
). Finally, we set CM(F) =
l
q=1
CM
q
(F).
Denition 3.10 Let F = (f
0
, . . . , f
l
; g) Ob(((M)) be a conductor of the smooth
manifold M. Let P
x
1
P
x
q
CM(f
i
0
, f
i
1
) CM(f
i
q1
, f
i
q
) and P
x
0
CM(f
i
0
, f
i
q
)
be such that (P
x
0
) (P
x
1
P
x
q
) = 2 q. A gradient ow trajectory : P
x
1

P
x
q
P
x
0
of F is a primitive trajectory (u, T, ) such that:
1) T K
q
is a metric ribbon tree with a root and q leaves.
2) The restriction of u : T M to every edge e of T satises for every e the
equation
u

() =
g
(f
j
e
f
i
e
)(u()), where d is the measure element on T and [i
e
+
1, j
e
]([j
e
+1, l][0, i
e
]) is the partition of the leaves v
0
, . . . , v
q

= 0, . . . , q into cyclically
convex intervals induced by e.
3) is an orientation of the real line det(

), where

: H
1,2
(T, u

TM) L
2
(T, u

TM)
is the connection whose restriction to every edge e of T is the rst variation v

v
Hess
f
j
e
f
i
e
(v). If q > 1,

is injective and has a (q 2)-dimensional cokernel canonically

isomorphic to the tangent space T
T
K
q
. The orientation is then the one induced from
this isomorphism and the one chosen on K
q
in 2.1. If q = 1,

is surjective and its one

dimensional kernel is generated by
u

, is then the orientation induced by this vector.

Recall that the tree T K
q
is oriented from the root to the leaves. Note also that from
Lemma 3.9, det(

) is canonically isomorphic to (
max
P
x
0
) (
l
i=1

max
P
x
i
)

so that a
18
gradient ow trajectory given by Denition 3.10 is indeed a primitive trajectory in the
sense of Denition 2.8.
From the genericness assumptions on g and the compactness theorem in Morse the-
ory, see [10], the conductor F has only nitely many gradient ow trajectories. We de-
note by m
q
: CM
q
(F) CM
1
(F) the sum of all these trajectories. Let then
CM
:
CM(F) CM(F) be the morphism induced from the opposite of these trajectories and
axioms A
1
, A
2
, A
3
of Denition 1.4. Hence, the restriction of
CM
to CM
q
(F) equals

q
l
2
=1
(1)
ql
2

q
1
i=1
(1)
i(l
2
1)
id
i1
m
l
2
id
q
1
i
, where q
1
+ l
2
1 = q, 1 q l. The
following theorem is due to Morse, Witten and Fukaya, see [5], [6], [10].
Theorem 3.11 Let M be a closed smooth manifold and F Ob(((M)) be a conductor
of M. Then,
CM

CM
= 0, so that (CM(F),
CM
) Ob(K
b
(T(M))).
Proof:
Let P
+
and P

in CM(F) be particles of cardinalities q

and q
+
such that (P

)q

=
(P
+
) q
+
+ 2, so that the space /(P
+
, P

) of pairs (u, T) satisfying properties 1 and

2 of Denition 3.10 is one-dimensional. From the compactness and glueing theorems in
Morse theory, see [10], [7], the union of primitive trajectories from P
+
to P

counted by

CM

CM
is in bijection with the boundary of this space. It suces thus to prove that
every such trajectory induces the outward normal orientation on /(P
+
, P

). Indeed, two
boundary components of a same connected component of /(P
+
, P

) then provide the

same trajectory but with opposite orientations, which implies the result. From Lemmas
1.5 and 1.6, we can assume that q

= 1, that is P

is elementary. Let
1

2
be such a
trajectory counted by
CM

CM
, P
0
= t(
2
) = s(
1
) and q
0
be the cardinality of P
0
. By
denition,
CM

CM
counts
1

2
with respect to the sign 1 + q
+
l
2
+ i(l
2
1), where
l
2
= q
+
+1 q
0
, and with respect to the orientation induced by the product K
q
0 K
l
2
on
the corresponding face of the associahedron K
q
+. From Lemma 2.3, this orientation diers
from the one induced by K
q
+ exactly from the quantity 1+q
0
l
2
+i(l
2
1) = 1+q
+
l
2
+i(l
2
1)
mod (2). The result follows when q
0
, l
2
2. But the latter relation also remains valid when
q
0
or l
2
= 1. Indeed, when q
0
= 1 (resp. l
2
= 1), the associated connection

1
(resp.

2
) is surjective and its one dimensional kernel is oriented such that it coincides with
the outward (resp. inward) normal vector of /(P
+
, P

). This follows by derivation with

respect to R of the preglueing relation u
1

R
u
2
() = u
1
( + R R
0
)
R
0
u
2
( R + R
0
)
and from our convention to orient from target to source. When q
0
= 1, the result is now
obvious and when l
2
= 1, it follows from the relation k = (1)
q
0
1
k, where k is
an orientation of K
q
0 and the outward normal vector of /(P
+
, P

).
Denition 3.12 Let M be a closed smooth manifold and F Ob(((M)) be a conductor
of M. The complex /(F) = (CM(F),
CM
) Ob(K
b
(T(M))) given by Theorem 3.11 is
called the Morse complex associated to F.
19
3.4 Morse continuations
3.4.1 Stashes multiplihedron
For every integer l 2, denote by J
l
the space of metric ribbon trees with l +1 free edges
given by Denition 2.1 which are painted in the sense of Denition 2.3.1 of [9], compare
Denition 5.1 of [4]. The trees are painted from the root to the leaves in such a way that
for every vertex v of the tree, the amount of painting is the same on every subtree rooted
at v with respect to the measure d. The space J
l
is a compactication of ]0, 1[

Kl
which has the structure of a (l 1)-dimensional convex polytope of the Euclidian space
isomorphic to Stashes multiplihedron, see [11], [9], [4] and references therin. We agree
that J
0
= J
1
= [0, 1]. Forgetting the painting provides a map for
l
: J
l
K
l
, we orient
its bers in the sense of propagation of the painting. This induces a product orientation
on

Jl
=]0, 1[

Kl
. The codimension one faces of J
l
dierent from 0 K
l
and 1 K
l
are of two dierent natures, see [8], [4]. The lower faces are canonically isomorphic to
products J
l
1
K
l
2
, l
1
+l
2
= l +1, they encode ribbon trees having one edge of length two
whose upper half is unpainted, see Denition 2.2 of [4]. The upper faces are canonically
isomorphic to products K
q
J
l
1
J
l
q
, q 1, they encode ribbon trees having q edges
of length two, whose lower half is painted, see Denition 2.3 of [4]. These faces inherit two
orientations, one from J
l
and one from the product structure. The following Lemmas 3.13
and 3.14, analogous to Lemmas 3.4 and 3.5 of [12], compare them.
Lemma 3.13 Let L = T = T

J
l
[ T

J
l
1
, T

K
l
2
and v

i
= v

0
be a lower
facet of the multiplihedron J
l
, where l
1
+ l
2
= l + 1, l
1
, l
2
2 and 1 i l
1
. Then, the
orientations of L induced by J
l
and by J
l
1
K
l
2
coincide if and only if l
1
l
2
+ i(l
2
1) is
even.
Proof:
This result follows from Lemma 2.3. Indeed, at the innitesimal level, ]0, 1[(K
l
1

K
l
2
) = J
l
1
K
l
2
whereas ]0, 1[K
l
= J
l
.
Lemma 3.14 Let U = TT
1
T
q
J
l
[ T K
q
, T
i
J
l
i
and v
i
= v
i
0
, 1 i q
be an upper facet of the multiplihedron J
l
, where l
1
+ +l
q
= l, q 2 and l
i
1 for every
1 i q. Then, the orientations of U induced by J
l
and K
q
J
l
1
J
l
q
coincide if
and only if

q
i=1
(q i)(l
i
1) is even.
Proof:
Let T
q
P
l
1
P
l
q
U, where T
q
K
q
is a comb represented by Figure 3 which
is completely painted and P
l
i
J
l
i
is a comb represented by Figure 3 which is painted
up to a level s
i
. Let T be an interior point of J
l
close to T
q
P
l
1
P
l
q
so that its
edges joining T
q
to P
l
1
, . . . , P
l
q
have length slightly less than two. The level of painting
s of T is an increasing function of s
i
and the length f
i
of the edge joining T
q
to P
l
i
,
1 i q. The interior of U writes

U
=

Kq
(]0, 1[
1

Kl
1
) (]0, 1[
q

Kl
q
) where
each factor ]0, 1[
i
is equipped with the orientation

s
i
, 1 i q. Since for xed s,
20
increasing f
i
amounts to decrease s
i
, this factor ]0, 1[
i
gets identied at T to the length
of the edge joining T
q
to P
l
i
but with the orientation

f
i
whereas

s
corresponds to
the outward normal vector to U. We thus have to compare the orientations of

Kl
and

Kq
(]0, 1[
1

Kl
1
) (]0, 1[
q

Kl
q
) at T, where ]0, 1[
i
is equipped with the orientation

f
i
. For this purpose, we proceed by nite induction. The glueing of T
q
with P
l
1
gives
the comb P
q+l
1
1
. However, writing (e
1
, . . . , e
q+l
1
3
) the direct basis of T
P
q+l
1
1
K
q+l
1
1
given by Figure 3, we see that

Kq
(]0, 1[
1

Kl
1
) comes equipped with the direct basis
(e
l
1
, . . . , e
q+l
1
3
, e
l
1
1
, e
1
, . . . , e
l
1
2
), so that the orientations of K
q
J
l
1
and K
q+l
1
1
dier
from (1)
(l
1
1)(q1)
.
Assume now that the orientations of K
q
J
l
1
J
l
i1
and K
q+(l
1
1)++(l
i1
1)
are
compared and let us compare the orientations of K
q+(l
1
1)++(l
i1
1)
J
l
i
and K
q+(l
1
1)++(l
i
1)
.
This general case diers from the rst one because we now glue the comb P
q+(l
1
1)++(l
i1
1)
with P
l
i
at the vertex v
I
of P
q+(l
1
1)++(l
i1
1)
, where I = 1+l
1
+ +l
i1
, so that the result
is no more a comb. We can however reduce this case to the rst one using the same trick as
in Lemma 2.3, that is relabeling the vertices. Indeed, the automorphism
I1
q+(l
1
1)++(l
i
1)
of K
q+(l
1
1)++(l
i
1)
given by Lemma 2.2 sends our facet onto a facet for which I = 1. This
automorphism preserves the product structure of the facets and induces the automorphism

I1
q+(l
1
1)++(l
i1
1)
on K
q+(l
1
1)++(l
i1
1)
and the identity on J
l
i
. From Lemma 2.2, this
automorphism
I1
q+(l
1
1)++(l
i
1)
contributes as (1)
(l
i
1)(I1)
to the sign we are comput-
ing. From the preceding case, we deduce that the orientations of K
q+(l
1
1)++(l
i1
1)
J
l
i
and K
q+(l
1
1)++(l
i
1)
dier from (1)
(l
i
1)(I1+q+(l
1
1)++(l
i1
1)1)
= (1)
(l
i
1)(qi)
. The
result follows by summation.
3.4.2 Morse continuations
Let H be an eective continuation from the conductor F = (f
0
, . . . , f
l
; g) Ob(((M)) to
the conductor

F = (

f
0
, . . . ,

f

l
; g) of M, see 3.1. Restricting ourselves to the subconductor
of

F image of H, we may assume that

l = l and I
H
= 0, . . . , l, so that H is just a
homotopy F
t
= (f
t
0
, . . . , f
t
l
; g
t
), t [0, 1], between F
0
= F and F
1
=

F. Let 1 q l,
0 i
0
< < i
q
l and P
x
1
P
x
q
CM(

f
i
0
,

f
i
1
) CM(

f
i
q1
,

f
i
q
) CM
q
(

F).
Let P
x
0
CM(f
i
0
, f
i
q
) be such that (P
x
0
) (P
x
1
P
x
q
) = 1 q. From the
genericness assumptions on g and the compactness theorem in Morse theory, see [10], [7],
there are only nitely many primitive trajectories = (u, T, ) : P
x
1
P
x
q
P
x
0
such that:
1) T J
q
is a metric ribbon tree with a root and q leaves, painted up to a level s.
2) The restriction of u : T M to every edge e of T satises for every e the
equation
_
_
_
u

() =
g
(

f
j
e


f
i
e
)(u()) if e is unpainted,
u

() =
g
(f
j
e
f
i
e
)(u()) if e is below the level s 1,
u

() =
g
t (f
t
j
e
f
t
i
e
)(u()) if e is at the level s t,
21
where d is the measure element on T, the levels s t are with respect to this measure
and [i
e
+ 1, j
e
] ([j
e
+ 1, l] [0, i
e
]) is the partition of the leaves v
0
, . . . , v
q

= 0, . . . , q
into cyclically convex intervals induced by e.
3) is an orientation of the real line det(

), where

: H
1,2
(T, u

TM) L
2
(T, u

TM)
is the connection whose restriction to every edge e of T is the rst variation
_

_
v()

v() Hess

f
j
e

f
i
e
(v()) if e is unpainted,
v()

v() Hess
f
j
e
f
i
e
(v()) if e is below the level s 1,
v()

v() Hess
f
t
j
e
f
t
i
e
(v()) if e is at the level s t.
This connection

is injective and has a (q 1)-dimensional cokernel canonically iso-

morphic to the tangent space T
T
J
q
. The orientation is then the one induced from this
isomorphism and the one chosen on J
q
in 3.4.1. From Lemma 3.9 we know that det(

)
is canonically isomorphic to (
max
P
x
0
) (
l
i=1

max
P
x
i
)

so that these triple are indeed

primitive trajectories in the sense of Denition 2.8.
Denition 3.15 These trajectories are the gradient trajectories of the continuation
H : F

F.
We denote by H
q
: CM
q
(

F) CM
1
(F) the sum of all these trajectories of continuation.
Let then /(H) : CM(

F) CM(F) be the morphism induced by these elementary

trajectories and axioms B
1
, B
2
of Denition 1.7. The restriction of /(H) to CM
q
(

F)
CM
q
(F) equals
l
1
++l
q
= q
(1)
P
q
j=1
(qj)(l
j
1)
H
l
1
H
l
q
. In case

l > l, we extend
/(H) by zero outside of the subcomplex associated to the subconductor of

F image of
H. The image of /(H) is the subconductor of CM(F) associated to the cokernel of H.
The following theorem is due to Morse, Witten and Fukaya, see [5], [10].
Theorem 3.16 Let M be a closed smooth manifold and H be an eective continuation
from the conductor F to the conductor

F of M. Then, /(H) : CM(

F) CM(F) is a
chain map.
Proof:
Let P
+
CM(

F) and P

CM(F) be particles of cardinalities q

+
and q

such that
(P

) q

= (P
+
) q
+
+ 1, so that the space /(P
+
, P

; H) of gradient trajectories
of H from P
+
to P

is one-dimensional. From the compactness and glueing theorems in

Morse theory, see [10], [7], the union of primitive trajectories from P
+
to P

counted by
/(H)
CM

CM
/(H) is in bijection with the boundary of this space. It suces thus to
prove that every such trajectory induces the outward normal orientation on /(P
+
, P

; H).
Indeed, two boundary components of a same connected component of /(P
+
, P

; H) then
provide the same trajectory but with opposite orientations, which implies the result. From
Lemma 1.9, we can assume that q

= 1, that is P

is elementary. Let
1

2
be such a
trajectory counted by
CM
/(H), P
0
= t(
2
) = s(
1
) and q
0
be the cardinality of P
0
.
By denition,
CM
/(H) counts
1

2
with respect to the sign (1)
1+
P
q
0
j=1
(q
0
j)(l
j
1)
,
22
where l
1
+ + l
q
0 = q
+
, and with respect to the orientation induced by the product
K
q
0 J
l
1
J
l
q
0
on the corresponding upper face of the multiplihedron J
q
+. From
Lemma 3.14, this orientation diers from the one induced by J
q
+ exactly from the quantity
(1)
P
q
0
j=1
(q
0
j)(l
j
1)
, so that
1

2
is counted by
CM
/(H) with respect to the sign 1.
This result remains valid when q
0
= 1, since in the neighborhood of the broken trajectory,
elements of /(P
+
, P

; H) write u
1

R
u
2
( +R) = u
1
( +2RR
0
)
R
0
u
2
( +R
0
). Thus,
lim
R+

u
1

R
u
2
( +R) = 2
u
1

and the vector

u
1

which orients m
1
coincides with the
outward normal orientation of /(P
+
, P

; H).
Likewise, /(H)
CM
counts a trajectory
1

2
with respect to the sign (1)
q
0
l
2
+i(l
2
1)
,
where q
0
+ l
2
= q
+
+ 1, and with respect to the orientation induced by the product
J
q
0 K
l
2
on the corresponding lower face of the multiplihedron J
q
+. From Lemma 3.13,
this orientation diers from the one induced by J
q
+ exactly from the same quantity, so that

1

2
is counted by /(H)
CM
with respect to the sign +1. This result remains valid
when l
2
= 1, since in the neighborhood of the broken trajectory, elements of /(P
+
, P

; H)
write u
1

R
u
2
( R) = u
1
( R
0
)
R
0
u
2
( 2R+R
0
). Thus, lim
R+

u
1

R
u
2
( R) =
2
u
2

and the vector

u
2

which orients m
1
coincides with the inward normal orientation
of /(P
+
, P

; H). The result follows from the relation k () = (1)

q
0
k, where k is
an orientation of J
q
0 and the inward normal vector of /(P
+
, P

; H).
3.5 Morse functor
Theorem 3.17 Let M be a closed smooth manifold. If H
0
, H
1
are two eective con-
tinuations from the conductor F to the conductor

F of M, then /(H
0
) and /(H
1
) are
chain homotopic maps CM(

F) CM(F). Likewise, if H
0
: F
0
F
1
and H
1
: F
1
F
2
are eective continuations, then /(H
1
H
0
) = /(H
0
) /(H
1
).
Theorem 3.17 means that / quotients out to a map Hom(((M)) Hom(K
b
(T(M)))
which together with the map given by Denition 3.12 provide a contravariant functor
/ : ((M) K
b
(T(M)).
Denition 3.18 The contravariant functor / : ((M) K
b
(T(M)) is called the
Morse functor.
Proof of Theorem 3.17:
Let (H
r
)
r[0,1]
be a generic homotopy between H
0
and H
1
. Equip [0, 1] J
l
with
the product orientation. In the same way as we dened the morphisms /(H) in the
previous paragraph, we dene, for every 1 q l, 0 i
0
< < i
q
l and r
[0, 1], the morphisms K
r
q
: CM
q
(

F) CM
1
(F) as the sum of gradient trajectories of the
continuation H
r
from a particle P
+
CM
q
(

F) to an elementary particle P

CM
1
(F)
such that (P

) (P
+
) = q. These trajectories can only appear for a nite set of
values of r ]0, 1[ since the homotopy (H
r
)
r[0,1]
is generic. They are oriented from the
orientation just xed on [0, 1] J
l
since the connection

is injective and has now a

q-dimensional cokernel canonically isomorphic to the tangent space T
T
([0, 1] J
q
). Then,
23
we denote by K : CM(

F) CM(F) the morphism dened by K =

l
q=1
(1)
q

l
1
++l
q
= q
(1)
P
q
j=1
(qj)(l
j
1)

q
i=1
(1)
P
i1
j=1
(l
j
1)
_
1
0
H
r
l
1
H
r
l
i1
K
r
l
i
H
r
l
i+1
H
r
l
q
, where the
integral is taken with respect to the counting measure. This morphism K is a homotopy
in the sense of Denition 1.12 between /(H
0
) and /(H
1
). In other words, assuming
that only one value r
0
of r ]0, 1[ appear, K is the morphism satisfying axioms C
1
, C
2
, C
3
of Denition 1.10 whose opposite is given on elementary particles by K
op
q
, 1 q l and
which satises the relation /(H
0
) /(H
1
) =
CM
K + K
CM
. Indeed, from the
genericness of the homotopy (H
r
)
r[0,1]
, there is only one gradient trajectory
0
: P
+
0

P

0
of the continuation H
r
0
. The elementary target particle P

0
belongs to a complex
CM(f
i
0
1
, f
j
0
). The morphism (K
CM
)
op
is non trivial only on P

0
whereas (
CM
K)
op
is non trivial only on complexes of the form CM(f
i1
, f
j
) with i i
0
and j j
0
. For every
particle P
1
P
q
CM(F), at most one elementary member belongs to CM(f
i
0
1
, f
j
0
)
or such CM(f
i1
, f
j
), so that axiom C
1
is satised. Axiom C
2
is by denition satised as
well as axiom C
3
with H = H
r
0
or any H
r
, r [0, 1] actually.
Let P
+
CM(

F) and P

CM(F) be particles of cardinalities q

+
and q

such that
(P

) q

= (P
+
) q
+
, so that the space /(P
+
, P

; K) of gradient trajectories of H
r
,
r [0, 1], from P
+
to P

is one-dimensional. From the compactness and glueing theorems

in Morse theory, see [10], [7], the union of primitive trajectories from P
+
to P

counted
by /(H
1
) /(H
0
) +
CM
K +K
CM
is in bijection with the boundary of this space.
It suces thus to prove that every such trajectory induces the outward normal orientation
on /(P
+
, P

; K). Indeed, two boundary components of a same connected component

of /(P
+
, P

; K) then provide the same trajectory but with opposite orientations, which
implies the result. From Lemma 1.11, we can assume that q

= 1, that is P

is elementary.
Let
1

2
be such a trajectory counted by
CM
K, P
0
= t(
2
) = s(
1
) and q
0
be
the cardinality of P
0
. By denition,
CM
K counts
1

2
with respect to the sign
(1)
1+q
0
+
P
q
0
j=1
(q
0
j)(l
j
1)+
P
i1
j=1
(l
j
1)
, where l
1
+ + l
q
0 = q
+
, and with respect to the
orientation induced by the product K
q
0 J
l
1
J
l
i1
([0, 1] J
l
i
)J
l
i+1
J
l
q
0
on
the corresponding face of [0, 1] J
q
+. From Lemma 3.14, this orientation diers from the
one induced by [0, 1] J
q
+ exactly from the quantity (1)
P
q
0
j=1
(q
0
j)(l
j
1)+
P
i1
j=1
(l
j
1)+q
0
1
,
so that
1

2
is counted by
CM
K with respect to the sign +1. The last 1 comes from
the commutation between the [0, 1]-factor and the outward normal vector. This remains
valid when q
0
= 1 for the same reason as in the proof of Theorem 3.16.
Likewise, K
CM
counts a trajectory
1

2
with respect to the sign (1)
q
0
l
2
+i(l
2
1)+1
,
where q
0
+ l
2
= q
+
+ 1, and with respect to the orientation induced by the product
[0, 1]J
q
0 K
l
2
on the corresponding face of [0, 1]J
q
+. From Lemma 3.13, this orientation
diers from the one induced by [0, 1] J
q
+ exactly from the same quantity, so that
1

2
is counted by /(H)
CM
with respect to the sign +1. The last +1 comes once more from
the commutation between the [0, 1]-factor and the outward normal vector and the result
remains valid when q
0
= 1 for the same reason as in the proof of Theorem 3.16. Hence the
rst part of Theorem 3.17.
Now, from Lemma 1.8, the composition /(H
0
) /(H
1
) is a chain map in the sense
24
of Dention 1.7 and from Lemma 1.9, in order to prove the second part of Theorem
3.17, we just have to check that /(H
1
H
0
) and /(H
0
) /(H
1
) are homotopic on
elementary members. Let
1

2
be an elementary trajectory counted by /(H
0
)/(H
1
),
P
0
= t(
2
) = s(
1
) and q
0
be the cardinality of P
0
. We write
2
= H
1
l
1
H
1
l
q
0
, where
l
1
+ + l
q
0 is the cardinality of the source of
2
. By denition, /(H
0
) /(H
1
) counts

1

2
with respect to the sign (1)
P
q
0
j=1
(q
0
j)(l
j
1)
and with respect to the orientation
induced by the product J
q
0 J
l
1
J
l
q
0
. This product is a face of the space J
2
l
which
encodes painted metric ribbon trees with l + 1 free edges among which one root and with
interior edges having lengths in [0, 2]. But this time, these trees are painted with two
dierent colors, one which encodes the propagation of the homotopy H
0
and the other one
which encodes the propagation of the homotopy H
1
. Thus, J
2
l
is the subset of the ber
product J
l

K
l
J
l
made of pairs of painted trees with same underlying tree and such that
the rst one is painted up to a lower level than the second one. From the glueing theorem in
Morse theory, see [10], [7], this trajectory
1

2
is a boundary point of a one dimensional
manifold whose underlying painted tree belongs to the space J
2
l
. The outward normal
direction in this manifold is given by increasing the dierence between the two colors 1
and 0. Moreover, from the compactness theorem in Morse theory, see [10], [7], for a value
of this dierence close to innity, any such trajectory belongs to one such manifold. Let us
choose a generic value s
1
s
0
close to innity. We can dene a morphism of continuation

/(H
1
H
0
) in the same way as in Denition 3.15 by counting gradient trajectories of
H
1
H
0
with dierence of painting s
1
s
0
. Such elementary trajectories are counted with
respect to the sign +1 and the orientation of J
l
1
++l
q
0
. From Lemma 3.14, this orientation
diers from the one of J
q
0 J
l
1
J
l
q
0
exactly by the sign (1)
P
q
0
j=1
(q
0
j)(l
j
1)
. As a
consequence, the morphisms

/(H
1
H
0
) and /(H
0
) /(H
1
) coincide. The result now
follows along the same line as Theorem 3.16.
References
[1] J. M. Boardman and R. M. Vogt. Homotopy invariant algebraic structures on topo-
logical spaces. Lecture Notes in Mathematics, Vol. 347. Springer-Verlag, Berlin, 1973.
[2] A. Floer. The unregularized gradient ow of the symplectic action. Comm. Pure Appl.
Math., 41(6):775813, 1988.
[3] A. Floer and H. Hofer. Coherent orientations for periodic orbit problems in symplectic
geometry. Math. Z., 212(1):1338, 1993.
[4] S. Forcey. Convex hull realizations of the multiplihedra. Preprint arXiv:0706.3226v6,
2007.
[5] K. Fukaya. Morse homotopy, A

-category, and Floer homologies. In Proceedings of

GARC Workshop on Geometry and Topology 93 (Seoul, 1993), volume 18 of Lecture
Notes Ser., pages 1102, Seoul, 1993. Seoul Nat. Univ.
25
[6] K. Fukaya and Y.-G. Oh. Zero-loop open strings in the cotangent bundle and Morse
homotopy. Asian J. Math., 1(1):96180, 1997.
[7] H. Hofer. A General Fredholm Theory and Applications. Preprint
arXiv:math/0509366, 2005.
[8] N. Iwase and M. Mimura. Higher homotopy associativity. In Algebraic topology (Ar-
cata, CA, 1986), volume 1370 of Lecture Notes in Math., pages 193220. Springer,
Berlin, 1989.
[9] S. Mau and C. Woodward. Geometric realizations of the multiplihedron and its com-
plexication. Preprint arXiv:0802.2120, 2008.
[10] M. Schwarz. Morse homology, volume 111 of Progress in Mathematics. Birkhauser
Verlag, Basel, 1993.
[11] J. D. Stashe. Homotopy associativity of H-spaces. I, II. Trans. Amer. Math. Soc.
108 (1963), 275-292; ibid., 108:293312, 1963.
[12] J.-Y. Welschinger. Open strings, Lagrangian conductors and Floer functor. Preprint
math.arXiv:0812.0276, 2008.
Unite de mathematiques pures et appliquees de l

Ecole normale superieure de Lyon,

CNRS - Universite de Lyon.
26