Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
r
X
i
v
:
1
0
0
9
.
3
2
6
0
v
1
[
m
a
t
h
.
A
T
]
1
6
S
e
p
2
0
1
0
Groups, cacti and framed little discs
Richard Hepworth
Copenhagen University
Abstract
Let G be a topological group. Then the based loopspace ΩG of G is an
algebra over the cacti operad, while the double loopspace Ω
2
BG of the
classifying space of G is an algebra over the framed little discs operad.
This paper shows that these two algebras are equivalent, in the sense
that they are weakly equivalent Ealgebras, where E is an operad weakly
equivalent to both framed little discs and cacti. We recover the equivalence
between cacti and framed little discs, and Menichi’s isomorphism between
the BValgebras H∗(ΩG) and H∗(Ω
2
BG).
1 Introduction
The framed little discs operad fD was introduced by Getzler in [5]. The double
loopspace Ω
2
X of any based space X is naturally an algebra over fD, and any
such algebra is weakly equivalent to one of the form Ω
2
BG for some topological
group G (with nondegenerate basepoint, which we assume throughout).
The cacti operad was introduced by Voronov [13] in order to understand
the BValgebras H
∗
(LM) of Chas and Sullivan [2]. We will use a variant of
cacti introduced by Salvatore in [12], and will denote it by C. Salvatore used a
version of the Deligne conjecture to show that for any topological group G, the
loopspace ΩG is an algebra over C.
So, given a topological group G we have an fDalgebra Ω
2
BG and a C
algebra ΩG. There is a standard weak equivalence ΩG ≃ Ω
2
BG, while C and
fD are weakly equivalent operads. It is therefore natural to ask whether ΩG
and Ω
2
BG are related as algebras. The theorem below shows that this is indeed
the case.
Theorem. There is an operad E equipped with weak equivalences of operads
fD
≃
←−−−−−
π1
E
≃
−−−−−→
π2
C
such that, for any topological group G, there is an Ealgebra εG equipped with
weak equivalences of Ealgebras
Ω
2
BG
≃
←−−−−−
p1
εG
≃
−−−−−→
p2
ΩG.
Here Ω
2
BG and ΩG are regarded as Ealgebras using π
1
and π
2
respectively, and
p
1
and p
2
are compatible with the standard weak equivalence h: ΩG
≃
−→ Ω
2
BG
in the sense that h ◦ p
2
≃ p
1
.
1
The operad E is deﬁned as follows. Given elements a ∈ fD(n) and c ∈ C(n),
write a for the space obtained from D
2
by deleting the interiors of the little
discs of a, and write c for the conﬁguration of circles underlying c. Then E(n) is
the space of triples (a, c, f), where a ∈ fD(n), c ∈ C(n), and where f : a → c
is a homotopy equivalence satisfying the following two boundary conditions.
First, the boundary of the ith little disc is sent into c by the inclusion of the
ith lobe. Second, the boundary of the big disc is sent into c by the pinch map.
A typical element (a, c, f) ∈ E(2) is depicted below.
a c f
1 2
1 2
There are forgetful maps
fD(n)
π1
←−−−−− E(n)
π2
−−−−−→ C(n).
The ﬁrst part of our theorem can now be restated as follows.
Theorem A. The spaces E = {E(n)} form a topological operad, and π
1
and π
2
are weak equivalences of operads.
In particular, we recover the fact that cacti and framed little discs are weakly
equivalent. This was stated by Voronov [13] and ﬁrst proved by Kaufmann [8].
A diﬀerent proof has since been given by Bargheer [1]. In those accounts, as
in Theorem A, one constructs some intermediate operad O equipped with weak
equivalences to both fD and C. In Kaufmann’s proof O is obtained using a
recognition principle of Fiedorowicz [4], while in Bargheer’s proof one takes
O = WfD, the BoardmanVogt Wconstruction on fD. The advantage of
Theorem A is that E is constructed using only the geometry of the original
operads fD and C. Further, it allows us to construct a natural Ealgebra εG
weakly equivalent to both ΩG and Ω
2
BG. Note, however, that Theorem A still
relies on Kaufmann’s computation of the weak homotopy type of the C(n).
Now we describe the Ealgebra εG. Let EG → BG be the universal principal
Gbundle. Then εG is the space of based maps D
2
→ EG that send S
1
into the
orbit of the basepoint. There are maps
Ω
2
BG
p1
←−−−−− εG
p2
−−−−−→ ΩG
given respectively by projecting from EG to BG and by restricting from D
2
to
S
1
. These are compatible with h: ΩG
≃
−→ Ω
2
BG in the sense that h ◦ p
2
≃ p
1
.
We can now restate the second part of our theorem as follows.
Theorem B. The space εG is an Ealgebra, and the maps p
1
and p
2
are weak
equivalences of Ealgebras.
We recover a result of Menichi [11] which states that h
∗
: H
∗
(ΩG) → H
∗
(Ω
2
BG)
is an isomorphism of BValgebras. (Menichi’s result holds in the more general
case that G is a grouplike topological monoid.)
2
The key ingredient in our paper is the following common feature of the operads
fD and C. To elements a ∈ fD(n) and c ∈ C(n) we can associate the spaces a
and c deﬁned above, which we call the realizations of a and c. The realizations
come together with incoming boundary maps
∂
1
, . . . , ∂
n
: S
1
→ a, ∂
1
, . . . , ∂
n
: S
1
→ c
and outgoing boundary maps
∂
out
: S
1
→ a, ∂
out
: S
1
→ c.
Together, these realizations and boundary maps form what we call realization
systems for fD and C. This means that they are compatible with the symmetry
maps, composition maps, and topology of the original operads fD and C. Most
importantly, the realization x ◦
i
y of the composite of x and y is the pushout
of ∂
i
: S
1
→ x and ∂
out
: S
1
→ y. This fact is easily veriﬁed for both fD and
C, and leads directly to the deﬁnition of composition in E.
The paper contains a detailed development of realization systems. The cru
cial result takes two operads P and Q with realization systems and constructs a
mapping operad M. Points of M(n) are triples (p, q, f) where p ∈ P(n), q ∈ Q(n)
and f : p → q is a map satisfying f ◦ ∂
i
= ∂
i
and f ◦ ∂
out
= ∂
out
. The operad
E is obtained in the case P = fD, Q = C by restricting to the suboperad of M
consisting of triples (p, q, f) in which f is a homotopy equivalence. The three
algebras studied in the paper, namely the fDalgebra Ω
2
BG, the Calgebra ΩG,
and the Ealgebra εG, are also described in terms of the realization systems for
fD and C.
The paper begins in Section 2 with a discussion of realization systems, and
proves that from a pair of operads with realization systems, one obtains a map
ping operad M. Sections 3 and 4 recall framed little discs and cacti, and describe
their realization systems. Then Section 5 applies the mapping operad construc
tion to P = fD and Q = C to obtain E.
Section 6 uses the realization systems to describe the actions of fDon Ω
2
BG,
of C on ΩG, and of E on εG. With these ingredients in place Theorem B follows
without diﬃculty.
It remains to complete the proof of Theorem A by showing that π
1
and
π
2
are weak equivalences. The structure of the proof is explained in detail in
Section 7. It is shown there that Theorem A follows from the computation of
a certain long exact sequence, which is carried out in Section 8, and from a
further result, Theorem C.
Theorem C states roughly that the projection π
1
×π
2
: E(n) → fD(n)×C(n)
is a quasiﬁbration. It relies on some basic results on mapping spaces which are
given in Section 9, and on a study of the ﬁbrewise structure of the realizations of
fD and C, which is carried out in Sections 10 and 11. The proof of Theorem C
is completed in Section 12.
An appendix recalls some basic results of ﬁbrewise topology.
2 Realization systems for operads
In this section we introduce realization systems for an operad O. A realization
system with boundaries X is a rule that assigns to each element x ∈ O(n) a space
3
x, the realization of x, equipped with (n + 1) diﬀerent maps X → x. The
rule carries further structure and is subject to a number of axioms, all of which
roughly amount to saying that the spaces x are compatible with symmetry and
composition in O, and vary continuously as x varies within O(n). This is made
precise in §2.1, and two simple examples are given in §2.2.
Then in §2.3 we deﬁne the mapping operad. Given operads P and Q, both
equipped with realizations, there is an associated mapping operad. It lies over
both P and Q, and its elements are triples consisting of points p ∈ P(n), q ∈ Q(n),
and a map p → q satisfying boundary conditions. The operad E of Theorem A
will be constructed as a suboperad of such a mapping operad.
For us, all operads are Σoperads with a 0th space, which we do not insist
be equal to a single point. Composition within operads will be denoted using
either the symbol γ or the symbol ◦
i
, for example γ(x; y
1
, . . . , y
n
) or x ◦
i
y.
2.1 Realization systems
Fix a compact, Hausdorﬀ topological space X and a topological operad O. The
following three deﬁnitions together deﬁne the notion of a realization system.
The deﬁnitions are somewhat abstract, and so the reader may ﬁnd it helpful to
bear in mind the example of framed little discs. In this case, the realization a
of a ∈ fD(n) is the complement of the interiors of the little discs; the incoming
boundary maps S
1
→ a are the boundaries of the little discs; and the outgoing
boundary map S
1
→ a is the boundary of the big disc.
The ﬁrst deﬁnition states what data must be given in order to deﬁne a
realization system. It is analogous to deﬁning an operad by specifying spaces,
symmetry maps, and composition maps.
Deﬁnition 2.1. A realization system for O with boundaries X is a rule RO that
for each n 0 and each x ∈ O(n) produces a topological space x called the
realization of x, and that furthermore equips these spaces with the following
structure:
• Boundary Maps. For each x ∈ O(n), incoming boundary maps
∂
1
, . . . , ∂
n
: X → x
and an outgoing boundary map ∂
out
: X → x.
• Symmetries. For each x ∈ O(n) and σ ∈ Σ
n
, a symmetry map σ
∗
: x →
xσ.
• Pasting and composition. Given x ∈ O(n) and y
i
∈ O(m
i
) for i =
1, . . . , n, a pushout diagram
¸
n
i=1
X
∂out
∂i
y
1
 ⊔ · · · ⊔ y
n

x
γ(x; y
1
, . . . , y
n
)
(1)
called the pasting square.
4
The system RO is required to satisfy the axioms of Deﬁnition 2.2, and to be
topologized in the sense of Deﬁnition 2.3, both below.
Our second deﬁnition lists the axioms that the structures above must satisfy.
It is analogous to the axioms satisﬁed by the various composition and symmetry
maps in an operad. Although we list no fewer than six axioms, they are almost
trivial to verify in the example of framed little discs.
Deﬁnition 2.2. A realization system RO must satisfy the axioms listed be
low. In what follows we will use underlines to indicate tuples of elements. For
example, y will denote y
1
, . . . , y
n
.
1. Unit: Taking x = 1
O
, the boundary maps ∂
1
, ∂
out
: X → 1
O
 are home
omorphisms and coincide. Applying these homeomorphisms to (1) in the
case n = 1 and x = 1
O
, and to (1) in the case y
i
= 1
O
, one obtains the
diagrams
X
∂out
y
X
∂out
y
and
¸
X
∂i
¸
X
∂i
x x
respectively.
2. Symmetries: The symmetry maps satisfy σ
∗
◦ τ
∗
= (τσ)
∗
and are com
patible with the boundary maps in the sense that
σ
∗
◦ ∂
i
= ∂
σ
−1
i
, σ
∗
◦ ∂
out
= ∂
out
.
3. Pasting and boundaries: The lower and righthand maps of diagram
(1) are compatible with the boundary maps, in the sense that the following
triangles commute.
x
γ(x; y)
X
∂out
∂out
x
x
x
x
x
x
x
x
x
x
x
x
¸
y
i

¸¸
X
∂j
∂i,j
.v
v
v
v
v
v
v
v
v
v
v
v
v
γ(x; y)
The unions run over i = 1, . . . , n and j = 1, . . . , m
i
; the symbol ∂
i,j
denotes ∂
m1+···+mi−1+j
.
4. Pasting and symmetries I: Given σ
1
∈ Σ
m1
, . . . , σ
n
∈ Σ
mn
, we can
formulate diagram (1) for x and y
1
, . . . , y
n
, and for x and y
1
σ
1
, . . . , y
n
σ
n
.
Then the resulting squares are isomorphic under the relevant symmetry
5
map, which means that the following diagram commutes.
¸
X
.t
t
t
t
t
t
K
K
K
K
K
K
¸
y
i

σ
∗
1
⊔···⊔σ
∗
n
¸
y
i
σ
i

x
.t
t
t
t
t
t
t
K
K
K
K
K
K
K
γ(x; y)
(σ1⊕···⊕σn)
∗
γ(x; yσ)
Here yσ = y
1
σ
1
, . . . , y
n
σ
n
.
5. Pasting and symmetries II: Similarly, given σ ∈ Σ
n
we can formulate
diagram (1) for x, y
1
, . . . , y
n
, and for xσ, y
σ
−1
1
, . . . , y
σ
−1
n
. Then the two
squares are isomorphic under the relevant symmetry maps, which means
that the following diagram commutes.
¸
X
σ⊔
E
E
E
E
E
E
¸
y
σ
−1
i

σ⊔
O
O
O
O
O
O
O
O
¸
X
¸
y
i

x
σ
∗
E
E
E
E
E
E
γ(x; y
σ
−1 )
σ(m1,...,mn)
∗
O
O
O
O
O
O
O
xσ
γ(xσ; y)
Here y
σ
−1 = y
σ
−1
1
, . . . , y
σ
−1
n
and σ
⊔
sends the σ
i
th cofactor to the ith
cofactor.
6. Associativity of pasting: Suppose we are given the data for an iterated
composition in O. Thus we have elements x ∈ O(n), y
i
∈ O(m
i
) and z
j
i
∈
O(l
j
i
), for i = 1, . . . , n and j = 1, . . . , m
i
. Then we can form composites
γ(x; y), γ(y
i
; z
i
), γ(γ(x; y); z), γ(x; γ(y
1
; z
1
), . . . , γ(y
n
; z
n
)).
The last two composites coincide, and consequently so do their realiza
tions,
γ(γ(x; y); z) = γ(x; γ(y
1
; z
1
), . . . , γ(y
n
; z
n
)).
By considering the pasting diagram (1) for each composite, we obtain two
maps from each of x, y
1
, . . . , y
n
, z
1
1
, . . . , z
mn
n
 into the single space
above. The two maps in each such pair coincide.
The next deﬁnition says what it means to topologize a realization system,
and makes use of ﬁbrewise topology. The standard reference for this is [6], and
we have recalled the necessary ideas in Appendix A. A space ﬁbred over B is
just a space X equipped with a map X → B, and the appendix recalls what it
means for such a ﬁbred space to be ﬁbrewise compact or ﬁbrewise Hausdorﬀ.
6
Deﬁnition 2.3 (Topology on a realization system). A topology on a realization
system RO consists of ﬁbred spaces
ρ
n
: RO(n) → O(n)
for n 0, such that for each x ∈ O(n) the ﬁbre ρ
−1
n
(x) is identiﬁed with x.
Each RO(n) must be ﬁbrewise compact Hausdorﬀ over O(n). The boundary
and symmetry maps determine functions
∂
1
, . . . , ∂
n
, ∂
out
: O(n) ×X → RO(n), σ
∗
: RO(n) → σ
∗
RO(n)
between spaces over O(n), and the pasting squares (1) become squares
¸
n
i=1
X
∂out
∂i
RO(m
1
) ⊔ · · · ⊔ RO(m
n
)
RO(n)
RO(m
1
+· · · +m
n
)
(2)
of spaces over O(n) × O(m
1
) × · · · × O(m
n
). All of the functions above are
required to be continuous, and the squares (2) are required to be ﬁbrewise
pushouts. (The spaces written in (2) should all be regarded as spaces over
O(n) × O(m
1
) × · · · × O(m
n
) by pulling back along the appropriate constant,
projection or composition map.)
2.2 Examples of realizations
This subsection attempts to illustrate the deﬁnition of realization systems by
giving two simple examples in the case when O is the operad of little ncubes.
The paper’s two main examples, which are the realizations for fD and C, are
presented in Sections 3 and 4 respectively.
Example 2.4 (The little cubes operad). Let C
n
denote the little ncubes op
erad. Deﬁne a realization system R
1
C
n
with boundaries S
n
as follows. Given
x ∈ C
n
(p), set x =
p
i=1
S
n
.
• The incoming boundary maps ∂
1
, . . . , ∂
p
: S
n
→
S
n
are the standard
insertions. The outgoing boundary map ∂
out
: S
n
→
S
n
is the collapse
map coll
x
induced by x.
• The symmetry map σ
∗
: x → xσ is the map
S
n
→
S
n
that sends
the ith cofactor to the (σ
−1
i
)th cofactor.
• Given x ∈ C
n
(p) and y
i
∈ C
n
(q
i
), the pasting square (1) is given by
¸
p
i=1
S
n
colly
i
¸
p
i=1
qi
j=1
S
n
p
i=1
S
n
colly
i
p
i=1
qi
j=1
S
n
where the vertical maps are the standard quotients.
7
The axioms of Deﬁnition 2.2 hold trivially. The system is topologized by setting
R
1
C
n
(p) = C
n
(p) ×
S
n
, and the conditions of Deﬁnition 2.3 are immediately
veriﬁed.
Example 2.5 (Little cubes operads again). Let C
n
again denote the little n
cubes operad. Deﬁne the realization system R
2
C
n
with boundaries S
n−1
, which
we regard as the boundary of the ncube I
n
, as follows. Given x ∈ C
n
(p), deﬁne
x to be the complement in the big cube I
n
of the interiors of the little cubes
of x.
• The incoming boundary maps ∂
1
, . . . , ∂
p
: S
n−1
→ x are the inclusions
of the boundaries of the little cubes of x. The outgoing boundary map
∂
out
: S
n−1
→ x is the inclusion of the boundary of the big cube.
• The symmetry map σ
∗
: x → xσ identiﬁes the two spaces as subsets of
the big cube.
• Given x ∈ C
n
(p) and y
i
∈ C
n
(q
i
), the pasting square (1)
¸
n
i=1
S
n−1
∂out
∂i
y
1
 ⊔ · · · ⊔ y
n

x
γ(x; y
1
, . . . , y
n
)
is given as follows. The bottom map is the inclusion of x into γ(x; y
1
, . . . , y
n
)
given by regarding both as subsets of the big cube I
n
. The righthand map
sends y
i
 into γ(x; y
1
, . . . , y
n
) by applying x
i
: I
n
→ I
n
, the ith little
cube of x. This is a pushout diagram.
The axioms of Deﬁnition 2.2 are simple to verify. The system is topologized by
deﬁning R
2
C
n
(p) to be the subspace of C
n
(p) × I
n
whose ﬁbre over x ∈ C
n
(p)
is precisely x ⊂ I
n
. The conditions of Deﬁnition 2.3 follow. (Compare with
the system RfD in Section 3.2.)
2.3 The mapping operad
Suppose now that we are given operads P and Q equipped with realization
systems RP and RQ, both with boundaries X. This subsection introduces the
mapping operad M associated to this data.
Deﬁnition 2.6. Fix n 0. Then M(n) denotes the set of triples (a, c, f), where
a ∈ P(n), c ∈ Q(n) and f : a → c is a continuous map satisfying the boundary
conditions f ◦ ∂
i
= ∂
i
and f ◦ ∂
out
= ∂
out
.
Deﬁnition 2.7. The action of Σ
n
on M(n) is deﬁned by the formula (a, c, f)σ =
(aσ, cσ, σ
∗
◦f ◦σ
∗−1
), where σ
∗
: a → aσ and σ
∗
: c → cσ are the symmetry
maps.
Deﬁnition 2.8. Composition in M is deﬁned by pasting functions in the fol
lowing way. Given (a, c, f) ∈ M(n) and (b
i
, d
i
, g
i
) ∈ M(m
i
) for i = 1, . . . , n, the
composite
γ ((a, c, f); (b
1
, d
1
, g
1
), . . . , (b
n
, d
n
, g
n
))
8
is the new triple
(γ(a; b
1
, . . . , b
n
), γ(c; d
1
, . . . , d
n
), γ(f; g
1
, . . . , g
n
)) ,
where γ(f; g
1
, . . . , g
n
) is the unique map making the cube
¸
X
F
F
F
F
F
F
F
F
F
F
¸
b
i

g1⊔···⊔gn
M
M
M
M
M
M
M
¸
X
¸
d
i

a
f
F
F
F
F
F
F
F
γ(a; b)
γ(f;g1,...,gn)
M
M
M
M
c
γ(c; d)
commute. The front and rear faces of this cube are the pushout squares (1)
for c, d
1
, . . . , d
n
and a, b
1
, . . . , b
n
respectively, while the left hand and upper
faces come from the compatibility between pasting and boundary maps. Thus
a unique such γ(f; g
1
, . . . , g
n
) exists.
To topologize M(n) we use another notion from ﬁbrewise topology. Given
spaces U and V ﬁbred over B, we write Map
B
(U, V ) for the set of pairs (b, f),
where b ∈ B is a point of the base and f : U
b
→ V
b
is a map between the
ﬁbres over that point. This set can be equipped with the ﬁbrewise compact
open topology, and we call the resulting space the ﬁbrewise mapping space. See
Appendix A and the reference there.
The realization systems RP and RQ give us ﬁbred spaces RP(n) → P(n) and
RQ(n) → Q(n). Write π
1
and π
2
for the projections of P(n) × Q(n) onto its
factors. Then the ﬁbrewise mapping space Map
P(n)×Q(n)
(π
∗
1
RP(n), π
∗
2
RQ(n))
consists of triples (a, c, f) where a ∈ P(n), c ∈ Q(n) and f : a → c. In
particular, it contains M(n) as a subset.
Deﬁnition 2.9. The space M(n) is topologized as a subspace of the ﬁbrewise
mapping space Map
P(n)×Q(n)
(π
∗
1
RP(n), π
∗
2
RQ(n)).
Theorem 2.10. The collection M = {M(n)} is topological operad, and the
projections
P ←− M −→ Q
are morphisms of operads.
Proposition 2.11. Suppose that for both RP and RQ the pasting squares (1)
are all homotopy pushouts, rather than just pushouts. Then the spaces
M
≃
(n) = {(a, c, f) ∈ M(n)  f is a homotopy equivalence}
form a suboperad M
≃
of M.
Proof of Proposition 2.11. That M
≃
(1) contains the unit element, and that M
≃
is closed under the Σ
n
action, follow from the unit and symmetry axioms re
spectively. Since homotopy equivalences are preserved by homotopy pushouts,
it follows that if f and g
1
, . . . , g
n
are all homotopy equivalences, then so is
γ(f; g
1
, . . . , g
n
). Thus M
≃
is closed under composition.
9
Proof of Theorem 2.10. We begin by showing that M is an operad in sets. This
is a direct consequence of the axioms for a realization system presented in Deﬁni
tion 2.2. We will only show that composition is associative, the proof of the rest
being similar but simpler. Take elements (x, u, f) ∈ M(n), (y
i
, v
i
, g
i
) ∈ M(m
i
)
and (z
j
i
, w
j
i
, h
j
i
) ∈ M(l
j
i
) for i = 1, . . . , n and j = 1, . . . , m
i
. Write
A = γ(γ(x, y); z) = γ(x; γ(y
1
; z
1
), . . . , γ(y
n
; z
n
))
and
B = γ(γ(u, v); w) = γ(u; γ(v
1
; w
1
), . . . , γ(v
n
; w
n
).
We must show that the two pasted maps
γ(γ(f, g); h) and γ(f; γ(g
1
; h
1
), . . . , γ(g
n
; h
n
))
coincide. But both maps are characterized as the unique map making each
square
x
f
u
A
_ _ _
B,
y
i

gi
v
i

A
_ _ _
B,
z
j
i

h
j
i
w
j
i

A
_ _ _ _
B
commute. Here the vertical maps come from Axiom 6 of Deﬁnition 2.2. In
particular, the two pasted maps coincide.
Now we will prove that the composition map M(n)×M(m
1
)×· · ·×M(m
n
) →
M(m
1
+· · ·+m
n
) is continuous. The proof that the permutation map σ: M(n) →
M(n) is continuous is similar.
We will use the following shorthand for spaces obtained from RP, and the
equivalent shorthand for spaces obtained from RQ.
1. P
Π
denotes the product P(n) ×P(m
1
) ×· · · ×P(m
n
).
2. RP
⊔
denotes the space obtained by pulling back each of RP(n) and RP(m
i
)
to P
Π
and then forming the disjoint union.
3. RP
◦
denotes the pullback of RP(m
1
+· · · +m
n
) to P
Π
.
4. ξ
P
: RP
⊔
→ RP
◦
denotes the map obtained from the pasting squares (2).
Note that ξ
P
and ξ
Q
are proper ﬁbrewise surjections, since X is compact and
the square (2) is a ﬁbrewise pushout. (The analogous claim need not hold for
ξ
Q
.)
An element x of M(n) × M(m
1
) × · · · × M(m
n
) can be regarded as a point
x
1
of the product
Map
P(n)×Q(n)
(RP(n), RQ(n)) ×
¸
Map
P(mi)×Q(mi)
(RP(m
i
), RQ(m
i
)),
which maps continuously into Map
P
Π
×Q
Π(RP
⊔
, RQ
⊔
), sending x
1
to a point x
2
.
Now ξ
P
and ξ
Q
give continuous embeddings
Map
P
Π
×Q
Π(RP
⊔
, RQ
⊔
)
ξ
Q
∗
−−→ Map
P
Π
×Q
Π(RP
⊔
, RQ
◦
)
ξ
∗
P
←− Map
P
Π
×Q
Π(RP
◦
, RQ
◦
).
10
The image of x
2
in the central space lifts to an element x
3
of the right hand
space. Finally we obtain a continuous map
Map
P
Π
×Q
Π(RP
◦
, RQ
◦
) → Map
P(
mi)×Q(
mi)
(RP(
¸
m
i
), RQ(
¸
m
i
))
sending x
3
to a point y. This y lies in M(m
1
+· · · +m
n
).
The assignment x → y just described is nothing more than the composition
map, and the description makes its continuity evident. This completes the
proof.
3 Framed little discs and their realizations
This section recalls Getzler’s framed little discs operad fD and introduces its
realization system RfD.
3.1 Framed little discs
Deﬁnition 3.1 (The framed little discs operad [5]). Getzler’s framed little
discs operad fD has for its nth space the collection fD(n) of embeddings
a:
¸
D
2
→ D
2
of n little discs into a single big disc. It is required that the
ith component a
i
: D
2
→ D
2
of the embedding be given by a combination of
translation, rotation and rescaling.
The unit 1
fD
is the identity embedding D
2
→ D
2
. The action of Σ
n
on
fD(n) is given by permuting the n little discs: the ith disc of aσ is the σ(i)th
disc of a. Composition in fD is given by composition of embeddings.
Example 3.2. A typical element a ∈ fD(3) is drawn below.
1
2
3
The dash on the ith little disc indicates the image of the basepoint of S
1
,
which we call the local marked point. The bullet • indicates the basepoint on
the boundary of the big disc, which we call the global marked point.
3.2 Realizations of framed little discs
Deﬁnition 3.3 (Realizations of framed little discs). The realization system
RfD is deﬁned as follows. Given a ∈ fD(n), the realization of a is the space
a ⊂ D
2
obtained by deleting the interiors of the little discs from the big disc.
• The incoming boundary maps ∂
1
, . . . , ∂
n
: S
1
→ a are given by ∂
i
=
a
i
S
1
. The outgoing boundary map ∂
out
: S
1
→ a is given by the inclusion
S
1
֒→ D
2
. In other words the ith incoming boundary map is given by
the boundary of the ith little disc, while the outgoing boundary map is
given by the boundary of the big disc.
• Given a ∈ fD(n) and σ ∈ Σ
n
, the symmetry map σ
∗
: a → aσ identiﬁes
a and aσ as subsets of D
2
.
11
• Given a ∈ fD(n) and b
i
∈ fD(m
i
) for i = 1, . . . , n, the pasting square
¸
n
i=1
S
1
∂out
∂i
b
1
 ⊔ · · · ⊔ b
n

a
γ(a; b
1
, . . . , b
n
)
(3)
is given as follows. The lower map is the inclusion of a into γ(a; b
1
, . . . , b
n
)
given by regarding both as subsets of D
2
. The right hand map sends b
i

into γ(a; b
1
, . . . , b
n
) by applying a
i
: D
2
→ D
2
. In other words it sends
b
i
 into the ith little disc of a. This is a pushout square, and in fact a
homotopy pushout.
Example 3.4. The realization of the element a ∈ fD(3) from Example 3.2,
together with its boundary maps, is shown below.
¸
∂
i
∂
out
Now take the following elements a, b
1
, b
2
.
a b
1
b
2
1
2
1 2
1
2
Then the pasting diagram (3) for γ(a; b
1
, b
2
) is depicted below:
Deﬁnition 3.5 (Topology on the realizations of framed little discs). For n 0
we deﬁne RfD(n) to be the subspace of fD(n)×D
2
whose ﬁbre over a ∈ fD(n)
is precisely a ⊂ D
2
. It is equipped with the projection map ρ
n
: RfD(n) →
fD(n). As a closed subspace of fD(n) ×D
2
, it is a ﬁbrewise compact Hausdorﬀ
space over fD(n).
Proposition 3.6. The axioms of Deﬁnition 2.2 and the conditions of Deﬁni
tion 2.3 are satisﬁed by the realizations of framed little discs.
Proof. The axioms of Deﬁnition 2.2 follow from a lengthy but trivial veriﬁcation.
The continuity required by Deﬁnition 2.3 follows immediately.
12
4 Cacti and their realizations
This section introduces the cacti operad and its realization system. Cacti were
introduced by Voronov in [13]. Several variants have been introduced since then,
in particular by McClureSmith [10], Kaufmann [8], and Salvatore [12]. We
begin in §4.1 with an informal introduction to cacti along the lines of Voronov’s
original deﬁnition. Then in §4.2 we recall in detail Salvatore’s variant of cacti,
which will be used in the rest of the paper. Finally in §4.3 we introduce the
realization system for cacti.
4.1 An introduction to cacti
This subsection gives an informal description of the cactus operad, along the
lines of Voronov’s original deﬁnition in [13]. We will not recall Voronov cacti in
detail, but hope to give the unfamiliar reader an idea of the theory.
Deﬁnition 4.1. A cactus with n lobes is a planar treelike conﬁguration of
parameterized circles, of varying radii, labelled by 1, . . . , n, equipped with a
global marked point on one of the circles. The individual circles are called the
lobes of the cactus, and the basepoints on the lobes are called the local marked
points. See [13] and [3].
Thus a cactus consists of n labelled circles of positive radius identiﬁed at
ﬁnitely many intersection points. Planar means that the circles meeting at
each intersection point are equipped with a cyclic order. Treelike means that
the space obtained by ﬁlling in each circle with a disc is contractible.
The cactus operad has for its nth term the collection of all cacti with n
lobes.
Example 4.2. Here is a typical cactus with 4 lobes.
1
2 3
4
The bullet • indicates the global marked point, while the dashes indicate the
local marked points.
Given cacti c and d, the composed cactus c ◦
i
d is deﬁned by the following
pasting process. Rescale d so that its total length becomes the length of the
ith lobe of c. Then c ◦
i
d is obtained from c and d by collapsing the ith lobe
of c onto d using the pinch map S
1
→ d that sends the basepoint to the global
marked point and then proceeds around the ‘outside’ of the cactus.
4.2 Salvatore’s cacti operad
In this subsection we recall the variant of the cacti operad developed by Salvatore
in [12]. The operad was denoted there by fMS, but here we will call it C. It
is a suboperad of the coendomorphism operad CoEnd(S
1
) of S
1
. We begin by
establishing some notation.
13
Deﬁnition 4.3. 1. For n 1 we deﬁne a space F(n) as follows. A point
x ∈ F(n) is a partition of S
1
into n closed 1manifolds I
j
(x) with equal
length and piecewise disjoint interiors, such that there is no cyclically
ordered sequence z
1
, z
2
, z
3
, z
4
with z
1
, z
3
∈ int I
j
(X) and z
2
, z
4
∈ int I
k
(x)
with j = k. Set F(0) = ∗.
2. Given x ∈ F(n) we deﬁne a map c(x): S
1
→ (S
1
)
n
as follows. Given
j ∈ {1, . . . , n}, collapse each component of S
1
\ I
j
(x) to a point and rescale
the quotient in order to identify it with S
1
. This gives a based map
π
j
: S
1
→ S
1
. Then c(x) is the product π
1
×· · · ×π
n
.
3. Write Mon(I, ∂I) for the space of nondecreasing maps I → I that preserve
∂I. Regard elements f ∈ Mon(I, ∂I) as based maps f : S
1
→ S
1
.
4. Given z ∈ S
1
write L
z
: S
1
→ S
1
for translation by z.
The constructions above induce an embedding
c: F(n) ×Mon(I, ∂I) ×(S
1
)
n
→ CoEnd(S
1
)(n)
that sends (x, f, z) to the composite (L
z1
×· · · ×L
zn
) ◦ c(x) ×f.
Deﬁnition 4.4 (Salvatore’s cacti operad). C is deﬁned to be the suboperad of
CoEnd(S
1
) whose nth term is the image of the embedding c: F(n)×Mon(I, ∂I)×
(S
1
)
n
→ CoEnd(S
1
)(n).
Note 4.5. Here is the connection between the operad C and the informal notion
of cacti given in the last subsection. Fix c ∈ C(n) and write c = c(x, f, z).
Consider the image c(S
1
) ⊂ (S
1
)
n
. It depends only on x and z, and can be
regarded as a Voronov cactus in which the lobes have equal length. For c(S
1
) is
a union of n parameterized circles. The jth of these circles is simply c(I
j
(x)).
This union of circles inherits a planar structure from the cyclic ordering of the
components of the I
j
(x), and the condition on these cyclic orderings ensures
that this union of circles is treelike. The global marked point is given by c(∗).
Thus c = c(x, f, z) determines a cactus in which the lobes have equal length,
depending only on x and z. We regard f as a reparametrization of the pinch
map of this cactus; in the case when f has constant speed l
j
on I
j
(x), we can
regard c as a cactus in which the jth lobe has length l
j
.
Note 4.6. Salvatore describes the precise relationship between C and Voronov’s
original deﬁnition of cacti. In particular, Theorem 5.3.6 of [8] holds with C in
place of Voronov cacti.
Note 4.7. Finally we recall the natural cell decomposition {D
X
} of F(n).
Fix x ∈ F(n). Pull back the partition {I
j
(x)} of S
1
to a partition of [0, 1].
This partition is speciﬁed by boundary points 0 = x
0
< x
1
< · · · < x
k
= 1 and
a labelling X(x) = (X
1
(x), . . . X
k
(x)) of the intervals [x
0
, x
1
], . . . , [x
k−1
, x
k
] by
elements from {1, . . . , n}.
The sequence X = X(x) of the last paragraph satisﬁes the following three
properties. First, all values 1, . . . , n appear. Second, adjacent terms are distinct.
Third, there is no subsequence of the form i, j, i, j with i = j. Every such
sequence X arises as X(x) for some element x ∈ F(n).
Let X be a sequence satisfying the three conditions above. Write D
X
⊂ F(n)
for the subset consisting of those x for which X(x) is contained in X. Then D
X
14
is homeomorphic to the product
¸
n
j=1
∆
d(j)
, where d(j) + 1 is the number of
elements of X labelled by j. The homeomorphism sends x ∈ D
X
to the point
whose image in ∆
d(j)
records the lengths of the intervals labelled by j.
4.3 The realizations of cacti
Deﬁnition 4.8 (Realizations of cacti). The realization system RC is deﬁned
as follows. Let c ∈ C(n) be a cactus with n lobes. The realization of c is the
topological space c(S
1
), which we denote by c.
• The ith incoming boundary map ∂
i
: S
1
→ c is deﬁned as follows. Write
c = c(x, f, z). Then c(I
i
(x)) ⊂ c has the form x
i
1
× · · · × S
1
× · · · ×
x
i
n
for some lobe coordinate (x
i
1
, . . . , x
i
i−1
, x
i
i+1
, . . . , x
i
n
), and ∂
i
is y →
(x
i
1
, . . . , x
i
i−1
, y, x
i
i+1
, . . . , x
i
n
). The outgoing boundary map ∂
out
: S
1
→ c
is given by c itself.
• Let c ∈ C(n) and let σ ∈ Σ
n
. Then the homeomorphism σ
∗
: c → cσ is
deﬁned by σ
∗
(t
1
, . . . , t
n
) = (t
σ1
, . . . , t
σn
).
• Let c ∈ C(n) and d
i
∈ C(m
i
) for i = 1, . . . , n. Then there is a pushout
square
¸
n
i=1
S
1
∂out
∂i
d
1
 ⊔ · · · ⊔ d
n

c
γ(c; d
1
, . . . , d
n
).
(4)
The lower map sends c = c(S
1
) into γ(c; d
1
, . . . , d
n
) = (d
1
× · · · ×
d
n
)c(S
1
) via d
1
×· · · ×d
n
. The right hand map sends y ∈ d
i
 to
(d
1
(x
i
1
), . . . , d
i−1
(x
i
i−1
), y, d
i+1
(x
i
i+1
), . . . , d
n
(x
i
n
)).
Example 4.9. The next ﬁgures illustrate the incoming
¸
∂
i
and outgoing
∂
out
boundary maps for the cactus of Example 4.2.
Deﬁnition 4.10 (Topology on the realizations of cacti). We deﬁne RC(n) to
be the space of pairs
RC(n) = {(c, z) ∈ C(n) ×(S
1
)
n
 z ∈ c(S
1
)}.
The ﬁbre of the projection map ρ
n
: RC(n) → C(n) over c ∈ C(n) is naturally
identiﬁed with c. As a closed subspace of C(n) ×(S
1
)
n
, it is ﬁbrewise compact
and Hausdorﬀ over C(n).
15
It remains to show that the last two deﬁnitions do indeed endow C with a
realization system RC. This is given by the next proposition.
Proposition 4.11. 1. Diagram (4) is a pushout, and in fact a homotopy
pushout.
2. The realizations of cacti satisfy the conditions of Deﬁnition 2.2.
3. The topology on realizations of cacti satisfy the conditions of Deﬁnition 2.3.
Proof. For the ﬁrst part, a simple inductive argument shows that it is suﬃcient
to prove the claim when all but one of the d
i
is the unit element. To prove this
case it suﬃces to show that, given c ∈ C(n) and d ∈ C(m), the diagram
S
1
∂out
∂i
d
β
c
α
c ◦
i
d
is a pushout. This follows from the deﬁnitions and from the fact that ∂
i
is a
coﬁbration.
We now prove the second part. Axioms 1, 2, 3, 4 and 5 follow immedi
ately from the deﬁnitions. Let us turn to axiom 6. Write Γ = γ(γ(x, y); z) =
γ(x; γ(y
i
; z
i
)). The compatibility of pasting and boundary maps means that the
pairs of maps
u
1
, u
2
: x → Γ, v
1
, v
2
: y
i
 → Γ, w
1
, w
2
: z
j
i
 → Γ
satisfy u
p
◦ ∂
out
= ∂
out
, v
p
◦ ∂
out
= ∂
i
, and w
p
◦ ∂
k
= ∂
L+k
, where L =
l
1
1
+· · · + l
m1
1
+ · · · +l
1
i
+ · · · +l
j−1
i
. Since ∂
out
: S
1
→ x, ∂
out
: S
1
→ y
i
 and
¸
∂
k
:
¸
S
1
→ z
j
i
 are all surjections, it follows that the two members of each
pair of maps coincide.
Now we prove the ﬁnal part. Continuity of ∂
i
: C(n) × S
1
→ RC(n) follows
from the fact that the lobe coordinates of a cactus depend continuously on
the cactus. This can be proved using the cell decomposition of F(n) given in
Note 4.7. Continuity of ∂
out
: C(n) × S
1
→ RC(n) and σ
∗
: RC(n) → RC(n) is
immediate.
Note that
¸
∂
i
: C(n) ×
¸
S
1
→ RC(n) are continuous surjections between
ﬁbrewise compact Hausdorﬀ spaces. By Proposition A.5, to prove continuity of
the righthand map of (2) it will suﬃce to show that the composite with
¸
∂
i
is
continuous. But this follows from the compatibility of pasting with boundaries.
Similar reasoning shows that the lower map of (2) is continuous.
5 The operad E
In Sections 3 and 4 we saw that the framed little discs operad fD and the cacti
operad C admit realization systems RfD and RC with boundaries S
1
. These
systems both consist of ﬁbrewise compact Hausdorﬀ spaces. We may therefore
form the mapping operad M of Section 2 in the case P = fD, Q = C. This
is a topological operad, and since the pasting squares (1) for RfD and RC
are homotopy pushouts, it has a suboperad M
≃
consisting of triples (a, c, f) in
which f is a homotopy equivalence.
16
Deﬁnition 5.1. Set E = M
≃
. Thus E(n) is the space of triples (a, c, f), where
a ∈ fD(n), c ∈ C(n), and f : a → c is a homotopy equivalence satisfying
f ◦ ∂
i
= ∂
i
and f ◦ ∂
out
= ∂
out
.
An element (a, c, f) ∈ E(2) is depicted in the introduction. Theorem 2.10
immediately gives us the following result. With this in hand, to prove Theo
rem A it remains to show that the projection maps π
1
and π
2
are weak homotopy
equivalences.
Proposition 5.2. E is a topological operad, and the projections
fD
π1
←−−−−− E
π1
−−−−−→ C
are morphisms of operads.
6 Algebras from topological groups
The aim of this paper is to compare the fDalgebra Ω
2
BG with the Calgebra
ΩG by constructing an intermediate Ealgebra εG. This section introduces all
three algebras using the realization systems for fD and C, and proves Theo
rem B, which states that the three are weakly equivalent as Ealgebras.
We begin in §6.1 with Ω
2
BG. There is little to say in this case, but we
include it for emphasis. Then §6.2 discusses ΩG and §6.3 discusses εG. Finally
§6.4 gives some deferred proofs.
6.1 The fDalgebra Ω
2
BG
In order to demonstrate how realization systems on an operad can be used to
deﬁne algebras over that operad, we will brieﬂy describe the fDalgebra Ω
2
BG.
Proofs are omitted. First note that, for a ﬁxed a ∈ fD(n), there is a pushout
diagram
¸
n
i=1
S
1
∂i
¸
n
i=1
D
2
a
a
D
2
.
Lemma 6.1. Given a ∈ fD(n) and x
1
, . . . , x
n
∈ Ω
2
BG, there is a unique map
ξ : D
2
→ BG with the following two properties.
1. Its restriction to a is constant.
2. Its restriction to the ith cofactor of
¸
n
i=1
D
2
is given by x
i
.
Deﬁnition 6.2. Deﬁne ν
n
: fD(n) ×(Ω
2
BG)
n
→ Ω
2
BG by
ν
n
(a; x
1
, . . . , x
n
) = ξ,
where ξ is the map of Lemma 6.1.
Proposition 6.3. The collection ν = {ν
n
} makes Ω
2
BG into a fDalgebra.
17
6.2 The Calgebra ΩG
We now explain how to make ΩG into a Calgebra. This algebra structure was
ﬁrst obtained by Salvatore in [12] using his proof of the topological cyclic Deligne
conjecture. We will describe it directly in terms of the realizations RC.
Notation 6.4. Let X be a space and let Y be a Gspace. Given a continuous
map f : X → Y and an element g ∈ G, we write g · f : X → Y for the map that
sends x ∈ X to g · f(x). Such maps will be called left translates of f.
Lemma 6.5. Let c be a cactus with n lobes and let γ
1
, . . . , γ
n
be elements of
ΩG. Then there is a unique map α: c → G satisfying:
1. α(•) = e, where • denotes the global marked point in c;
2. for each i = 1, . . . , n, the composite α ◦ ∂
i
: S
1
→ G is a left translate of
γ
i
.
Proof. The realization c is a treelike conﬁguration of circles, so after relabelling
the lobes we may assume that ∂
1
(S
1
) contains • and that each ∂
i+1
(S
1
) meets
∂
1
(S
1
)∪· · · ∪∂
i
(S
1
) in a single point. There is a unique α
1
: ∂
1
(S
1
) → G sending
• to e and such that α
1
◦ ∂
1
is a left translate of γ
i
. It has a unique extension
α
2
: ∂
1
(S
1
) ∪ ∂
2
(S
1
) → G for which α
2
◦ ∂
2
is a left translate of γ
2
. Proceeding
in this way, the claim follows.
Deﬁnition 6.6. Deﬁne ω
n
: C(n) ×(ΩG)
n
→ ΩG by
ω
n
(c; γ
1
, . . . , γ
n
) = α ◦ ∂
out
,
where α: c → G is the map of Lemma 6.5.
Proposition 6.7. The collection ω = {ω
n
} makes ΩG into a Calgebra.
The proof will be given in §6.4. The action ω encodes some natural properties
of ΩG. For example, given s ∈ S
1
, let c
s
∈ C(1) denote the cactus in which
• = ∂
1
(s) and ∂
out
has constant speed. Then
ω
1
(c
−
, −): S
1
×ΩG → ΩG
is the circle action that sends (s, γ) to the loop t → γ(s)
−1
· γ(s + t). This
circle action was considered by Menichi in [11] and by Salvatore in [12]. Now let
c
P
∈ C(2) denote the cactus in which two circles of radius 1/2 are joined at their
basepoint, and the global marked point lies on the ﬁrst circle at its basepoint:
Then the map
ω
2
(c
P
; −, −): ΩG×ΩG → ΩG
is the ordinary Pontrjagin product given by concatenating loops. More generally,
there is a natural inclusion of the little intervals operad into C, under which the
algebra structure ω pulls back to the standard E
1
algebra structure on ΩG.
18
6.3 The Ealgebra εG
Now we will deﬁne the space εG. We will give it the structure of an Ealgebra,
and show that it is weakly equivalent to both Ω
2
BG and ΩG as Ealgebras.
The algebra structure on εG is an almost literal mixture of the existing algebra
structures on Ω
2
BG and ΩG.
Let EG → BG be the universal principal G bundle. Fix a basepoint ∗ of EG
and use it to deﬁne a basepoint of BG and an inclusion ι
G
: G ֒→ EG, g → g · ∗.
Deﬁnition 6.8. Let εG be the space of maps φ: D
2
→ EG for which φ
S
1
factors through ι
G
, and for which φ(∗) = ι
G
(e). There are projections
p
1
: εG → Ω
2
BG, p
2
: εG → ΩG,
where p
1
is obtained by projecting from EG to BG, and p
2
is deﬁned by ι
G
◦
p
2
(φ) = φ
S
1 .
We will again make use of the pushout diagram
¸
n
i=1
S
1
∂i
¸
n
i=1
D
2
a
a
D
2
(5)
determined by an element a ∈ fD(n).
Lemma 6.9. Given (a, c, f) ∈ E(n) and φ
1
, . . . , φ
n
∈ εG, there is a unique map
φ: D
2
→ EG with the following properties:
1. Its restriction to a factors through f : a → c.
2. Its restriction to the ith cofactor of
¸
n
i=1
D
2
is a left translate of φ
i
.
3. Its value on ∗ ∈ D
2
is ι
G
(e).
Note that ψ ∈ εG.
Proof. The conditions mean that ψ
a
must factor as ι
G
◦ α ◦ f, where α is the
map obtained by applying Lemma 6.5 to c and p
2
(φ
1
), . . . , p
2
(φ
n
). There is a
unique extension of this map to D
2
satisfying the second condition.
Deﬁnition 6.10. Deﬁne ε
n
: E(n) ×(εG)
n
→ εG by
ε
n
((a, c, f), (φ
1
, . . . , φ
n
)) = ψ,
where ψ is the map of Lemma 6.9.
The following two propositions together prove Theorem B. Their proofs are
given in the next subsection.
Proposition 6.11. The collection ε = {ε
n
} makes εG into a Ealgebra.
Proposition 6.12. The projection maps p
1
: εG → Ω
2
BG, p
2
: εG → ΩG are
homotopy equivalences of E algebras.
19
6.4 Proof of Propositions 6.7, 6.11 and 6.12
In this subsection we resume the practice of using underlines to indicate tuples
of elements. For example, given γ
i
1
, . . . , γ
i
mi
for each i = 1, . . . , n, we write γ for
γ
1
1
, . . . , γ
n
mn
and we write γ
i
for γ
i
1
, . . . , γ
i
mi
.
Proof of Proposition 6.7. Axioms 1, 2 and 3 of Deﬁnition 2.2 show that the
unit element 1
C
acts as the identity on ΩG, and that the equivariance property
ω
n
(cσ; γ
1
, . . . , γ
n
) = ω
n
(c; γ
σ
−1
1
, . . . , γ
σ
−1
(n)
) holds.
Now let us show that ω is associative. Let c ∈ C(n), d
i
∈ C(m
i
) and
γ
i
1
, . . . , γ
i
mi
∈ ΩG for i = 1, . . . , n. We must show that
ω
m1+···+mn
(γ(c; d); γ) = ω
n
(c; ω
mi
(d
i
; γ
i
)).
We can use Lemma 6.5 to obtain three diﬀerent maps.
1. Let α: γ(c; d
1
, . . . , d
n
) → Gdenote the map obtained using γ(c; d
1
, . . . , d
n
)
and γ
1
1
, . . . , γ
n
mn
.
2. Let β : c → Gdenote the map obtained using c and the ω
mi
(d
i
; γ
i
1
, . . . , γ
i
mi
).
3. Let γ
i
: d
i
 → G denote the map obtained using d
i
and γ
i
1
, . . . , γ
i
mi
.
We must show that α ◦ ∂
out
= β ◦ ∂
out
. We can ﬁnd g
1
, . . . , g
n
∈ G and two
commutative squares:
¸
S
1
∂out
∂i
d
1
 ⊔ · · · ⊔ d
n

gi·γ
i
c
β
G
¸
S
1
∂out
∂i
d
1
 ⊔ · · · ⊔ d
n

µ
c
λ
γ(c; d
1
, . . . , d
n
)
The ﬁrst square comes from the deﬁnition of β. The second is a pasting square
(1), and in particular is a pushout. Comparing the squares we see that β
factors as α
′
◦ λ for some α
′
: γ(c; d
1
, . . . , d
n
) → G. The compatibility of λ and
µ with the boundary maps means that α
′
in fact satisﬁes the properties that
characterize α, so that α
′
= α. Thus β = α ◦ λ, and by the compatibility of λ
with the boundary maps we have β ◦ ∂
out
= α ◦ λ ◦ ∂
out
= α ◦ ∂
out
as required.
Let us show that ω
n
is continuous. Lemma 6.5 gives us a function
α: RC(n) ×(ΩG)
n
→ G
and it suﬃces to show that this is continuous. The map
¸
∂
i
:
¸
S
1
× C(n) →
RC(n) is proper surjection of ﬁbred spaces, so by Proposition A.5 it will suﬃce
to show that each composite α ◦ ∂
i
is continuous. This composite is given by
(t, c, γ
1
, . . . , γ
n
) → g
i
· γ
i
(t) for some g
i
∈ G. It will therefore suﬃce to show
that the assignments C(n) × (ΩG)
n
→ G, (c, γ
1
, . . . , γ
n
) → g
i
are continuous.
These maps factor through a map F(n) ×(ΩG)
n
→ G, whose continuity can be
proved using the cell decomposition of F(n) given in Note 4.7.
20
Proof of Proposition 6.11. To begin we must show that ε satisﬁes the unit,
equivariance and associativity rules. We will only prove associativity as the
other two properties can be proved in a similar but much simpler way. Take
(a, c, f) ∈ E(n) and (b
i
, d
i
, g
i
) ∈ E(m
i
) for i = 1, . . . , n, and write their compos
ite as (A, C, F). Take φ
i
j
∈ εG for i = 1, . . . , n and j = 1, . . . , m
i
. We must
show that
ε
m1+···+mn
(A, C, F); φ
= ε
n
(a, c, f); ε
mi
(b
i
, d
i
, g
i
); φ
i
. (6)
Consider the following diagram of pushout squares.
¸
i
¸
j
S
1
¸
i
¸
j
D
2
¸
i
S
1
¸
i
b
i

¸
i
D
2
a
A
D
2
The upper square comes from (5) for the b
i
; the right hand two squares compose
to give (5) for A, and the lower two squares compose to give (5) for a; the left
hand square is the pasting square for A. Using this diagram we can characterise
the right hand side of (6) as the map ψ: D
2
→ EG that
1. when restricted to a, factors through f : a → c;
2. when restricted to b
i
, factors through g
i
: b
i
 → d
i
;
3. on the (i, j)th cofactor D
2
is a left translate of φ
i
j
;
4. sends ∗ to ι
G
(e).
The ﬁrst two of these properties, together with the deﬁnition of F, show that
the restriction of ψ to A factors through F : A → C. This property, together
with with the third and fourth properties above, characterize the left hand side
of (6). Equality follows. Thus ε makes εG into a Ealgebra in sets.
To show that ε makes εG into a Ealgebra in the topological setting we will
show that ε
n
: E(n)×(εG)
n
→ εG is continuous. To do this we will show that the
adjoint D
2
×E(n) ×(εG)
n
→ EG is continuous. Since the lower and righthand
maps of diagram (5) give inclusions of closed subsets
¸
D
2
×E(n) ֒→ D
2
×E(n)
and π
∗
1
RfD(n) ֒→ D
2
×E(n), it suﬃces to show continuity of the restrictions
ξ :
¸
D
2
×E(n) ×(εG)
n
→ EG, η: π
∗
1
RfD(n) ×(εG)
n
→ EG.
For ξ this follows from continuity of the map ((a, c, f), (φ
1
, . . . , φ
n
)) → (g
1
, . . . , g
n
)
in the proof of Proposition 6.7. For η it follows from continuity of two maps
α: RC(n) ×(ΩG)
n
→ G, β : π
∗
1
RfD(n) → RC(n).
Here β is given in the ﬁbre over (a, c, f) by x → f(x), and is continuous by
part 5 of Proposition A.8, while α was deﬁned, and shown to be continuous, in
the proof of Proposition 6.7.
21
Proof of Proposition 6.12. It is immediate that both p
1
and p
2
are Ealgebra
morphisms. We must now show that both p
1
and p
2
are homotopy equivalences.
We ﬁrst show that p
1
, p
2
are ﬁbrations. For p
2
the homotopy lifting problem
is adjoint to an extension problem that can always be solved. For p
1
the homo
topy lifting problem is adjoint to the problem of ﬁnding a lift F in a diagram
of the form
X ×D
2
×{0}
h
EG
X ×D
2
×I
H
F
r
r
r
r
r
r
BG
with the additional property that F(X × ∗ × I) is constant with value ι
G
(e).
To solve this, ﬁrst choose an arbitrary lift F
′
. Write F
′
X ×∗ ×I = ι
G
◦ f, and
then deﬁne F by F(x, d, s) = F
′
(x, d, s) · f(x, s)
−1
.
It now suﬃces to show that π
1
and π
2
have contractible ﬁbres. But p
−1
1
(x)
is homeomorphic to the space of based maps D
2
→ G, while p
−1
2
(γ) is the space
of maps D
2
→ EG whose restriction to S
1
is ι
G
◦ γ. These are contractible, by
contractibility of D
2
and EG respectively.
7 Proof of Theorem A
This section gives the proof of Theorem A, stating certain results that will be
proved in the remaining sections of the paper. The theorem states that the
projections π
1
: E → fD and π
2
: E → C are weak homotopy equivalences of
operads. This means that for each n 0, the maps π
1
: E(n) → fD(n) and
π
2
: E(n) → C(n) are weak equivalences of spaces. To show this we will consider
the product
Π: E(n) → fD(n) ×C(n)
of π
1
and π
2
. It would be ideal for us if Π were a ﬁbration, but this is not the
case: a typical ﬁbre is nonempty, but the ﬁbre over a pair (a, c) can be empty
if the little discs of a meet the boundary of the big disc, for then the boundary
conditions can be overdetermined. This issue is easily remedied.
Deﬁnition 7.1. Let fD
◦
(n) ⊂ fD(n) consist of those elements whose little
discs do not meet the boundary of the big disc, and let E
◦
(n) = π
−1
1
fD
◦
(n) ⊂
E(n).
Lemma 7.2. The inclusions fD
◦
(n) ֒→ fD(n) and E
◦
(n) ֒→ E(n) are homo
topy equivalences.
Proof. Given a ∈ fD
◦
(n), write a
1/2
for the element obtained by halving the
radii of the little discs of a. Write ρ
a
: a
1/2
 → a for the map that preserves
a ⊂ a
1/2
 and that projects points inside a little disc of a to the boundary of
that little disc. Then the maps fD(n) → fD
◦
(n), a → a
1/2
and E(n) → E
◦
(n),
(a, c, f) → (a
1/2
, c, f ◦ ρ
a
) are homotopy inverse to the inclusions above.
To prove Theorem A it will therefore suﬃce to prove that the restrictions
π
1
: E
◦
(n) → fD
◦
(n) and π
2
: E
◦
(n) → C(n) are weak equivalences. To do this
we now consider the map
Π: E
◦
(n) → fD
◦
(n) ×C(n),
22
which has much better ﬁbrewise properties than its predecessor.
Theorem C. Fix (a, c) ∈ fD
◦
(n) × C(n) and write Map
∂
(a, c) for the ﬁbre
of E
◦
(n) over (a, c). Then there is a neighbourhood U of (a, c) over which we
can ﬁnd a ﬁbrewise homotopy equivalence
E
◦
(n)U ≃ U ×Map
∂
(a, c).
Thus each point of fD
◦
(n) ×C(n) has arbitrarily small neighbourhoods over
which Π is a quasiﬁbration. By Corollary 2.4 of [9], such maps are themselves
quasiﬁbrations.
Corollary 7.3. The projection Π: E
◦
(n) → fD
◦
(n) ×C(n) is a quasiﬁbration.
We will use the quasiﬁbration Π to show that π
1
and π
2
are weak equiva
lences. Write PRB
n
for the pure ribbon braid group. Then both fD
◦
(n) and
C(n) are EilenbergMacLane spaces K(PRB
n
; 1). (The relevant background is
recalled in §8.1 and §8.2.) It will therefore suﬃce to show that π
i
(E
◦
(n)) is
trivial for i = 1, and that Π
∗
: π
1
(E
◦
(n)) → PRB
n
×PRB
n
identiﬁes π
1
(E
◦
(n))
with the diagonal subgroup of PRB
n
×PRB
n
.
Write F
n
for a typical ﬁbre of Π. For Π to be a quasiﬁbration means that
F
n
is weakly equivalent to the homotopy ﬁbre of Π, or in other words that there
is a long exact sequence of homotopy groups:
· · · −→ π
i
(F
n
) −→ π
i
(E
◦
(n)) −→ π
i
(fD
◦
(n)) ×π
i
(C(n)) −→ · · · (7)
We will establish the criteria listed in the last paragraph by exploiting this long
exact sequence.
Proposition 7.4. There is a homotopy equivalence F
n
→ PRB
n
. In particular
F
n
has contractible components.
It follows immediately from (7) that π
i
(E
◦
(n)) is trivial for i > 1, and that
π
1
(E
◦
(n)) and π
0
(E
◦
(n)) are computed using an exact sequence
0 → π
1
(E
◦
(n))
p
−→ PRB
n
×PRB
n
q
−→ PRB
n
r
−→ π
0
(E
◦
(n)) → 0. (8)
There is an action of PRB
n
× PRB
n
on PRB
n
, and q is given by applying
this action to the constant element ∗. Exactness at the last two terms of the
sequence means that q
−1
(∗) is the image of p and r
−1
(∗) is the image of q. We
can therefore show that π
0
(E
◦
(n)) = ∗ and that p identiﬁes π
1
(E
◦
(n)) with the
diagonal subgroup of PRB
n
× PRB
n
by proving the next proposition, which
completes the proof of Theorem A.
Proposition 7.5. The action of PRB
n
×PRB
n
on PRB
n
is given by (γ, δ)·φ =
δ · φ · γ
−1
.
The promised recollections on the homotopy type of fD
◦
(n) and C(n), to
gether with the proof of Propositions 7.4 and 7.5, are given in §8.
Now let us discuss the proof of Theorem C. This theorem states that E
◦
(n)
is locally ﬁbrewise homotopy equivalent to a product, or in other words, that
it is a homotopy ﬁbre bundle. Recall that E
◦
(n) consists of maps from ﬁbres
of RfD
◦
(n) to ﬁbres of RC(n) satisfying certain boundary conditions. Our
23
proof of Theorem C will follow from general results on spaces of maps satisfying
boundary conditions, together with a study of the local ﬁbrewise structure of
RfD
◦
(n) and RC(n).
In Section 9 we consider pairs (X, f
X
) and (Y, f
Y
), where X and Y are
spaces and f
X
: A → X and f
Y
: A → Y are maps from some ﬁxed space A. We
then introduce two mapping spaces, one consisting of maps g : X → Y relative
to A, or in other words satisfying g ◦ f
X
= f
Y
, and a second consisting of
maps g : X → Y homotopyrelative to A, which are equipped with homotopies
g◦f
X
≃ f
Y
. We show that under certain conditions the two mapping spaces are
homotopy equivalent, and that homotopy type of the space of homotopyrelative
maps does not change when we replace (X, f
X
) or (Y, f
Y
) with a homotopy
equivalent pair (Z, f
Z
).
The program outlined above extends without modiﬁcation to the ﬁbre
wise setting, where we can consider the pairs (RfD
◦
(n), ∂) over fD
◦
(n) and
(RC(n), ∂) over C(n); here ∂ denotes the combined boundary map. The corre
sponding relative mapping space is E
◦
(n), and it is homotopy equivalent to the
homotopyrelative mapping space E
◦
h
(n).
In Sections 10 and 11 we study the pairs (RfD
◦
(n), ∂) and (RC(n), ∂) re
spectively. We show that, locally over fD
◦
(n) and C(n), these pairs are ho
motopy equivalent to trivial ﬁbred pairs of the form (a ×fD
◦
(n), ∂ ×Id) and
(c ×C(n), ∂ ×Id) respectively.
Finally, in Section 12 we combine the results of the preceding three sections
to prove Theorem C.
8 Computing the long exact sequence
This section is given to the proof of Propositions 7.4 and 7.5. Recall that these
propositions compute the homotopy exact sequence associated to the quasiﬁbra
tion Π: E
◦
(n) → fD
◦
(n) × C(n), and lead to the proof of Theorem A. In §8.1
and §8.2 we recall the pure ribbon braid group and its relationship to framed
discs and cacti. Then in §8.3 and §8.4 we prove the propositions.
8.1 The pure ribbon braid group
This subsection collects some facts about the pure ribbon braid group that will
be useful in the subsections to follow.
Recall, for example from [7], that E. Artin’s braid group B
n
on n strands is
the group on generators σ
1
, . . . , σ
n−1
subject to the relations σ
i
σ
j
= σ
j
σ
i
for
i − j 2 and σ
i
σ
i+1
σ
i
= σ
i+1
σ
i
σ
i+1
. There is a homomorphism B
n
→ Σ
n
sending σ
i
to the transposition of i and i +1. Its kernel is the pure braid group
PB
n
, which has generators
α
ij
= σ
j−1
σ
j−2
· · · σ
i+1
σ
2
i
σ
−1
i+1
· · · σ
−1
j−2
σ
−1
j−1
for 1 i < j n. The pure ribbon braid group is the direct product PRB
n
=
PB
n
×Z
n
. The generator of the ith cyclic factor is written as ζ
i
.
Let F
n
denote the free group on generators x
1
, . . . , x
n
. There is an embed
ding B
n
֒→ Aut(F
n
) sending σ
i
to the transformation
x
i
→ x
i
x
i+1
x
−1
i
, x
i+1
→ x
i
, x
l
→ x
l
for l = i, i + 1.
24
This embedding identiﬁes B
n
with the subgroup consisting of those automor
phisms f : F
n
→ F
n
for which f(x
1
· · · x
n
) = x
1
· · · x
n
and for which there is a
permutation π of {1, . . . , n} so that each f(x
i
) is conjugate to x
π(i)
. The same
embedding identiﬁes PB
n
with the analogous subgroup where the permutation
π is the identity map.
In later subsections we will use the following alternative description of the
pure ribbon braid group, which is more directly related to the topological situ
ation at hand. Given w = (w
1
, . . . , w
n
) ∈ (F
n
)
n
let α(w) denote the homomor
phism F
n
→ F
n
that sends x
i
to w
i
x
i
w
−1
i
. Set
W
n
= {w ∈ (F
n
)
n
 α(w) ∈ Aut(F
n
), α(w)(x
1
· · · x
n
) = x
1
· · · x
n
}.
There is a bijection Λ: W
n
∼
=
−→ PRB
n
. The ﬁrst component of Λ(w) is α(w) ∈
PB
n
; the second component is (m
1
, . . . , m
n
) ∈ Z
n
, where m
i
is the sum of the
exponents of the letter x
i
in the word w
i
. The isomorphism Λ translates the
group structure on PRB
n
into the operation
(v
1
, . . . , v
n
) · (w
1
, . . . , w
n
) = (α(v)(w
1
)v
1
, . . . , α(v)(w
n
)v
n
).
8.2 Discs, cacti and the pure ribbon braid group
This subsection recalls how both fD(n) (or equivalently fD
◦
(n)) and C(n) are
EilenbergMacLane spaces K(PRB
n
, 1). We begin by choosing basepoints.
Deﬁnition 8.1. Let a
n
∈ fD
◦
(n) be an element whose little discs all have
the same (small) radius, are arranged horizontally across the big disc in order
1, . . . , n, and which are embedded without rotation, so that the local marked
point lies at the top of each little disc. For i = 1, . . . , n let l
i
denote the line
segment in a
n
 passing from the basepoint • of the big disc to the basepoint
∂
i
(∗) of the ith little disc.
a
n 1 n
l1
··· ln
We will sometimes regard the l
i
as paths l
i
: [0, 1] → a
n
, and sometimes as
subsets l
i
⊂ a
n
. The fundamental group π
1
(a
n
, •) can be identiﬁed with F
n
;
the ith generator x
i
is represented by the loop which travels along l
i
, then
around the ith circle and back down l
i
.
Deﬁnition 8.2. Let c
n
denote the cactus in which all lobes have equal length
and meet at a unique point, and whose outgoing boundary ∂
out
: S
1
→ c
n
 has
constant speed. The intersection point coincides with the global marked point
and also the local marked point on each lobe. The lobes are cyclically ordered
n, . . . , 1 at the intersection point, and the global marked point is chosen to lie
on the nth lobe. Thus the outgoing boundary map ∂
out
: S
1
→ c traverses the
lobes n, n −1, . . . , 1 in turn at constant speed.
c
n
n 1
25
The fundamental group π
1
(c
n
, •) can be identiﬁed with the free group F
n
on
generators x
1
, . . . , x
n
. Here x
i
is represented by the loop ∂
i
: S
1
→ c.
It is well known that fD(n) (or fD
◦
(n)) is a K(PRB
n
, 1). For instance,
fD(n) is homotopy equivalent to the product of n circles with the conﬁguration
space Conf(n, R
2
) of n ordered points in R
2
, and Conf(n, R
2
) is a K(PB
n
, 1)
(see section 1.4.1 of [7], for example). The generator α
ij
corresponds to the loop
based at a
n
in which the ith and jth little discs are brought side by side beneath
the other little discs, and are then rotated around one another anticlockwise
before returning to their original position. Throughout this maneouver the
framing of the little discs does not change. The generator ζ
i
corresponds to the
loop in which the framing of the ith little disc is rotated through one full turn.
A result of R. Kaufmann [8, Proposition 3.3.19] states that C(n) is a K(PRB
n
, 1).
The generator α
ij
corresponds to the loop based at c
n
in which the jth lobe
travels across the (i + 1), . . . , (j − 1)th lobes so that it is adjacent to the ith
lobe; the ith and jth lobes then rotate around one another anticlockwise be
fore the jth retraces its steps to its original position. (This description of the
generator is taken from the proof of Kaufmann’s result quoted above. In order
to obtain this description we have reversed the path α
i
in the proof there; this
does not aﬀect the outcome of that proposition, since the braid group admits
an automorphism inverting all of its standard generators.) The generator ζ
i
corresponds to the loop in which the parameterization of the ith little disc is
rotated through one full turn.
8.3 The ﬁbre of Π: E
◦
(n) → fD
◦
(n) ×C(n)
The purpose of this subsection is to prove Proposition 7.4, which describes the
homotopy type of the ﬁbre of Π.
Deﬁnition 8.3. Let F
n
denote the ﬁbre of Π: E
◦
(n) → fD
◦
(n) × C(n) over
(a
n
, c
n
). It is the space of homotopy equivalences f : a
n
 → c
n
 such that
f ◦ ∂
i
= ∂
i
and f ◦ ∂
out
= ∂
out
, equipped with the compactopen topology.
Deﬁnition 8.4. Deﬁne a map F
n
→ W
n
as follows. A point f ∈ F
n
is sent to
(w
1
, . . . , w
n
), where w
i
∈ F
n
= π
1
(c
n
, •) is obtained by applying f to the line
segment l
i
. The boundary conditions guarantee that the image of f ◦l
i
is a loop
in c
n
.
We will prove Proposition 7.4 by showing that this map F
n
→ W
n
is a
homotopy equivalence. The proof is given in the next three lemmas.
Lemma 8.5. The ﬁbre F
n
has weakly contractible components.
Proof. It will suﬃce to prove that, for any f ∈ F
n
, the loopspace Ω
f
(F
n
) is
contractible. Note that Ω
f
(F
n
) is the space of sections of the ﬁbration Ω
f
c
n
 →
a
n
 obtained by pulling back the free loopspace ﬁbration Ωc
n
 → Lc
n
 → c
n

along f. Its ﬁbre over x is Ω
f(x)
c
n
. We must show that the space of sections
of this ﬁbration is contractible.
Since c
n
 is a K(F
n
, 1), Ω
f
c
n
 is ﬁbre homotopy equivalent to a covering
space with ﬁbre π
1
(c
n
, f(x))
∼
= F
n
over x. We must show that this covering
space admits a unique section. The covering space is determined by an action
of the fundamental group π
1
(a
n
, •) of the base on the ﬁbre π
1
(c
n
, •) over •.
The sections of the covering space correspond to the ﬁxed points of this action.
26
If φ ∈ Aut(F
n
) is the automorphism determined by f, then the action of
π
1
(a
n
, •) = F
n
on π
1
(c
n
, •) = F
n
is given by x
i
(ω) = φ(x
i
)ωφ(x
i
)
−1
. Since φ
is an automorphism, ω is ﬁxed by the action if and only if it commutes with every
element of F
n
. The identity element is the unique element with this property,
and this completes the proof.
Lemma 8.6. The map F
n
→ W
n
is surjective.
Proof. Write J
n
⊂ a
n
 for the union of the boundaries of the little discs with
the line segments l
i
. Write K
n
⊂ a
n
 for the union of J
n
with the boundary
of the big disc D
2
. Choose w = (w
1
, . . . , w
n
) ∈ W
n
and represent each w
i
by a
based loop γ
i
: I → c
n
. Let f : J
n
→ c
n
 be the map that satisﬁes f ◦ ∂
i
= ∂
i
for each i and that is given by γ
i
on l
i
. Now J
n
is a strong deformation retract
of a
n
, and so we may extend f to a map g : a
n
 → c
n
. This map satisﬁes
g ◦ ∂
i
= ∂
i
. Also, since w ∈ W
n
it follows that g is a homotopy equivalence,
and that g ◦ ∂
out
is based homotopic to ∂
out
. Then we can form a homotopy
between gK
n
and a map K
n
→ c
n
 that commutes with the ∂
i
and ∂
out
, and
that is given by the γ
i
on l
i
. Since K
n
֒→ a
n
 is a coﬁbration, we may extend
the homotopy to one between g and some h: a
n
 → c
n
. By construction h is
an element of F
n
and its image in W
n
is exactly w.
Lemma 8.7. Let f
1
, f
2
∈ F
n
. If the images of these elements in W
n
coincide,
then they lie in the same component of F
n
.
Proof. Let K
n
⊂ a
n
 be as in the preceding proof. Now f
1
and f
2
coincide on
the image of every boundary map and are homotopic rel. endpoints on each l
i
.
Since K
n
֒→ a
n
 is a coﬁbration, we may therefore apply a homotopy to f
2
,
through elements of F
n
, until it coincides with f
1
on all of K
n
. Let us therefore
assume that f
1
and f
2
coincide on K
n
.
We may now combine f
1
and f
2
to obtain a map f : a
n
 ∪
Kn
a
n
 → c
n
. Let
a
n
 denote the space obtained from a
n
 ×I by identifying K
n
×I with a single
copy of K
n
. It has ‘boundary’ a
n
 ∪
Kn
a
n
. To produce a homotopy between
f
1
and f
2
it will suﬃce to extend f to all of a
n
.
Choose an identiﬁcation map D
2
→ a
n
 that is a bijection away from the
boundary S
1
. Then we obtain identiﬁcation maps S
2
→ a
n
 ∪
Kn
a
n
 and
B
3
→ a
n
. To extend f to a
n
 it will suﬃce to extend the composite S
2
→
a
n
 ∪
Kn
a
n

f
−→ c
n
 to B
3
. But since c
n
 is a K(F
n
, 1) this is always possible.
This completes the proof.
8.4 The action of π
1
(fD
◦
(n) ×C(n)) on F
n
This subsection will give the proof of Proposition 7.5, which states that the
fundamental group PRB
n
× PRB
n
of fD
◦
(n) × C(n) acts on π
0
(F
n
)
∼
= PRB
n
by the rule (γ, δ) · φ = δ · φ · γ
−1
. The proof consists of Lemma 8.8, where we
show that (γ, γ) · 1 = 1, and Lemma 8.9, where we show that (γ, 1) · φ = φ· γ
−1
.
In general, if F → E → B is a ﬁbration, then the action of π
1
(B) on π
0
(F) is
deﬁned as follows. Take a loop γ in B and a point f in F. Then γ can be lifted
to a path in E with initial point f ∈ F and ﬁnal point g ∈ F, and the eﬀect
of [γ] on [f] is given by [γ] · [f] = [g]. If F → E → B is only a quasiﬁbration
then the lifting process just described may not be possible. However, when it is
possible, then it still leads to a description of the action of π
1
(B) on π
0
(F).
27
Choose a basepoint f
n
∈ F
n
, where f
n
is any function sending each l
i
to the
constant path.
Lemma 8.8. The action of Proposition 7.5 satisﬁes (γ, γ) · 1 = 1 for all γ ∈
PRB
n
.
Proof. It will suﬃce to prove this for γ = ζ
i
and for γ = α
ij
. Loops in fD
◦
(n)
and C(n) representing these elements were described in §8.2.
Let us begin with the case γ = ζ
i
. Let t → a
n
(t) and t → c
n
(t) be the loops
in fD
◦
(n) and C(n) that represent ζ
i
. Then t → (a
n
(t), c
n
(t), f
n
) determines a
loop in E
◦
(n) and our claim follows.
Now ﬁx γ = α
ij
and let t → a
n
(t), t → c
n
(t) be the loops in fD
◦
(n) and
C(n) respectively that represent α
ij
. Then for each t ∈ S
1
we can ﬁnd maps
f
n
(t): a
n
(t) → c
n
(t) such that t → (a
n
(t), c
n
(t), f
n
(t)) deﬁnes a loop in E
◦
(n).
For at all times t ∈ S
1
we can ﬁnd an oriented embedding of c
n
(t) into a
n
(t)
that sends the basepoint to the basepoint, such that the image of the ith lobe
separates the ith little disc from the others, and such that the highest point
each lobe is its local marked point. Then the required f
n
(t) can be constructed
by projecting out from the centre of the ith little disc onto the ith lobe, and
inwards from the boundary of the big disc onto the cactus.
For example take n = 2, i = 1, j = 2. Then a
n
(t) and c
n
(t) are depicted
here:
1 2
1
2
1 2
1
2
1 2
1 2
1
2
1 2
1
2
1 2
t = 0 0 < t < π t = π π < t < 2π t = 2π
And the family of embeddings c
n
(t) → a
n
(t) is depicted here:
Lemma 8.9. The action of Proposition 7.5 satisﬁes (γ, 1) · φ = φ · γ
−1
for all
γ ∈ PRB
n
.
Proof. We identify PRB
n
with W
n
. Represent φ ∈ W
n
by (a
n
, c
n
, f) ∈ F
n
.
Represent γ by a loop t → a
n
(t) in fD
◦
(n) based at a
n
, so that (γ, 1) is
represented by t → (a
n
(t), c
n
). The representative of γ can be chosen so that
there is a continuous family of homeomorphisms h
n
(t): a
n
 → a
n
(t), t ∈ [0, 1],
satisfying h
n
(t) ◦ ∂
i
= ∂
i
and h
n
(t) ◦ ∂
out
= ∂
out
for all t. Then we can form a
28
path in E
◦
(n) given by t → (a
n
(t), c
n
, f ◦ h
n
(t)
−1
). The endpoint of this path
is (a
n
, c
n
, f ◦ h
n
(1)
−1
) ∈ F
n
.
We must show that f ◦ h
n
(1)
−1
∈ F
n
represents φ · γ
−1
. To h
n
(1) we assign
an element w = (w
1
, . . . , w
n
) ∈ W
n
as follows. The arc h
n
(1)(l
i
) in a
n
 has the
same endpoints as l
i
, and so is homotopic rel. endpoints to the concatenation
of a loop ω
i
based at • with the arc l
i
. Then w
i
= [ω
i
]. Using the description
of the group operation on W
n
, it follows that f ◦ h
n
(1)
−1
represents precisely
φ · w
−1
.
So we must show that w = γ. In fact the assignment h
n
(1) → w is a
homomorphism from the group of boundaryﬁxing homeomorphisms of a
n
 into
W
n
, and so it suﬃces to prove w = γ in the special cases γ = α
ij
and γ = ζ
i
.
In these cases the family of homeomorphisms h
n
(t) can be seen explicity, and
the identities w = γ follow.
9 Relative mapping spaces
This section begins our work towards a proof of Theorem C. Recall that Theo
rem C describes the ﬁbrewise homotopy type of E
◦
(n) locally over fD
◦
(n)×C(n).
Recall also that E
◦
(n) is a ﬁbrewise mapping space, whose ﬁbre over (a, c) con
sists of homotopy equivalences f : a → c with the property that f ◦ ∂ = ∂.
Here we have written ∂ for the maps
∂
out
⊔ ∂
1
⊔ · · · ⊔ ∂
n
:
¸
n
i=0
S
1
→ a,
∂
out
⊔ ∂
1
⊔ · · · ⊔ ∂
n
:
¸
n
i=0
S
1
→ c.
In order to understand E
◦
(n), this section will study spaces of maps satisfying
a condition of the form f ◦ ∂ = ∂.
Fix a compact Hausdorﬀ space A. We will consider spaces under A, by
which we mean pairs (X, f
X
) consisting of a compact Hausdorﬀ space X and a
continuous map f
X
: A → X. There are no further assumptions on f
X
. Given
spaces (X, f
X
) and (Y, f
Y
) under A, we will consider spaces of maps g : X → Y
satisfying the strict condition g ◦ f
X
= f
Y
, which we call maps relative to A,
and also maps g : X → Y equipped with a homotopy g ◦f
X
≃ f
Y
, which we call
maps homotopyrelative to A.
The section has two parts. The ﬁrst part introduces what it means for two
spaces under A to be homotopy equivalent. The second part introduces two
mapping spaces between spaces under A, one space of maps relative to A, and
another space of maps homotopyrelative to A. We compare the two notions,
and we show that the homotopy type of the homotopyrelative mapping space
is preserved if one of the spaces under A is replaced by a homotopyequivalent
space under A.
Note 9.1. Here we will prove nonﬁbrewise versions of all results. However, all
results immediately generalize to the setting of spaces ﬁbred over B by replacing
‘compact’, ‘Hausdorﬀ’ and ‘compactopen’ with their ﬁbrewise analogues.
Notation 9.2. Homotopies will always be parameterized by [0, 1] = I, and
a homotopy H from f to g will be written H: f ⇒ g. Given homotopies
H
1
: g
0
⇒ g
1
and H
2
: g
1
⇒ g
2
, we write H
2
·H
1
: g
0
⇒ g
2
for their concatenation.
Given maps
A
k
−→ B
g,h
−−→ C
l
−→ D
29
and a homotopy H: g ⇒ h, we write H◦ k: g ◦ k ⇒ h◦ k and l ◦ H: l ◦ g ⇒ l ◦ h
and for the induced homotopies between the composites.
9.1 Homotopy equivalences of spaces under A
In this subsection we will study the following notion of homotopy equivalence
between spaces under A. It is somewhat nonstandard in the sense that, given
spaces (X, f
X
) and (Y, f
Y
) under A, we will consider maps g : X → Y that do
not necessarily satisfy g ◦ f
X
= f
Y
, but only satisfy a homotopy version of the
condition. Nevertheless this is the notion best suited to our purposes.
Deﬁnition 9.3. A homotopy equivalence from (X, f
X
) to (Y, f
Y
) is a 6tuple
(φ, ψ, G, H, K, L) consisting of maps
φ: X → Y, ψ: Y → X,
homotopies
G: ψ ◦ φ ⇒ Id
X
, H: φ ◦ ψ ⇒ Id
Y
,
and homotopies
K: φ ◦ f
X
⇒ f
Y
, L: ψ ◦ f
Y
⇒ f
X
such that the two homotopies L · (ψ ◦ K) and G ◦ f
X
from ψ ◦ φ ◦ f
X
to f
X
are homotopic relative to their endpoints, and such that the two homotopies
K · (φ ◦ L) and H ◦ f
Y
from φ ◦ ψ ◦ f
Y
to f
Y
are homotopic relative to their
endpoints. We write (X, f
X
) ≃ (Y, f
Y
) if there is a homotopy equivalence from
(X, f
X
) to (Y, f
Y
). See Lemma 9.5 below.
Example 9.4. 1. Let (X, f
X
) and (Y, f
Y
) be spaces under A. If X and
Y are homotopy equivalent relative to A, then (X, f
X
) and (Y, f
Y
) are
homotopy equivalent in the sense above.
2. If (X, f
X
) and (X, g
X
) are spaces under A with f
X
≃ g
X
, then (X, f
X
)
and (X, g
X
) are homotopy equivalent.
3. For any c
1
, c
2
∈ C(n) the pairs (c
1
, ∂) and (c
2
, ∂) are homotopy equiv
alent in the present sense, even though there are no maps between them
relative to A = S
1
⊔
¸
n
i=1
S
1
. See Section 11.
Lemma 9.5. Homotopy equivalence is an equivalence relation on spaces under
A.
Proof. The relation is evidently reﬂexive and symmetric. If (φ, ψ, G, H, K, L) is
a homotopy equivalence from (X, f
X
) to (Y, f
Y
) and (φ
′
, ψ
′
, G
′
, H
′
, K
′
, L
′
) is a
homotopy equivalence from (Y, f
Y
) to (Z, f
Z
), then
(φ
′
φ, ψψ
′
, G· (ψG
′
φ), H
′
· (φ
′
Hψ
′
), K
′
· (φ
′
K), L · (ψL
′
))
is a homotopy equivalence from (X, f
X
) to (Z, f
Z
). Thus the relation is transi
tive.
Lemma 9.6. Suppose given a space (X, f
X
) under A and a coﬁbrant embedding
i : T ֒→ X of a contractible space. Let (X/T, f
X/T
) be the space under A in which
f
X/T
is the composite of f
X
with X → X/T. Then (X, f
X
) and (X/T, f
X/T
)
are homotopy equivalent.
30
Proof. Let f : I × X → X extend the identity map {0} × X → X and a null
homotopy I ×T → T. Let φ: X → X/T be the collapse map and let ψ: X/T →
X be induced by f
1
= f{1}×X. Let G be the homotopy given by f. Let H be
the homotopy whose composite with φ is given by f; this H exists because f(I ×
T) ⊂ T. Let K to be the constant homotopy and let L be the homotopy obtained
from G. Then (φ, ψ, G, H, K, L) is the required homotopy equivalence.
9.2 Spaces of maps between spaces under A
Now we introduce two mapping spaces between spaces under A.
Deﬁnition 9.7. Let (X, f
X
) and (Y, f
Y
) be spaces under A. Set
Map
f
(X, Y ) = {g : X → Y  g ◦ f
X
= f
Y
}
and
Map
h
f
(X, Y ) = {(g, H)  g : X → Y, H: g ◦ f
X
⇒ f
Y
}.
These are topologized as subspaces of Map(X, Y ) and Map(X ∪
fX
A × I, Y )
respectively, both equipped with the compactopen topology. There is an inclu
sion Map
f
(X, Y ) ֒→ Map
h
f
(X, Y ) given by taking the constant homotopy.
Proposition 9.8. The inclusion Map
f
(X, Y ) ֒→ Map
h
f
(X, Y ) is a homotopy
equivalence so long as for any space K the map f
X
× Id
K
: A × K → X × K
has the homotopy extension property. This restricts to a homotopy equivalence
between the subspaces consisting of those elements for which the map g is a
homotopy equivalence.
Proof. This is a standard result, but we give a proof here in order to indicate
just how the given assumptions are used.
Since X and Y are compact Hausdorﬀ, continuous maps (X∪
fX
A×I)×K →
Y are in bijection with continuous maps K → Map((X∪
fX
A×I), Y ). Thus the
evaluation map α: (X∪
fX
A×I) ×Map
h
f
(X, Y ) → Y is continuous and extends
to a continuous map β : X×I ×Map
h
f
(X, Y ) → Y . By choosing an appropriate
retraction we can form a map γ : (A ×I) ×I ×Map
h
f
(X, Y ) → Y such that
γ((a, t), 0, m) = α(a, t, m),
γ((a, 0), s, m) = α(a, s, m),
γ((a, t), 1, m) = α(a, 1, m),
and such that if m ∈ Map
f
(X, Y ) then γ((a, t), s, m) = m(a) is independent of t
and s. Now β and γ combine to give a map ω: (X∪
fX
A×I)×I×Map
h
f
(X, Y ) →
Y that has an adjoint φ: I ×Map
h
f
(X, Y ) → Map
h
f
(X, Y ).
By construction, φ{0} ×Map
h
f
(X, Y ) is the identity; φ{1} ×Map
h
f
(X, Y )
has image contained in Map
f
(X, Y ); and φI ×Map
f
(X, Y ) has image con
tained in Map
f
(X, Y ). Thus φ{1} ×Map
h
f
(X, Y ) is the required homotopy
inverse.
The next result shows that the homotopy type of Map
h
f
(X, Y ) depends only
on the homotopy equivalence class of (X, f
X
) and (Y, f
Y
).
31
Proposition 9.9. Let (φ, ψ, G, H, K, L) be a homotopy equivalence from (Y, f
Y
)
to (
˜
Y , f
˜
Y
). Then the two maps
Φ: Map
h
f
(X, Y ) → Map
h
f
(X,
˜
Y ), Ψ: Map
h
f
(X,
˜
Y ) → Map
h
f
(X, Y )
given by Φ(g, F) = (φ ◦ g, K · (φ ◦ F)) and Ψ(˜ g,
˜
F) = (ψ ◦ ˜ g, L · (ψ ◦
˜
F)) are
inverse homotopy equivalences. This restricts to a homotopy equivalence between
the subspaces consisting of homotopy equivalences. The analogous result holds
for homotopy equivalences from (X, f
X
) to (
˜
X, f
˜
X
).
Proof. We will show that Ψ◦ Φ ≃ Id. For Ψ◦ Φ is the map
(g, F) → (ψ ◦ φ ◦ g, L · [(ψ ◦ K) · (ψ ◦ φ ◦ F)]).
We have the following homotopies relative to endpoints:
L · [(ψ ◦ K) · (ψ ◦ φ ◦ F)] ≃ [L · (ψ ◦ K)] · (ψ ◦ φ ◦ F)
≃ (G◦ f
Y
) · (ψ ◦ φ ◦ F)
≃ F · (G◦ g ◦ f
X
).
The ﬁrst of these is a reparametrization, the second comes from our assumption
on the two homotopies φ ◦ ψ ◦ f
Y
⇒ f
Y
, and the third is obtained by using G
and F simultaneously. It follows that Ψ◦ Φ is homotopic to
(g, F) → (ψ ◦ φ ◦ g, F · (G◦ g ◦ f
X
)).
This is homotopic to the identity, as required. That Φ ◦ Ψ is homotopic to the
identity follows from the analogous argument, as does the second part of the
proposition.
10 The ﬁbrewise structure of RfD
◦
(n)
Let
∂ :
n
¸
i=0
S
1
×fD
◦
(n) → RfD
◦
(n)
denote the combined boundary map ∂
out
⊔
¸
n
i=1
∂
i
. In this section we will
study the ﬁbrewise properties of RfD
◦
(n) over fD
◦
(n), relative to ∂, using
the language of Section 9. To be precise, we will work in the setting of spaces
ﬁbred over fD
◦
(n) (and its open subsets), and in this setting we will consider
(RfD
◦
(n), ∂) as a space under
¸
n
i=0
S
1
× fD
◦
(n). We prove that each a ∈
fD
◦
(n) has a neighbourhood U over which (RfD
◦
(n), ∂) is isomorphic to (a ×
fD
◦
(n), ∂ ×Id). Moreover, we prove that (RfD
◦
(n), ∂) satisﬁes the hypothesis
of Proposition 9.8.
Proposition 10.1. 1. The combined boundary map ∂ extends to a ﬁbrewise
open embedding
¸
n
i=0
S
1
×[0, 1) ×fD
◦
(n) ֒→ RfD
◦
(n). In particular it is
a ﬁbrewise coﬁbration.
2. RfD
◦
(n) → fD
◦
(n) is a ﬁbre bundle. Moreover, local trivializations
RfD
◦
(n)U
∼
= U ×a can be chosen compatible with the boundary maps.
32
Proof. We will construct a ﬁbrewise open embedding
¸
n
i=0
S
1
×[0, 1) → RfD
◦
(n)
extending ∂. The ﬁrst claim follows.
Find a continuous function d: fD
◦
(n) → (0, 1) such that, for each a ∈
fD
◦
(n), the little discs of a can be dilated by a factor (1 + d(a)) yet still be
disjoint and lie within the interior of the disc of radius (1 −d(a)). Extend the i
th little disc of a to an aﬃne linear map a
i
: R
2
→R
2
. The required embedding
is then given in the ﬁbre over a by (θ, r) → [1−d(a)r]θ on the 0th cofactor and
by (θ, r) → a
i
([1 +d(a)r]θ) on the ith cofactor, for i > 0.
We now turn to the second claim, beginning with a special case. Fix a ∈
fD
◦
(1). Let φ: S
1
→ S
1
denote the framing of a. There is a homeomorphism
S
1
×[0, 1] → a given by (θ, r) → (1−r)∂
1
(θ) +rφ(θ). In a neighbourhood U of
a the map (θ, r) → (1 −r)∂
1
(θ) +rφ(θ)) is a homeomorphism S
1
×[0, 1] → b.
By combining the two homeomorphisms we obtain a trivialization of RfD
◦
(n)
over U.
Let a ∈ fD
◦
(n). Let A ⊂ fD
◦
(n) consist of those b for which each little
disc of b lies in the interior of the corresponding little disc of a. This is an open
subset of fD
◦
(n), and such open subsets cover fD
◦
(n). The required local
trivialization of RfD
◦
(n) near a can now be formed by restricting attention to
each little disc of a and applying the result for the case n = 1 given above.
11 The ﬁbrewise structure of RC(n)
The last section studied the ﬁbrewise properties of realizations of framed little
discs. In this section we turn to the analogous issue for cacti. Let
∂ :
n
¸
i=0
S
1
×C(n) → RC(n)
denote the combined boundary map ∂
out
⊔
¸
n
i=1
∂
i
. We will study the ﬁbrewise
properties of RC(n) over C(n), relative to ∂, using the language of Section 9.
To be precise, we will work in the setting of spaces ﬁbred over C(n) (and its
open subsets), and in this setting we will consider (RC(n), ∂) as a space under
¸
n
i=0
S
1
×C(n). Our result for cacti is much more involved than its counterpart
for framed little discs.
Proposition 11.1. Fix c ∈ C(n) and form the ﬁbrewise spaces (RC(n), ∂) and
(c ×C(n), ∂ ×Id) under
¸
n
i=0
S
1
×C(n). Then there is an open neighbourhood
U of c over which (RC(n), ∂) and (c ×C(n), ∂ ×Id) are homotopy equivalent.
Corollary 11.2. RC(n) → C(n) is a quasiﬁbration.
Proof. In Proposition 11.1 the local existence of the maps φ and ψ, and the
homotopies G and H, show that RC(n) → C(n) is locally ﬁbrewise homotopy
equivalent to a product. It is therefore locally a quasiﬁbration. Since maps that
are locally quasiﬁbrations are themselves quasiﬁbrations, the claim follows.
11.1 Proof of Proposition 11.1
Fix c ∈ C(n). For each r ∈ {1, . . . , n} choose closed intervals I
r
⊂
¯
I
r
⊂ S
1
such that I
r
lies in the interior of
¯
I
r
, such that ∂
r
(
¯
I
r
) does not meet the image
33
of any incoming boundary maps besides ∂
r
, and such that ∂
out
(•) does not lie
in any
¯
I
r
. Since the lobe coordinates of a cactus depend continuously on the
cactus, we may choose a neighbourhood U of c over which the above properties
of I
r
⊂
¯
I
r
still hold. Set P =
¸
r
∂
r
(S
1
\ int(I
r
)) ⊂ RC(n)U.
Thus P is a ﬁbrewise subspace of RC(n)U whose ﬁbre over d is a copy of
d with an interval removed from each lobe; in other words, each ﬁbre is a tree.
The ﬁbrewise quotient RC(n)/P is given in each ﬁbre by the bouquet of circles
(
¸
I
r
)/(
¸
∂I
r
). It is therefore natural to expect that RC(n)U → (RC(n)U)/P
is a ﬁbrewise homotopy equivalence with the product (
¸
I
r
)/(
¸
∂I
r
) ×U. Our
proof of Proposition 11.1 will proceed along these lines.
Form the ﬁbrewise spaces ((RC(n)U)/P, ∂
1
) and ((c/P
c
) ×U, ∂
2
) under
¸
n
i=0
S
1
×U, where ∂
1
and ∂
2
are the composites
∂
1
:
¸
n
i=0
S
1
×U
∂
−→ RC(n)U → (RC(n)U)/P,
∂
2
:
¸
n
i=0
S
1
×U
∂×Id
−−−→ c ×U → (c/P
c
) ×U.
We will construct three homotopy equivalences,
(RC(n)U, ∂) ≃ ((RC(n)U)/P, ∂
1
) ≃ ((c/P
c
) ×U, ∂
2
) ≃ (c ×U, ∂) . (9)
Since homotopy equivalence is a transitive relation, this will suﬃce to prove the
proposition.
There are ﬁbrewise pushout squares
¸
r
U ×∂I
r
P
¸
r
U ×I
r
RC(n)U,
¸
r
U ×∂I
r
P
c
×U
¸
r
U ×I
r
c ×U.
(10)
In both cases the horizontal maps are the restrictions of
¸
n
r=1
∂
r
and the right
hand map is the inclusion. The left hand maps of these squares are coﬁbrations,
so the same is true for the right hand maps. We make the following claim:
1. P is ﬁbrewise contractible.
From the claim and from Lemma 9.6 we obtain the ﬁrst and last homotopy
equivalence of (9). It remains to ﬁnd the middle homotopy equivalence. The
diagrams (10) also show that the ﬁbrewise spaces (RC(n)U)/P and (c/P
c
)×U
are ﬁbrewise isomorphic, and moreover the isomorphism is such that the triangle
¸
n
i=1
S
1
×U
∂2
P
P
P
P
P
P
P
P
P
P
P
∂1
n
n
n
n
n
n
n
n
n
n
n
n
(RC(n)U)/P
∼
=
(c/P
c
) ×U
commutes. The triangle says nothing about the value of ∂
1
and ∂
2
on the ﬁrst
cofactor of
¸
n
i=0
S
1
×U. Now we make a further claim.
2. The composites
S
1
×U ֒→
¸
n
i=0
S
1
×U
∂1
−→(RC(n)U)/P
∼
= (c/P
c
) ×U,
S
1
×U ֒→
¸
n
i=0
S
1
×U
∂2
−→(c/P
c
) ×U
are homotopic; here the ﬁrst map is the inclusion of the 0th cofactor.
34
From this claim and from the commutative triangle it follows that ((RC(n)U)/P, ∂
1
)
is isomorphic to ((c/P
c
) ×U, ∂
′
2
), where ∂
′
2
is some map homotopic to ∂
2
, so
that ((c/P
c
) ×U, ∂
′
2
) is itself homotopy equivalent to ((c/P
c
) ×U, ∂
2
). Thus
((RC(n)U)/P, ∂
1
) is homotopy equivalent to ((c/P
c
) ×U, ∂
2
), as required.
It remains to prove our claims. The second claim states that two maps
S
1
× U → (c/P
c
) × U are homotopic. By construction, both are families
(parameterized by U) of based maps from S
1
to a bouquet of circles, lying in
the same homotopy class. The claim follows since the space of based maps
S
1
→
S
1
in a ﬁxed homotopy class is contractible.
The ﬁrst claim states that P is ﬁbrewise contractible. Note that the ﬁbres
of P are connected. For, given d ∈ U, any two points of d can be joined by
a path that does not pass through any of the I
r
. Let us assume without loss
that each I
r
⊂ S
1
is equal to [1/2, 1] ⊂ S
1
. Then the lobe coordinates are maps
U → [0, 1/2]
n−1
and the ∂
r
I
r
are maps [0, 1/2] → [0, 1/2]
n
. Rescaling [0, 1/2]
to [0, 1] we see that the ﬁbred space P → U is a conﬁguration of orthogonal
line segments in the sense described in the next section. The second claim then
follows from Proposition 11.3 below.
11.2 Conﬁgurations of orthogonal line segments
Let A be a topological space. Suppose we are given functions x
i
: A → R
n−1
for i = 1, . . . , n. Write these as x
i
= (x
i
1
, . . . , x
i
i−1
, x
i
i+1
, . . . , x
i
n
). We obtain
functions ψ
i
: A×I →R
n
given by
(a, t) → (x
i
1
(a), . . . , x
i
i−1
(a), t, x
i
i+1
(a), . . . , x
i
n
(a)).
Assume that for each a ∈ A the space
¸
i
ψ
i
({a} × I) is connected. Deﬁne
Q → A to be the ﬁbrewise space Q =
¸
i
ψ
i
(A×I) ⊂ A×R
n
. We refer to Q as
a conﬁguration of orthogonal line segments in R
n
.
Proposition 11.3. Q is ﬁbrewise contractible.
Proof. Since Q obviously admits sections, it will suﬃce to show that it is ﬁbre
wise convex, or in other words that there is a ﬁbrewise h: Q×
A
Q×[0, 1] → Q
satisfying h((p, q), 0) = p and h((p, q), 1) = q.
For the beginning of this proof let us assume that A is a single point and
omit it from the notation.
Note that the ith coordinates of the points where ψ
i
[0, 1] meets other ψ
j
[0, 1]
are exactly the x
j
i
.
We say that i ∈ {1, . . . , n} is a leaf if ψ
i
[0, 1] meets exactly one other ψ
j
[0, 1].
Leaves exist so long as n 2. This is clear when n = 2, since Q is connected.
We prove the general case by induction. Let π: R
n−1
→R
n−2
be the map that
forgets the ﬁnal coordinate. Then πx
1
, . . . , πx
n−1
deﬁne a conﬁguration of line
segments in R
n−1
, and so have a leaf which without loss is (n − 1). Then the
x
1
n−1
, . . . , x
n−2
n−1
coincide. So if (n − 1) is not a leaf then x
n
n−1
must diﬀer from
the common value of the x
i
n−1
, and it follows that n is a leaf. In either case, a
leaf exists.
An adapted path γ : [0, 1] → Q is one for which there is v 0 and a decom
position of [0, 1] into intervals [α, β] on which γ has the form t → (x
i
1
, . . . , t
0
+
ǫ
i
vt, . . . , x
i
n
) for some value of i, some t
0
, and some ǫ
i
∈ {±1} depending only
on i.
35
Any two points x, y of Q can be joined by a unique adapted path. This is
clear when n = 1 and n = 2, and in general is proved by induction. For we
can assume that n is a leaf. The claim is then immediate if x and y both lie in
ψ
n
[0, 1], and follows by induction if they both lie in
¸
in−1
ψ
i
[0, 1]. In the ﬁnal
case assume without loss that x lies in ψ
n
[0, 1] and that y lies in
¸
in−1
ψ
i
[0, 1].
Write x
′
for the point where ψ
n
[0, 1] meets
¸
in−1
ψ
i
[0, 1]. Then an adapted
path from x
′
to y exists by induction, and can easily be modiﬁed to produce an
adapted path from x to y. Uniqueness is proved by a similar induction.
Now remove the restriction that A is a single point. By applying the results
just obtained in each ﬁbre, we obtain a function h: Q ×
A
Q × [0, 1] → Q de
termined by the fact that each h(x, y, −) is the adapted path from x to y. We
must show that it is continuous. For ﬁxed i ∈ {1, . . . , n} the subset A
i
of A on
which i is a leaf is closed, so it suﬃces to prove continuity over A
i
. Continuity
over A
i
follows from an argument similar to the construction of adapted paths.
This completes the proof.
12 Proof of Theorem C
We can now complete the proof of Theorem C. We continue to write ∂ for the
combined boundary map ∂
out
⊔
¸
n
i=1
∂
i
.
Deﬁnition 12.1. Let E
◦
h
(n) denote the space of quadruples (a, c, f, H) where
a ∈ fD
◦
(n), c ∈ C(n), f : a → c is a homotopy equivalence, and H:
¸
n
i=0
S
1
×
[0, 1] → c is a homotopy from f ◦ ∂ to ∂. We topologize E
◦
h
(n) as a subspace of
the ﬁbrewise mapping space. Taking the constant homotopy gives an inclusion
E
◦
(n) ֒→ E
◦
h
(n).
Fix a ∈ fD(n) and c ∈ C(n). Recall that Map
∂
(a, c) denotes the ﬁbre
of E(n) over (a, c). In other words, it is the space of homotopy equivalences
f : a → c satisfying f ◦ ∂ = ∂.
Deﬁnition 12.2. Given a ∈ fD
◦
(n) and c ∈ C(n), let Map
h
∂
(a, c) denote the
ﬁbre of E
◦
(n) over (a, c). In other words, it is the space of pairs (f, H) with
f : a → c a homotopy equivalence and H: [0, 1] ×
¸
n
i=0
S
1
→ c a homotopy
from f ◦ ∂ to ∂. It is equipped with the compact open topology. Taking the
constant homotopy gives an inclusion Map
∂
(a, c) ֒→ Map
h
∂
(a, c).
The proof of Theorem A is completed by the following three propositions.
Proposition 12.3. The inclusion E
◦
(n) ֒→ E
◦
h
(n) is a ﬁbrewise homotopy
equivalence of spaces over fD
◦
(n) ×C(n).
Proposition 12.4. Every pair (a, c) ∈ fD
◦
(n) × C(n) has a neighbourhood W
over which there is a ﬁbrewise homotopy equivalence
E
◦
h
(n) ≃ Map
h
∂
(a, c) ×W.
Proposition 12.5. The inclusion Map
∂
(a, c) ֒→ Map
h
∂
(a, c) is a homotopy
equivalence.
Proof of Proposition 12.3. This is an application of the ﬁbrewise version of
Proposition 9.8; the assumption holds by the ﬁrst part of Proposition 10.1.
36
Proposition 12.5 is also proved by an application of Proposition 9.8.
Proof of Proposition 12.4. Let U be a neighbourhood of a over which (RfD
◦
(n), ∂)
and (a ×fD
◦
(n), ∂ ×Id) are homotopy equivalent as in Proposition 10.1, and
let V be a neighbourhood of c over which (RC(n), ∂) and (c ×C(n), ∂ ×Id) are
homotopy equivalent, as in Proposition 11.1. Set W = U ×V .
Write E
′
for the space of quadruples ((b, d), f, H), where (b, d) ∈ W, f : a →
c is a homotopy equivalence, and H: f ◦ ∂ ⇒ ∂ is a homotopy. Topologize E
′
as a subspace of the ﬁbrewise mapping space. Proposition 9.9, together with
the previous paragraph, implies that E
◦
h
(n)W is ﬁbre homotopy equivalent to
E
′
. Part 8 of Proposition A.8 shows that E
′
is ﬁbrewise homeomorphic to the
space given in the statement. This completes the proof.
A Fibrewise topology
This appendix will recall some basic notions of ﬁbrewise topology, and establish
some facts regarding the ﬁbrewise mapping space. We refer almost entirely to
the book of James [6].
Deﬁnition A.1. Let B be a space. Then a ﬁbred space or ﬁbrewise space over
B is a space X with a map p: X → B called the projection. The ﬁbre of X over
b ∈ B is X
b
= p
−1
(b). A ﬁbrewise map φ: X → Y between spaces ﬁbred over
B is a map X → Y that commutes with the projections.
Deﬁnition A.2. Given a map f : A → B and a space X ﬁbred over B, the
pullback f
∗
X is the ﬁbred space f
∗
X → A given by X ×
p,f
A with its natural
projection. When A ֒→ B is the inclusion of a subspace we will write the
pullback as X
A
or X
A
.
Deﬁnition A.3. Given X, Y ﬁbred over B, then X ⊔ Y and X ×
B
Y become
spaces ﬁbred over B in the obvious way. The ﬁbrewise pushout of a diagram
Z ← X → Y of spaces and maps ﬁbred over B is the ordinary pushout Y ∪
X
Z,
regarded as a space ﬁbred over B. A square
X
Y
Z
W
of spaces and maps ﬁbred over B is a ﬁbrewise pushout square if the induced
ﬁbrewise Y ∪
X
Z → W is a homeomorphism.
Deﬁnition A.4. The ﬁbrewise space X is ﬁbrewise compact if p: X → B is
proper, or in other words if p is closed and has compact ﬁbres. It is ﬁbrewise
Hausdorﬀ if distinct points in the same ﬁbre can be separated by open sets of
X; this is always the case if X itself is Hausdorﬀ.
Proposition A.5 ([6, 3.7]). Let X and Y be ﬁbrewise compact Hausdorﬀ spaces.
A continuous bijection φ: X → Y is in fact a homeomorphism. If φ: X → Y
is a continuous ﬁbrewise surjection, then a ﬁbrewise function ψ: Y → Z is
continuous if and only if the composite ψ ◦ φ is continuous.
37
Corollary A.6. Suppose given a commutative square
X
Y
Z
W
of ﬁbrewise compact Hausdorﬀ spaces ﬁbred over B, and suppose that in each
ﬁbre the square is a pushout. Then the square is a ﬁbrewise pushout. Moreover,
a ﬁbrewise function W → V is continuous if and only if the composites Y → V
and Z → V are continuous.
Deﬁnition A.7 ([6, p.63]). Let X and Y be spaces ﬁbred over B. The ﬁbre
wise mapping space Map
B
(X, Y ) is the space of pairs (b, f), where b ∈ B and
f : X
b
→ Y
b
is continuous. It is equipped with the topology generated by basic
open sets (K, V ; W). Here W is an open subset of B, K is a ﬁbrewise compact
subspace of X
W
, and V is an open subspace of Y
W
. Then (K, V ; W) consists
of all pairs (b, f) for which b ∈ W and f(K
b
) ⊂ V
b
.
Proposition A.8 ([6]). Let B be a topological space and let X, Y, Z and X
1
, X
2
be ﬁbrewise compact Hausdorﬀ spaces over B.
1. Let θ : X → Y and φ: Y → Z be ﬁbrewise maps. Then there are ﬁbrewise
maps
φ
∗
: Map
B
(X, Y ) → Map
B
(X, Z), θ
∗
: Map
B
(Y, Z) → Map
B
(X, Z).
2. If θ above is a ﬁbrewise surjection then θ
∗
is an embedding.
3. The bijection
Map
B
(X
1
⊔ X
2
, Y ) → Map
B
(X
1
, Y ) ×
B
Map
B
(X
2
, Y )
is a ﬁbrewise homeomorphism.
4. Take a map F : A → B. Then there is a continuous map
Map
A
(F
∗
X, F
∗
Y ) → Map
B
(X, Y )
sending (a, f) to (F(a), f).
5. A ﬁbrewise map
h: X ×
B
Y → Z
is continuous if and only if its adjoint
ˆ
h: X → Map
B
(Y, Z)
is continuous.
6. Let C be a further space and regard X × C, Y × C as ﬁbred over B × C.
Then the map
Map
B
(X, Y ) ×C → Map
B×C
(X ×C, Y ×C)
sending ((b, f), c) to ((b, c), f) is continuous.
38
7. Let P and Q be topological spaces, with P compact Hausdorﬀ. Then the
isomorphism of sets
Map
B
(B ×P, B ×Q)
∼
= B ×Map(P, Q)
is a homeomorphism.
Proof. All references in this proof are to [6]. Note that 3.22 and the comments
after 3.12 guarantee that a ﬁbrewise compact ﬁbrewise Hausdorﬀ space is also
ﬁbrewise locally compact and ﬁbrewise regular. Then parts 1, 2, 3, 5 and 6
are found in p.69, 9.4, 9.6, 9.7 and 9.13 respectively. Part 4 is an immediate
consequence of the deﬁnitions.
For part 6 take a standard open set (K, V ; W) in Map
B×C
(X × C, Y × C)
and an element ((b, f), c) of its preimage. We will ﬁnd a neighbourhood of
((b, f), c) whose image is contained in (K, V ; W). To do this we will reduce
(K, V ; W) many times. By reducing W and V , assume that W = W
1
×W
2
and
V = V
1
×W
2
; this can be done because K
(b,c)
is compact. Now using 3.14 and
reducing W
1
if needed, ﬁnd an open subset J of X
W1
such that K
(b,c)
⊂ J and
J
b
⊂ f
−1
(V
1
), where J is ﬁbrewise compact over W
1
. Thus K
W1×W2
\J ×W
2
is
ﬁbrewise compact over W
1
×W
2
and in particular its image in W
1
×W
2
is closed.
But this image does not contain (b, c), and so by reducing W
1
and W
2
we may
assume that K
W1×W2
⊂ J ×W
2
⊂ J ×W
2
. Now the open set (J, V
1
; W
1
) ×W
2
contains ((b, f), c) and its image is contained in (K, V ; W) as required.
For the ﬁnal part, continuity of f : Map
B
(B ×P, B ×Q) → B ×Map(P, Q)
follows from part 6, and continuity of its inverse follows from part 7.
References
[1] T. Bargheer. Cultivating operads in string topology. Master’s Thesis,
University of Copenhagen, 2008.
[2] Moira Chas and Dennis Sullivan. String topology, 1999. Preprint, available
from arXiv:9911159.
[3] Ralph L. Cohen and Alexander A. Voronov. Notes on string topology. In
String topology and cyclic homology, Adv. Courses Math. CRM Barcelona,
pages 1–95. Birkh¨auser, Basel, 2006.
[4] Z. Fiedorowicz. Constructions of E
n
operads. Preprint, available from
arXiv:9808089, 1998.
[5] E. Getzler. BatalinVilkovisky algebras and twodimensional topological
ﬁeld theories. Comm. Math. Phys., 159(2):265–285, 1994.
[6] I. M. James. Fibrewise topology, volume 91 of Cambridge Tracts in Math
ematics. Cambridge University Press, Cambridge, 1989.
[7] Christian Kassel and Vladimir Turaev. Braid groups, volume 247 of Grad
uate Texts in Mathematics. Springer, New York, 2008. With the graphical
assistance of Olivier Dodane.
[8] Ralph M. Kaufmann. On several varieties of cacti and their relations.
Algebr. Geom. Topol., 5:237–300 (electronic), 2005.
39
[9] J. P. May. Weak equivalences and quasiﬁbrations. In Groups of self
equivalences and related topics (Montreal, PQ, 1988), volume 1425 of Lec
ture Notes in Math., pages 91–101. Springer, Berlin, 1990.
[10] James E. McClure and Jeﬀrey H. Smith. Cosimplicial objects and little
ncubes. I. Amer. J. Math., 126(5):1109–1153, 2004.
[11] Luc Menichi. A BatalinVilkovisky algebra morphism from double loop
spaces to free loops. To appear in Trans. Amer. Math. Soc., see also
arXiv:0908.1883, 2009.
[12] Paolo Salvatore. The topological cyclic Deligne conjecture. Algebr. Geom.
Topol., 9(1):237–264, 2009.
[13] Alexander A. Voronov. Notes on universal algebra. In Graphs and patterns
in mathematics and theoretical physics, volume 73 of Proc. Sympos. Pure
Math., pages 81–103. Amer. Math. Soc., Providence, RI, 2005.
40
The operad E is deﬁned as follows. Given elements a ∈ f D(n) and c ∈ C(n), write a for the space obtained from D2 by deleting the interiors of the little discs of a, and write c for the conﬁguration of circles underlying c. Then E(n) is the space of triples (a, c, f ), where a ∈ f D(n), c ∈ C(n), and where f : a → c is a homotopy equivalence satisfying the following two boundary conditions. First, the boundary of the ith little disc is sent into c by the inclusion of the ith lobe. Second, the boundary of the big disc is sent into c by the pinch map. A typical element (a, c, f ) ∈ E(2) is depicted below.
1 1 2
2
a There are forgetful maps
c
f
− −→ − −− f D(n) ← −1 − E(n) − −2− C(n). The ﬁrst part of our theorem can now be restated as follows. Theorem A. The spaces E = {E(n)} form a topological operad, and π1 and π2 are weak equivalences of operads. In particular, we recover the fact that cacti and framed little discs are weakly equivalent. This was stated by Voronov [13] and ﬁrst proved by Kaufmann [8]. A diﬀerent proof has since been given by Bargheer [1]. In those accounts, as in Theorem A, one constructs some intermediate operad O equipped with weak equivalences to both f D and C. In Kaufmann’s proof O is obtained using a recognition principle of Fiedorowicz [4], while in Bargheer’s proof one takes O = Wf D, the BoardmanVogt Wconstruction on f D. The advantage of Theorem A is that E is constructed using only the geometry of the original operads f D and C. Further, it allows us to construct a natural Ealgebra εG weakly equivalent to both ΩG and Ω2 BG. Note, however, that Theorem A still relies on Kaufmann’s computation of the weak homotopy type of the C(n). Now we describe the Ealgebra εG. Let EG → BG be the universal principal Gbundle. Then εG is the space of based maps D2 → EG that send S 1 into the orbit of the basepoint. There are maps Ω2 BG ← − − εG − − − ΩG − −− − −→ given respectively by projecting from EG to BG and by restricting from D2 to ≃ S 1 . These are compatible with h : ΩG − Ω2 BG in the sense that h ◦ p2 ≃ p1 . → We can now restate the second part of our theorem as follows. Theorem B. The space εG is an Ealgebra, and the maps p1 and p2 are weak equivalences of Ealgebras. We recover a result of Menichi [11] which states that h∗ : H∗ (ΩG) → H∗ (Ω2 BG) is an isomorphism of BValgebras. (Menichi’s result holds in the more general case that G is a grouplike topological monoid.) 2
p1 p2
π
π
The key ingredient in our paper is the following common feature of the operads f D and C. To elements a ∈ f D(n) and c ∈ C(n) we can associate the spaces a and c deﬁned above, which we call the realizations of a and c. The realizations come together with incoming boundary maps ∂1 , . . . , ∂n : S 1 → a, and outgoing boundary maps ∂out : S 1 → a, ∂out : S 1 → c. ∂1 , . . . , ∂n : S 1 → c
Together, these realizations and boundary maps form what we call realization systems for f D and C. This means that they are compatible with the symmetry maps, composition maps, and topology of the original operads f D and C. Most importantly, the realization x ◦i y of the composite of x and y is the pushout of ∂i : S 1 → x and ∂out : S 1 → y. This fact is easily veriﬁed for both f D and C, and leads directly to the deﬁnition of composition in E. The paper contains a detailed development of realization systems. The crucial result takes two operads P and Q with realization systems and constructs a mapping operad M. Points of M(n) are triples (p, q, f ) where p ∈ P(n), q ∈ Q(n) and f : p → q is a map satisfying f ◦ ∂i = ∂i and f ◦ ∂out = ∂out . The operad E is obtained in the case P = f D, Q = C by restricting to the suboperad of M consisting of triples (p, q, f ) in which f is a homotopy equivalence. The three algebras studied in the paper, namely the f Dalgebra Ω2 BG, the Calgebra ΩG, and the Ealgebra εG, are also described in terms of the realization systems for f D and C. The paper begins in Section 2 with a discussion of realization systems, and proves that from a pair of operads with realization systems, one obtains a mapping operad M. Sections 3 and 4 recall framed little discs and cacti, and describe their realization systems. Then Section 5 applies the mapping operad construction to P = f D and Q = C to obtain E. Section 6 uses the realization systems to describe the actions of f D on Ω2 BG, of C on ΩG, and of E on εG. With these ingredients in place Theorem B follows without diﬃculty. It remains to complete the proof of Theorem A by showing that π1 and π2 are weak equivalences. The structure of the proof is explained in detail in Section 7. It is shown there that Theorem A follows from the computation of a certain long exact sequence, which is carried out in Section 8, and from a further result, Theorem C. Theorem C states roughly that the projection π1 × π2 : E(n) → f D(n)× C(n) is a quasiﬁbration. It relies on some basic results on mapping spaces which are given in Section 9, and on a study of the ﬁbrewise structure of the realizations of f D and C, which is carried out in Sections 10 and 11. The proof of Theorem C is completed in Section 12. An appendix recalls some basic results of ﬁbrewise topology.
2
Realization systems for operads
In this section we introduce realization systems for an operad O. A realization system with boundaries X is a rule that assigns to each element x ∈ O(n) a space 3
x, the realization of x, equipped with (n + 1) diﬀerent maps X → x. The rule carries further structure and is subject to a number of axioms, all of which roughly amount to saying that the spaces x are compatible with symmetry and composition in O, and vary continuously as x varies within O(n). This is made precise in §2.1, and two simple examples are given in §2.2. Then in §2.3 we deﬁne the mapping operad. Given operads P and Q, both equipped with realizations, there is an associated mapping operad. It lies over both P and Q, and its elements are triples consisting of points p ∈ P(n), q ∈ Q(n), and a map p → q satisfying boundary conditions. The operad E of Theorem A will be constructed as a suboperad of such a mapping operad. For us, all operads are Σoperads with a 0th space, which we do not insist be equal to a single point. Composition within operads will be denoted using either the symbol γ or the symbol ◦i , for example γ(x; y1 , . . . , yn ) or x ◦i y.
2.1
Realization systems
Fix a compact, Hausdorﬀ topological space X and a topological operad O. The following three deﬁnitions together deﬁne the notion of a realization system. The deﬁnitions are somewhat abstract, and so the reader may ﬁnd it helpful to bear in mind the example of framed little discs. In this case, the realization a of a ∈ f D(n) is the complement of the interiors of the little discs; the incoming boundary maps S 1 → a are the boundaries of the little discs; and the outgoing boundary map S 1 → a is the boundary of the big disc. The ﬁrst deﬁnition states what data must be given in order to deﬁne a realization system. It is analogous to deﬁning an operad by specifying spaces, symmetry maps, and composition maps. Deﬁnition 2.1. A realization system for O with boundaries X is a rule RO that for each n 0 and each x ∈ O(n) produces a topological space x called the realization of x, and that furthermore equips these spaces with the following structure: • Boundary Maps. For each x ∈ O(n), incoming boundary maps ∂1 , . . . , ∂n : X → x and an outgoing boundary map ∂out : X → x. • Symmetries. For each x ∈ O(n) and σ ∈ Σn , a symmetry map σ ∗ : x → xσ. • Pasting and composition. Given x ∈ O(n) and yi ∈ O(mi ) for i = 1, . . . , n, a pushout diagram
n i=1 ∂i
X
∂out
/ y1  ⊔ · · · ⊔ yn 
(1)
x called the pasting square. 4
/ γ(x; y1 , . . . , yn )
. . In what follows we will use underlines to indicate tuples of elements. the boundary maps ∂1 . . It is analogous to the axioms satisﬁed by the various composition and symmetry maps in an operad. 4.2. ∂out / y x 2. σ ∗ ◦ ∂out = ∂out . 3. . they are almost trivial to verify in the example of framed little discs. . yn σn . A realization system RO must satisfy the axioms listed below. y) < xx xx x xx xx ∂out x xx yi  o ∂j X vv vv vv vv ∂ i. For example. . . . and for x and y1 σ1 .2. and to be topologized in the sense of Deﬁnition 2.3. Pasting and boundaries: The lower and righthand maps of diagram (1) are compatible with the boundary maps. . Deﬁnition 2. 1. Applying these homeomorphisms to (1) in the case n = 1 and x = 1O . both below. . y) The unions run over i = 1. . . . mi . . . the symbol ∂i. and to (1) in the case yi = 1O . . . yn . . .j denotes ∂m1 +···+mi−1 +j . y will denote y1 . yn . n and j = 1. we can formulate diagram (1) for x and y1 . Unit: Taking x = 1O .j vv vv zv γ(x.The system RO is required to satisfy the axioms of Deﬁnition 2. Pasting and symmetries I: Given σ1 ∈ Σm1 . Symmetries: The symmetry maps satisfy σ ∗ ◦ τ ∗ = (τ σ)∗ and are compatible with the boundary maps in the sense that σ ∗ ◦ ∂i = ∂σ−1 i . . ∂out : X → 1O  are homeomorphisms and coincide. x O ∂out X / γ(x. Our second deﬁnition lists the axioms that the structures above must satisfy. in the sense that the following triangles commute. . σn ∈ Σmn . . Although we list no fewer than six axioms. . Then the resulting squares are isomorphic under the relevant symmetry 5 . . one obtains the diagrams X ∂out / y and ∂i X x X ∂i X respectively.
mn )∗ O' / γ(xσ. yσ−1 . z). . . Then we can form composites γ(x. yn . . y). . γ(x. By considering the pasting diagram (1) for each composite. . . γ(yn . z1 . . Thus we have elements x ∈ O(n).map. Then the two 1 squares are isomorphic under the relevant symmetry maps. zn n  into the single space above. y1 . . . . we obtain two 1 m maps from each of x. yσn and σ⊔ sends the σi th cofactor to the ith 1 cofactor. z) = γ(x. . . Pasting and symmetries II: Similarly. γ(y1 . y). 6. and the appendix recalls what it means for such a ﬁbred space to be ﬁbrewise compact or ﬁbrewise Hausdorﬀ. yi ∈ O(mi ) and zi ∈ j O(li ).. . yn σn . y) ∗ (σ1 ⊕···⊕σn ) Here yσ = y1 σ1 .. and we have recalled the necessary ideas in Appendix A. The last two composites coincide. γ(y1 . . . γ(γ(x. yσ) γ(x. . . The two maps in each such pair coincide. . γ(γ(x. mi . which means that the following diagram commutes. for i = 1. . . The standard reference for this is [6]. . . yn . X EE EE E σ⊔ E" / X yσ−1  i OOOσ⊔ OOO OO' / yi  x E EE EE σ∗ E" xσ / γ(x. given σ ∈ Σn we can formulate −1 diagram (1) for x. . . . . . zn )). . Associativity of pasting: Suppose we are given the data for an iterated j composition in O. yσn . 6 . The next deﬁnition says what it means to topologize a realization system. . . . zn )). and consequently so do their realizations. . y). . . and for xσ. . 5. . z1 ). γ(yn . . and makes use of ﬁbrewise topology.. y1 . . y) −1 Here yσ−1 = yσ−1 . n and j = 1. zi ). which means that the following diagram commutes. γ(yi .. z1 ). yσ−1 ) OOO OOO σ(m1 . . X yi  ttt yttt KK KK KK % / yi σi  ∗ ∗ σ1 ⊔···⊔σn x K KK KKK ttt K% ttt zt / γ(x. A space ﬁbred over B is just a space X equipped with a map X → B. .
(The spaces written in (2) should all be regarded as spaces over O(n) × O(m1 ) × · · · × O(mn ) by pulling back along the appropriate constant. the pasting square (1) is given by p i=1 Sn collyi / p i=1 qi j=1 Sn p i=1 Sn collyi / p i=1 qi j=1 Sn where the vertical maps are the standard quotients. which are the realizations for f D and C. The outgoing boundary map ∂out : S n → S n is the collapse map collx induced by x. . . and the pasting squares (1) become squares n i=1 ∂i X ∂out / RO(m1 ) ⊔ · · · ⊔ RO(mn ) (2) RO(n) / RO(m1 + · · · + mn ) of spaces over O(n) × O(m1 ) × · · · × O(mn ).4 (The little cubes operad). • The incoming boundary maps ∂1 . ∂out : O(n) × X → RO(n). . Let Cn denote the little ncubes operad. Example 2. All of the functions above are required to be continuous. . σ ∗ : RO(n) → σ ∗ RO(n) between spaces over O(n). are presented in Sections 3 and 4 respectively.Deﬁnition 2. Given p x ∈ Cn (p).3 (Topology on a realization system). n Each RO(n) must be ﬁbrewise compact Hausdorﬀ over O(n). ∂n . . A topology on a realization system RO consists of ﬁbred spaces ρn : RO(n) → O(n) for n 0. . set x = i=1 S n . . projection or composition map. such that for each x ∈ O(n) the ﬁbre ρ−1 (x) is identiﬁed with x. • The symmetry map σ ∗ : x → xσ is the map −1 the ith cofactor to the (σi )th cofactor.) 2. ∂p : S n → S n are the standard insertions. 7 . Sn → S n that sends • Given x ∈ Cn (p) and yi ∈ Cn (qi ). and the squares (2) are required to be ﬁbrewise pushouts.2 Examples of realizations This subsection attempts to illustrate the deﬁnition of realization systems by giving two simple examples in the case when O is the operad of little ncubes. Deﬁne a realization system R1 Cn with boundaries S n as follows. The paper’s two main examples. The boundary and symmetry maps determine functions ∂1 . .
The conditions of Deﬁnition 2. f ). c. di . (b1 . Composition in M is deﬁned by pasting functions in the following way. . . .) 2. the composite γ ((a. both with boundaries X.6. f ). This is a pushout diagram. The axioms of Deﬁnition 2. σ ∗ ◦ f ◦ σ ∗ −1 ). . . cσ. This subsection introduces the mapping operad M associated to this data.The axioms of Deﬁnition 2. .2 are simple to verify. . (bn .2 hold trivially. dn . gi ) ∈ M(mi ) for i = 1. f )σ = (aσ. Fix n 0. . Let Cn again denote the little ncubes operad. g1 ). c. yn ) by applying xi : I n → I n . The system is topologized by setting R1 Cn (p) = Cn (p) × S n . ∂p : S n−1 → x are the inclusions of the boundaries of the little cubes of x. . .5 (Little cubes operads again).8. y1 . The righthand map sends yi  into γ(x. . c. Deﬁnition 2. y1 . the ith little cube of x. . • Given x ∈ Cn (p) and yi ∈ Cn (qi ). .2. Example 2.3 are immediately veriﬁed.7. . . Given (a. (Compare with the system Rf D in Section 3. c ∈ Q(n) and f : a → c is a continuous map satisfying the boundary conditions f ◦ ∂i = ∂i and f ◦ ∂out = ∂out . the pasting square (1) n i=1 ∂i S n−1 ∂out / y1  ⊔ · · · ⊔ yn  x / γ(x. . . c. . . which we regard as the boundary of the ncube I n . The outgoing boundary map ∂out : S n−1 → x is the inclusion of the boundary of the big cube. gn )) 8 . Deﬁnition 2. • The incoming boundary maps ∂1 . yn ) given by regarding both as subsets of the big cube I n . The system is topologized by deﬁning R2 Cn (p) to be the subspace of Cn (p) × I n whose ﬁbre over x ∈ Cn (p) is precisely x ⊂ I n . . Deﬁnition 2. n. yn ) is given as follows. where σ ∗ : a → aσ and σ ∗ : c → cσ are the symmetry maps. Deﬁne the realization system R2 Cn with boundaries S n−1 . y1 . deﬁne x to be the complement in the big cube I n of the interiors of the little cubes of x.3 follow. and the conditions of Deﬁnition 2. Then M(n) denotes the set of triples (a. where a ∈ P(n). Given x ∈ Cn (p). d1 . .3 The mapping operad Suppose now that we are given operads P and Q equipped with realization systems RP and RQ. • The symmetry map σ ∗ : x → xσ identiﬁes the two spaces as subsets of the big cube. . f ) ∈ M(n) and (bi . The bottom map is the inclusion of x into γ(x. . as follows. The action of Σn on M(n) is deﬁned by the formula (a. .
. γ(c. . . bn ). Theorem 2. 9 . . f ) where a ∈ P(n). and we call the resulting space the ﬁbrewise mapping space.. d1 . This set can be equipped with the ﬁbrewise compactopen topology. . c. . b) MM MM & / γ(c. Given spaces U and V ﬁbred over B.. . V ) for the set of pairs (b. . Then the spaces M≃ (n) = {(a. and the projections P ←− M −→ Q are morphisms of operads.10. Since homotopy equivalences are preserved by homotopy pushouts. . b1 . it follows that if f and g1 . and that M≃ is closed under the Σn action. To topologize M(n) we use another notion from ﬁbrewise topology. The front and rear faces of this cube are the pushout squares (1) for c.11. The realization systems RP and RQ give us ﬁbred spaces RP(n) → P(n) and RQ(n) → Q(n). . . π2 RQ(n)) consists of triples (a. while the left hand and upper faces come from the compatibility between pasting and boundary maps. The space M(n) is topologized as a subspace of the ﬁbrewise ∗ ∗ mapping space MapP(n)×Q(n) (π1 RP(n). γ(f .11. Proposition 2. . . Write π1 and π2 for the projections of P(n) × Q(n) onto its ∗ ∗ factors.gn ) / γ(a. . where b ∈ B is a point of the base and f : Ub → Vb is a map between the ﬁbres over that point.is the new triple (γ(a. f ). g1 . where γ(f . See Appendix A and the reference there. d1 . rather than just pushouts. . . c ∈ Q(n) and f : a → c. . . . gn ). dn ). then so is γ(f . . . gn ) is the unique map making the cube XF FFF F FFF F F X a F FF FF FF f # c / bi  MMMg1 ⊔···⊔gn MMM M& / di  γ(f . Proof of Proposition 2. . Then the ﬁbrewise mapping space MapP(n)×Q(n) (π1 RP(n).. That M≃ (1) contains the unit element. g1 . . . . . dn and a. b1 .9. f ) ∈ M(n)  f is a homotopy equivalence} form a suboperad M≃ of M. The collection M = {M(n)} is topological operad. bn respectively. . . In particular. we write MapB (U. . . .g1 . g1 . g1 . follow from the unit and symmetry axioms respectively. Deﬁnition 2. . .. . gn are all homotopy equivalences. gn ) exists. Suppose that for both RP and RQ the pasting squares (1) are all homotopy pushouts. c. Thus M≃ is closed under composition. it contains M(n) as a subset. d) commute. Thus a unique such γ(f . . . gn )) . π2 RQ(n)).
γ(yn . In particular. mi . 4. . . and the equivalent shorthand for spaces obtained from RQ. . . which maps continuously into MapPΠ ×QΠ (RP⊔ . . (yi . RQ⊔ ) − → MapPΠ ×QΠ (RP⊔ . .2. We will use the following shorthand for spaces obtained from RP. h1 ). w) = γ(u. We will only show that composition is associative. − − ξQ ∗ ξ∗ 10 . But both maps are characterized as the unique map making each square x f / u yi  gi / vi  j zi  hj i / wj  i A _ _ _/ B. sending x1 to a point x2 . g). u. Now we will prove that the composition map M(n)×M(m1 )×· · ·×M(mn ) → M(m1 +· · ·+mn ) is continuous. ξP : RP⊔ → RP◦ denotes the map obtained from the pasting squares (2). We must show that the two pasted maps γ(γ(f. PΠ denotes the product P(n) × P(m1 ) × · · · × P(mn ). We begin by showing that M is an operad in sets. This is a direct consequence of the axioms for a realization system presented in Deﬁnition 2. RQ(n)) × MapP(mi )×Q(mi ) (RP(mi ). v). wn ). y). . . Now ξP and ξQ give continuous embeddings P MapPΠ ×QΠ (RP⊔ . γ(v1 . w1 ). . . . RP⊔ denotes the space obtained by pulling back each of RP(n) and RP(mi ) to PΠ and then forming the disjoint union. γ(gn . z) = γ(x. 1. the proof of the rest being similar but simpler. RQ◦ ) ← MapPΠ ×QΠ (RP◦ . gi ) ∈ M(mi ) j j j and (zi . z1 ). h) and γ(f . . 3. . 2. A _ _ _/ B. wi . the two pasted maps coincide. vi . Note that ξP and ξQ are proper ﬁbrewise surjections. Write i A = γ(γ(x. .) An element x of M(n) × M(m1 ) × · · · × M(mn ) can be regarded as a point x1 of the product MapP(n)×Q(n) (RP(n). γ(g1 .2. hj ) ∈ M(li ) for i = 1. . RQ(mi )). γ(vn . γ(y1 . A _ _ _ _/ B commute. . RQ◦ ). The proof that the permutation map σ : M(n) → M(n) is continuous is similar. RQ⊔ ). f ) ∈ M(n). zn )) and B = γ(γ(u. . RP◦ denotes the pullback of RP(m1 + · · · + mn ) to PΠ . hn )) coincide. n and j = 1. . Here the vertical maps come from Axiom 6 of Deﬁnition 2.10. . (The analogous claim need not hold for ξQ .Proof of Theorem 2. since X is compact and the square (2) is a ﬁbrewise pushout. Take elements (x. .
The bullet • indicates the basepoint on the boundary of the big disc. 11 . The action of Σn on f D(n) is given by permuting the n little discs: the ith disc of aσ is the σ(i)th disc of a. RQ( mi )) sending x3 to a point y. while the outgoing boundary map is given by the boundary of the big disc. which we call the local marked point. rotation and rescaling.3 (Realizations of framed little discs). The realization system Rf D is deﬁned as follows. . 3 Framed little discs and their realizations This section recalls Getzler’s framed little discs operad f D and introduces its realization system Rf D. . A typical element a ∈ f D(3) is drawn below. the realization of a is the space a ⊂ D2 obtained by deleting the interiors of the little discs from the big disc. The assignment x → y just described is nothing more than the composition map. 3. The outgoing boundary map ∂out : S 1 → a is given by the inclusion S 1 ֒→ D2 . • Given a ∈ f D(n) and σ ∈ Σn . In other words the ith incoming boundary map is given by the boundary of the ith little disc. and the description makes its continuity evident. This y lies in M(m1 + · · · + mn ).The image of x2 in the central space lifts to an element x3 of the right hand space. Getzler’s framed little discs operad f D has for its nth space the collection f D(n) of embeddings a: D2 → D2 of n little discs into a single big disc. . the symmetry map σ ∗ : a → aσ identiﬁes a and aσ as subsets of D2 . The unit 1f D is the identity embedding D2 → D2 . which we call the global marked point. . This completes the proof. Given a ∈ f D(n). ∂n : S 1 → a are given by ∂i = ai S 1 .1 Framed little discs Deﬁnition 3. Example 3.2 Realizations of framed little discs Deﬁnition 3. It is required that the ith component ai : D2 → D2 of the embedding be given by a combination of translation. 1 3 The dash on the ith little disc indicates the image of the basepoint of S 1 .2. • The incoming boundary maps ∂1 .1 (The framed little discs operad [5]). Finally we obtain a continuous map MapPΠ ×QΠ (RP◦ . RQ◦ ) → MapP( mi )×Q( mi ) (RP( mi ). 2 3. Composition in f D is given by composition of embeddings.
and in fact a homotopy pushout. bn ) is given as follows. b1 .3 follows immediately. . . b2 ) is depicted below: Deﬁnition 3. The realization of the element a ∈ f D(3) from Example 3. b2 . the pasting square n i=1 ∂i S1 ∂out / b1  ⊔ · · · ⊔ bn  (3) a / γ(a. . This is a pushout square.3 are satisﬁed by the realizations of framed little discs. b1 .• Given a ∈ f D(n) and bi ∈ f D(mi ) for i = 1. The lower map is the inclusion of a into γ(a.4. .2 follow from a lengthy but trivial veriﬁcation. b1 . . . In other words it sends bi  into the ith little disc of a. b1 . ∂i ∂out Now take the following elements a. The axioms of Deﬁnition 2. b1 . 1 2 a b1 1 1 2 2 b2 Then the pasting diagram (3) for γ(a. It is equipped with the projection map ρn : Rf D(n) → f D(n). Proposition 3. it is a ﬁbrewise compact Hausdorﬀ space over f D(n). is shown below. . The axioms of Deﬁnition 2. Example 3. bn ) by applying ai : D2 → D2 . As a closed subspace of f D(n) × D2 . together with its boundary maps. Proof. . For n 0 we deﬁne Rf D(n) to be the subspace of f D(n) × D2 whose ﬁbre over a ∈ f D(n) is precisely a ⊂ D2 . . . The continuity required by Deﬁnition 2. 12 . .2 and the conditions of Deﬁnition 2. . .2. . .5 (Topology on the realizations of framed little discs). n. bn ) given by regarding both as subsets of D2 .6. . The right hand map sends bi  into γ(a.
Then in §4. the composed cactus c ◦i d is deﬁned by the following pasting process.3 we introduce the realization system for cacti. Several variants have been introduced since then.1. See [13] and [3]. It is a suboperad of the coendomorphism operad CoEnd(S 1 ) of S 1 . We begin in §4. Kaufmann [8]. in particular by McClureSmith [10]. The operad was denoted there by f MS. .2 Salvatore’s cacti operad In this subsection we recall the variant of the cacti operad developed by Salvatore in [12]. The individual circles are called the lobes of the cactus. and the basepoints on the lobes are called the local marked points. along the lines of Voronov’s original deﬁnition in [13]. Given cacti c and d. but hope to give the unfamiliar reader an idea of the theory. n. Planar means that the circles meeting at each intersection point are equipped with a cyclic order. Deﬁnition 4. Rescale d so that its total length becomes the length of the ith lobe of c. 3 4. . Thus a cactus consists of n labelled circles of positive radius identiﬁed at ﬁnitely many intersection points. The cactus operad has for its nth term the collection of all cacti with n lobes. Here is a typical cactus with 4 lobes. A cactus with n lobes is a planar treelike conﬁguration of parameterized circles. Example 4.1 with an informal introduction to cacti along the lines of Voronov’s original deﬁnition. Finally in §4. and Salvatore [12]. We begin by establishing some notation. . Treelike means that the space obtained by ﬁlling in each circle with a disc is contractible.1 An introduction to cacti This subsection gives an informal description of the cactus operad. . which will be used in the rest of the paper. 13 . 4.2. 2 1 4 The bullet • indicates the global marked point. Then c ◦i d is obtained from c and d by collapsing the ith lobe of c onto d using the pinch map S 1 → d that sends the basepoint to the global marked point and then proceeds around the ‘outside’ of the cactus. of varying radii. but here we will call it C. Cacti were introduced by Voronov in [13]. while the dashes indicate the local marked points.2 we recall in detail Salvatore’s variant of cacti. We will not recall Voronov cacti in detail. labelled by 1.4 Cacti and their realizations This section introduces the cacti operad and its realization system. equipped with a global marked point on one of the circles.
n appear. . n}. . 1]. f. C is deﬁned to be the suboperad of CoEnd(S 1 ) whose nth term is the image of the embedding c : F(n)×Mon(I. 1. . j. Regard elements f ∈ Mon(I. This gives a based map πj : S 1 → S 1 . . The global marked point is given by c(∗). . Salvatore describes the precise relationship between C and Voronov’s original deﬁnition of cacti. [xk−1 . A point x ∈ F(n) is a partition of S 1 into n closed 1manifolds Ij (x) with equal length and piecewise disjoint interiors. Second. Write Mon(I.6 of [8] holds with C in place of Voronov cacti. Fix x ∈ F(n). For n 1 we deﬁne a space F(n) as follows. In particular. . Note 4. ∂I) for the space of nondecreasing maps I → I that preserve ∂I. Xk (x)) of the intervals [x0 . The constructions above induce an embedding c : F(n) × Mon(I. . Thus c = c(x. x1 ]. j with i = j. in the case when f has constant speed lj on Ij (x). . and can be regarded as a Voronov cactus in which the lobes have equal length. Note 4. For c(S 1 ) is a union of n parameterized circles. . .5. 2. This union of circles inherits a planar structure from the cyclic ordering of the components of the Ij (x). Then DX 14 . Here is the connection between the operad C and the informal notion of cacti given in the last subsection. Third. The jth of these circles is simply c(Ij (x)). Write DX ⊂ F(n) for the subset consisting of those x for which X(x) is contained in X. Given z ∈ S 1 write Lz : S 1 → S 1 for translation by z. Pull back the partition {Ij (x)} of S 1 to a partition of [0. ∂I) × (S 1 )n → CoEnd(S 1 )(n) that sends (x. f. Set F(0) = ∗. z4 with z1 . z4 ∈ int Ik (x) with j = k. f. First. z) determines a cactus in which the lobes have equal length. . Then c(x) is the product π1 × · · · × πn . . . Note 4. .3. . ∂I) as based maps f : S 1 → S 1 . Let X be a sequence satisfying the three conditions above.4 (Salvatore’s cacti operad). z3 ∈ int Ij (X) and z2 . Finally we recall the natural cell decomposition {DX } of F(n). This partition is speciﬁed by boundary points 0 = x0 < x1 < · · · < xk = 1 and a labelling X(x) = (X1 (x). xk ] by elements from {1. . Given j ∈ {1. z3 . collapse each component of S 1 \ Ij (x) to a point and rescale the quotient in order to identify it with S 1 . Given x ∈ F(n) we deﬁne a map c(x) : S 1 → (S 1 )n as follows.7. We regard f as a reparametrization of the pinch map of this cactus. there is no subsequence of the form i.3. It depends only on x and z. ∂I)× (S 1 )n → CoEnd(S 1 )(n). adjacent terms are distinct. . z2 . 4.Deﬁnition 4. i. z). Every such sequence X arises as X(x) for some element x ∈ F(n). all values 1. Fix c ∈ C(n) and write c = c(x. z) to the composite (Lz1 × · · · × Lzn ) ◦ c(x) × f . . . Theorem 5. and the condition on these cyclic orderings ensures that this union of circles is treelike. Deﬁnition 4. we can regard c as a cactus in which the jth lobe has length lj . 3. The sequence X = X(x) of the last paragraph satisﬁes the following three properties. n}. Consider the image c(S 1 ) ⊂ (S 1 )n . such that there is no cyclically ordered sequence z1 .6. depending only on x and z.
. . di+1 (xi ). . xi . . . . • The ith incoming boundary map ∂i : S 1 → c is deﬁned as follows.is homeomorphic to the product n ∆d(j) . . . d1 . di−1 (xi ). xi . . . xi . which we denote by c. Let c ∈ C(n) be a cactus with n lobes. . . Deﬁnition 4. . The realization of c is the topological space c(S 1 ). • Let c ∈ C(n) and let σ ∈ Σn . . tn ) = (tσ1 . . We deﬁne RC(n) to be the space of pairs RC(n) = {(c. . . The right hand map sends y ∈ di  to (d1 (xi ). f. xi ).8 (Realizations of cacti). dn ) = (d1 × · · · × dn )c(S 1 ) via d1 × · · · × dn . z) ∈ C(n) × (S 1 )n  z ∈ c(S 1 )}. d1 . . n. z). y. The next ﬁgures illustrate the incoming ∂i and outgoing ∂out boundary maps for the cactus of Example 4. . . . 4. The homeomorphism sends x ∈ DX to the point whose image in ∆d(j) records the lengths of the intervals labelled by j. . The outgoing boundary map ∂out : S 1 → c 1 n i−1 i+1 is given by c itself. • Let c ∈ C(n) and di ∈ C(mi ) for i = 1. . . . xi . it is ﬁbrewise compact and Hausdorﬀ over C(n). xi ). Then there is a pushout square n i=1 ∂i S1 ∂out / d1  ⊔ · · · ⊔ dn  (4) c / γ(c.10 (Topology on the realizations of cacti). .9. The lower map sends c = c(S 1 ) into γ(c. tσn ). 1 i−1 i+1 n Example 4. . . .2. .3 The realizations of cacti Deﬁnition 4. y. . . Then the homeomorphism σ ∗ : c → cσ is deﬁned by σ ∗ (t1 . where d(j) + 1 is the number of j=1 elements of X labelled by j. . . . . . . The realization system RC is deﬁned as follows. The ﬁbre of the projection map ρn : RC(n) → C(n) over c ∈ C(n) is naturally identiﬁed with c. dn ). . and ∂i is y → n 1 n i−1 i+1 (xi . . . . . Then c(Ii (x)) ⊂ c has the form xi × · · · × S 1 × · · · × 1 xi for some lobe coordinate (xi . . 15 . Write c = c(x. As a closed subspace of C(n) × (S 1 )n . dn (xi )).
We may therefore form the mapping operad M of Section 2 in the case P = f D. 4 and 5 follow immediately from the deﬁnitions. Since ∂out : S 1 → x. j w1 . c. The compatibility of pasting and boundary maps means that the pairs of maps u1 . Note that ∂i : C(n) × S 1 → RC(n) are continuous surjections between ﬁbrewise compact Hausdorﬀ spaces. Continuity of ∂i : C(n) × S 1 → RC(n) follows from the fact that the lobe coordinates of a cactus depend continuously on the cactus. and since the pasting squares (1) for Rf D and RC are homotopy pushouts. But this follows from the compatibility of pasting with boundaries. given c ∈ C(n) and d ∈ C(m). 5 The operad E In Sections 3 and 4 we saw that the framed little discs operad f D and the cacti operad C admit realization systems Rf D and RC with boundaries S 1 . By Proposition A. The topology on realizations of cacti satisfy the conditions of Deﬁnition 2. To prove this case it suﬃces to show that. and in fact a homotopy 2. Now we prove the ﬁnal part. it follows that the two members of each pair of maps coincide. For the ﬁrst part. w2 : zi  → Γ satisfy up ◦ ∂out = ∂out . Q = C. v2 : yi  → Γ. 3. Diagram (4) is a pushout. a simple inductive argument shows that it is suﬃcient to prove the claim when all but one of the di is the unit element. Proposition 4. These systems both consist of ﬁbrewise compact Hausdorﬀ spaces.5. where L = j−1 m 1 1 l1 + · · · + l1 1 + · · · + li + · · · + li . z i )).11. 16 . Write Γ = γ(γ(x. 1.3. γ(yi .7. 2. Continuity of ∂out : C(n) × S 1 → RC(n) and σ ∗ : RC(n) → RC(n) is immediate. Let us turn to axiom 6. z) = γ(x. f ) in which f is a homotopy equivalence. vp ◦ ∂out = ∂i . to prove continuity of the righthand map of (2) it will suﬃce to show that the composite with ∂i is continuous. u2 : x → Γ.It remains to show that the last two deﬁnitions do indeed endow C with a realization system RC. it has a suboperad M≃ consisting of triples (a. This follows from the deﬁnitions and from the fact that ∂i is a coﬁbration. the diagram S1 ∂i ∂out / d β c α / c ◦i d is a pushout. We now prove the second part. Similar reasoning shows that the lower map of (2) is continuous. and wp ◦ ∂k = ∂L+k . This is a topological operad. This can be proved using the cell decomposition of F(n) given in Note 4. Axioms 1. y). Proof. pushout. 3. The realizations of cacti satisfy the conditions of Deﬁnition 2. ∂out : S 1 → yi  and j ∂k : S 1 → zi  are all surjections. v1 . This is given by the next proposition.2.
f ). 17 .1 The f Dalgebra Ω2 BG In order to demonstrate how realization systems on an operad can be used to deﬁne algebras over that operad. . Proposition 6. c. Deﬁne νn : f D(n) × (Ω2 BG)n → Ω2 BG by νn (a. . which states that the three are weakly equivalent as Ealgebras. First note that. c. Proofs are omitted.1 with Ω2 BG. The collection ν = {νn } makes Ω2 BG into a f Dalgebra. With this in hand. Theorem 2.2. to prove Theorem A it remains to show that the projection maps π1 and π2 are weak homotopy equivalences. for a ﬁxed a ∈ f D(n).2 discusses ΩG and §6. Lemma 6. . and the projections − −→ − −− f D ← −1 − E − −1 − C are morphisms of operads.3 discusses εG. Given a ∈ f D(n) and x1 . We begin in §6. f ) ∈ E(2) is depicted in the introduction. E is a topological operad.1. 1. Deﬁnition 6.1. and f : a → c is a homotopy equivalence satisfying f ◦ ∂i = ∂i and f ◦ ∂out = ∂out . we will brieﬂy describe the f Dalgebra Ω2 BG. There is little to say in this case. . Finally §6. c ∈ C(n). Set E = M≃ . . and proves Theorem B. This section introduces all three algebras using the realization systems for f D and C. Then §6. . π π 6 Algebras from topological groups The aim of this paper is to compare the f Dalgebra Ω2 BG with the Calgebra ΩG by constructing an intermediate Ealgebra εG. 6. .1. x1 . xn ) = ξ. 2.Deﬁnition 5. Its restriction to a is constant. there is a unique map ξ : D2 → BG with the following two properties. .3. where ξ is the map of Lemma 6. where a ∈ f D(n). but we include it for emphasis.10 immediately gives us the following result. there is a pushout diagram n 1 / n D2 i=1 i=1 S ∂i a a / D2 . An element (a. Thus E(n) is the space of triples (a.2. xn ∈ Ω2 BG.4 gives some deferred proofs. Its restriction to the ith cofactor of n i=1 D2 is given by xi . Proposition 5.
Proof. . The collection ω = {ωn } makes ΩG into a Calgebra. . the composite α ◦ ∂i : S 1 → G is a left translate of γi . Proposition 6. α(•) = e.4. 18 . Now let cP ∈ C(2) denote the cactus in which two circles of radius 1/2 are joined at their basepoint. Deﬁnition 6. . for each i = 1. . Then ω1 (c− . under which the algebra structure ω pulls back to the standard E1 algebra structure on ΩG. let cs ∈ C(1) denote the cactus in which • = ∂1 (s) and ∂out has constant speed.5. The realization c is a treelike conﬁguration of circles. . γn ) = α ◦ ∂out . We will describe it directly in terms of the realizations RC. . For example. The proof will be given in §6. given s ∈ S 1 . γn be elements of ΩG. The action ω encodes some natural properties of ΩG.4. and the global marked point lies on the ﬁrst circle at its basepoint: Then the map ω2 (cP . −. −) : S 1 × ΩG → ΩG is the circle action that sends (s. There is a unique α1 : ∂1 (S 1 ) → G sending • to e and such that α1 ◦ ∂1 is a left translate of γi . 2.5. Such maps will be called left translates of f . It has a unique extension α2 : ∂1 (S 1 ) ∪ ∂2 (S 1 ) → G for which α2 ◦ ∂2 is a left translate of γ2 . Deﬁne ωn : C(n) × (ΩG)n → ΩG by ωn (c. where • denotes the global marked point in c. the claim follows. γ) to the loop t → γ(s)−1 · γ(s + t). Let c be a cactus with n lobes and let γ1 . . where α : c → G is the map of Lemma 6. γ1 .7. −) : ΩG × ΩG → ΩG is the ordinary Pontrjagin product given by concatenating loops. . Let X be a space and let Y be a Gspace. . This algebra structure was ﬁrst obtained by Salvatore in [12] using his proof of the topological cyclic Deligne conjecture. . Then there is a unique map α : c → G satisfying: 1. Proceeding in this way. so after relabelling the lobes we may assume that ∂1 (S 1 ) contains • and that each ∂i+1 (S 1 ) meets ∂1 (S 1 )∪· · ·∪∂i (S 1 ) in a single point.2 The Calgebra ΩG We now explain how to make ΩG into a Calgebra. . This circle action was considered by Menichi in [11] and by Salvatore in [12]. Lemma 6. there is a natural inclusion of the little intervals operad into C. More generally. n. . Notation 6. Given a continuous map f : X → Y and an element g ∈ G.6. we write g · f : X → Y for the map that sends x ∈ X to g · f (x).6.
Proof. f ) ∈ E(n) and φ1 . Let εG be the space of maps φ : D2 → EG for which φS 1 factors through ιG . Proposition 6. c. g → g · ∗. . n i=1 D2 is a left translate of φi . We will again make use of the pushout diagram n i=1 ∂i S1 / n i=1 D2 (5) a a determined by an element a ∈ f D(n).12. The projection maps p1 : εG → Ω2 BG. . where α is the map obtained by applying Lemma 6. Their proofs are given in the next subsection. . / D2 Lemma 6. The conditions mean that ψa must factor as ιG ◦ α ◦ f .5 to c and p2 (φ1 ). Let EG → BG be the universal principal G bundle. The collection ε = {εn } makes εG into a Ealgebra. and p2 is deﬁned by ιG ◦ p2 (φ) = φS 1 . . Its value on ∗ ∈ D2 is ιG (e).10. Given (a. . φn )) = ψ. p2 : εG → ΩG. We will give it the structure of an Ealgebra. Note that ψ ∈ εG. Its restriction to a factors through f : a → c. and show that it is weakly equivalent to both Ω2 BG and ΩG as Ealgebras. Proposition 6. 19 .3 The Ealgebra εG Now we will deﬁne the space εG. .9. . p2 : εG → ΩG are homotopy equivalences of E algebras. where p1 is obtained by projecting from EG to BG. There is a unique extension of this map to D2 satisfying the second condition. f ). The following two propositions together prove Theorem B. and for which φ(∗) = ιG (e). c. 2. Deﬁnition 6. Fix a basepoint ∗ of EG and use it to deﬁne a basepoint of BG and an inclusion ιG : G ֒→ EG.9. Deﬁnition 6. . . Its restriction to the ith cofactor of 3.6. there is a unique map φ : D2 → EG with the following properties: 1.11. . .8. p2 (φn ). where ψ is the map of Lemma 6. . (φ1 . The algebra structure on εG is an almost literal mixture of the existing algebra structures on Ω2 BG and ΩG. There are projections p1 : εG → Ω2 BG. φn ∈ εG. Deﬁne εn : E(n) × (εG)n → εG by εn ((a.
γ i )). . . . . . . . γn ) → gi · γi (t) for some gi ∈ G. . . . . . γ1 . . .5 to obtain three diﬀerent maps.6. d1 . 6. c. (c. Proof of Proposition 6. dn ) The ﬁrst square comes from the deﬁnition of β. di ∈ C(mi ) and i i γ1 . and by the compatibility of λ with the boundary maps we have β ◦ ∂out = α ◦ λ ◦ ∂out = α ◦ ∂out as required. . . . . n. γmn and we write γ for γ1 . . . . γmi ∈ ΩG for i = 1. γmi ). . . 1. We must show that ωm1 +···+mn (γ(c. d1 . .11 and 6. Let c ∈ C(n). and in particular is a pushout. . . . Comparing the squares we see that β factors as α′ ◦ λ for some α′ : γ(c. Let γ i : di  → G denote the map obtained using di and γ1 . . dn ) 1 n and γ1 . i i 2. We can ﬁnd g1 . . . . dn ) → G. . 20 . . .2 show that the unit element 1C acts as the identity on ΩG. . γmi for each i = 1. γn ) → gi are continuous. . . whose continuity can be proved using the cell decomposition of F(n) given in Note 4.7. . Let α : γ(c. These maps factor through a map F(n) × (ΩG)n → G. γmi . . . . . . γσ−1 1 . . . . . . The second is a pasting square (1). d1 . γ) = ωn (c. The map ∂i : S 1 × C(n) → RC(n) is proper surjection of ﬁbred spaces. . γ1 . . . It will therefore suﬃce to show that the assignments C(n) × (ΩG)n → G. Lemma 6. gn ∈ G and two commutative squares: S1 ∂i ∂out / d1  ⊔ · · · ⊔ dn  gi ·γ i ∂i S1 ∂out / d1  ⊔ · · · ⊔ dn  µ c β /G c λ / γ(c.4 Proof of Propositions 6. The compatibility of λ and µ with the boundary maps means that α′ in fact satisﬁes the properties that characterize α. .12 In this subsection we resume the practice of using underlines to indicate tuples i i of elements. .5 gives us a function α : RC(n) × (ΩG)n → G and it suﬃces to show that this is continuous. . γ1 . γσ−1 (n) ) holds. . i i 3. . Now let us show that ω is associative. . . . 2 and 3 of Deﬁnition 2. . Let β : c → G denote the map obtained using c and the ωmi (di . n. . we write γ for i i i 1 n γ1 . . γ1 . . γn ) = ωn (c. dn ) → G denote the map obtained using γ(c. . For example. . This composite is given by (t. given γ1 .5 it will suﬃce to show that each composite α ◦ ∂i is continuous. . . We must show that α ◦ ∂out = β ◦ ∂out . . Thus β = α ◦ λ. . . and that the equivariance property ωn (cσ. d1 . We can use Lemma 6. Let us show that ωn is continuous. so that α′ = α. Axioms 1. . γmi . so by Proposition A. ωmi (di . γmn .7.7. d). .
. 21 . in the proof of Proposition 6. show that the restriction of ψ to A factors through F : A → C. εmi (bi . Using this diagram we can characterise the right hand side of (6) as the map ψ : D2 → EG that 1. . . We must show that εm1 +···+mn (A. c. n. φn )) → (g1 . equivariance and associativity rules. To show that ε makes εG into a Ealgebra in the topological setting we will show that εn : E(n)×(εG)n → εG is continuous. . while α was deﬁned. 2. and write their composi ite as (A. n and j = 1. . Take φj ∈ εG for i = 1. This property. f ) by x → f (x). j 4. . . mi .7. . j)th cofactor D2 is a left translate of φi . (6) S1 / i j D2 1 iS / i bi  / i D2 a / A / D2 The upper square comes from (5) for the bi . gi ). c. together with the deﬁnition of F . . We will only prove associativity as the other two properties can be proved in a similar but much simpler way. the left hand square is the pasting square for A. φi Consider the following diagram of pushout squares. gi ) ∈ E(mi ) for i = 1. . Here β is given in the ﬁbre over (a. i j . and the lower two squares compose to give (5) for a. c. on the (i. when restricted to a. f ) ∈ E(n) and (bi . c. di . .8. Equality follows. ∗ η : π1 Rf D(n) × (εG)n → EG. . To begin we must show that ε satisﬁes the unit. ∗ β : π1 Rf D(n) → RC(n). For ξ this follows from continuity of the map ((a. . Since the lower and righthand maps of diagram (5) give inclusions of closed subsets D2 × E(n) ֒→ D2 × E(n) ∗ and π1 Rf D(n) ֒→ D2 × E(n). and is continuous by part 5 of Proposition A. 3. . C. C.7. f ). . Thus ε makes εG into a Ealgebra in sets. F ). gn ) in the proof of Proposition 6. the right hand two squares compose to give (5) for A. To do this we will show that the adjoint D2 × E(n) × (εG)n → EG is continuous. characterize the left hand side of (6). (φ1 . and shown to be continuous. The ﬁrst two of these properties. factors through f : a → c. For η it follows from continuity of two maps α : RC(n) × (ΩG)n → G. sends ∗ to ιG (e). together with with the third and fourth properties above. Take (a. . .Proof of Proposition 6. it suﬃces to show continuity of the restrictions ξ: D2 × E(n) × (εG)n → EG. . φ = εn (a.11. . di . factors through gi : bi  → di . when restricted to bi . F ). f ). .
s) = F ′ (x. but the ﬁbre over a pair (a.1. Deﬁnition 7. f ◦ ρa ) are homotopy inverse to the inclusions above. It is immediate that both p1 and p2 are Ealgebra morphisms. Let f D◦ (n) ⊂ f D(n) consist of those elements whose little −1 discs do not meet the boundary of the big disc. To solve this.2. To prove Theorem A it will therefore suﬃce to prove that the restrictions π1 : E◦ (n) → f D◦ (n) and π2 : E◦ (n) → C(n) are weak equivalences. c. but this is not the case: a typical ﬁbre is nonempty. d. For p2 the homotopy lifting problem is adjoint to an extension problem that can always be solved. and then deﬁne F by F (x. We ﬁrst show that p1 . c. write a1/2 for the element obtained by halving the radii of the little discs of a. The inclusions f D◦ (n) ֒→ f D(n) and E◦ (n) ֒→ E(n) are homotopy equivalences. To do this we now consider the map Π : E◦ (n) → f D◦ (n) × C(n). Then the maps f D(n) → f D◦ (n). These are contractible. It now suﬃces to show that π1 and π2 have contractible ﬁbres. Write F ′ X × ∗ × I = ιG ◦ f . (a. This means that for each n 0. We must now show that both p1 and p2 are homotopy equivalences. The theorem states that the projections π1 : E → f D and π2 : E → C are weak homotopy equivalences of operads.Proof of Proposition 6. Write ρa : a1/2  → a for the map that preserves a ⊂ a1/2  and that projects points inside a little disc of a to the boundary of that little disc. Given a ∈ f D◦ (n). 7 Proof of Theorem A This section gives the proof of Theorem A. p2 are ﬁbrations. d. Proof. s) · f (x.12. by contractibility of D2 and EG respectively. while p−1 (γ) is the space 2 of maps D2 → EG whose restriction to S 1 is ιG ◦ γ. c) can be empty if the little discs of a meet the boundary of the big disc. To show this we will consider the product Π : E(n) → f D(n) × C(n) of π1 and π2 . But p−1 (x) 1 is homeomorphic to the space of based maps D2 → G. ﬁrst choose an arbitrary lift F ′ . For p1 the homotopy lifting problem is adjoint to the problem of ﬁnding a lift F in a diagram of the form h / X × D2 × {0} EG r8 rr r rr F / BG X × D2 × I H with the additional property that F (X × ∗ × I) is constant with value ιG (e). stating certain results that will be proved in the remaining sections of the paper. the maps π1 : E(n) → f D(n) and π2 : E(n) → C(n) are weak equivalences of spaces. 22 . This issue is easily remedied. and let E◦ (n) = π1 f D◦ (n) ⊂ E(n). It would be ideal for us if Π were a ﬁbration. f ) → (a1/2 . a → a1/2 and E(n) → E◦ (n). s)−1 . for then the boundary conditions can be overdetermined. Lemma 7.
c). We will use the quasiﬁbration Π to show that π1 and π2 are weak equivalences. (The relevant background is recalled in §8. c) ∈ f D◦ (n) × C(n) and write Map∂ (a. Then there is a neighbourhood U of (a. such maps are themselves quasiﬁbrations. Recall that E◦ (n) consists of maps from ﬁbres of Rf D◦ (n) to ﬁbres of RC(n) satisfying certain boundary conditions. c). are given in §8. We can therefore show that π0 (E◦ (n)) = ∗ and that p identiﬁes π1 (E◦ (n)) with the diagonal subgroup of P RBn × P RBn by proving the next proposition.5. Now let us discuss the proof of Theorem C.5. Write P RBn for the pure ribbon braid group. Exactness at the last two terms of the sequence means that q −1 (∗) is the image of p and r−1 (∗) is the image of q. δ)·φ = δ · φ · γ −1 . The projection Π : E◦ (n) → f D◦ (n) × C(n) is a quasiﬁbration. The promised recollections on the homotopy type of f D◦ (n) and C(n).3. It follows immediately from (7) that πi (E◦ (n)) is trivial for i > 1. c) over which we can ﬁnd a ﬁbrewise homotopy equivalence E◦ (n)U ≃ U × Map∂ (a. Fix (a. Corollary 7.which has much better ﬁbrewise properties than its predecessor. which completes the proof of Theorem A. or in other words. Write Fn for a typical ﬁbre of Π. and q is given by applying this action to the constant element ∗. or in other words that there is a long exact sequence of homotopy groups: · · · −→ πi (Fn ) −→ πi (E◦ (n)) −→ πi (f D◦ (n)) × πi (C(n)) −→ · · · (7) We will establish the criteria listed in the last paragraph by exploiting this long exact sequence. and that Π∗ : π1 (E◦ (n)) → P RBn × P RBn identiﬁes π1 (E◦ (n)) with the diagonal subgroup of P RBn × P RBn . c) for the ﬁbre of E◦ (n) over (a. Proposition 7. Theorem C. The action of P RBn ×P RBn on P RBn is given by (γ. → p q r (8) There is an action of P RBn × P RBn on P RBn .1 and §8. In particular Fn has contractible components. Then both f D◦ (n) and C(n) are EilenbergMacLane spaces K(P RBn . There is a homotopy equivalence Fn → P RBn . By Corollary 2. For Π to be a quasiﬁbration means that Fn is weakly equivalent to the homotopy ﬁbre of Π. and that π1 (E◦ (n)) and π0 (E◦ (n)) are computed using an exact sequence → → 0 → π1 (E◦ (n)) − P RBn × P RBn − P RBn − π0 (E◦ (n)) → 0.2.) It will therefore suﬃce to show that πi (E◦ (n)) is trivial for i = 1.4 of [9]. Our 23 . 1). Proposition 7. together with the proof of Propositions 7.4 and 7. This theorem states that E◦ (n) is locally ﬁbrewise homotopy equivalent to a product.4. that it is a homotopy ﬁbre bundle. Thus each point of f D◦ (n) × C(n) has arbitrarily small neighbourhoods over which Π is a quasiﬁbration.
Recall that these propositions compute the homotopy exact sequence associated to the quasiﬁbration Π : E◦ (n) → f D◦ (n) × C(n). and lead to the proof of Theorem A. and it is homotopy equivalent to the homotopyrelative mapping space E◦ (n). where X and Y are spaces and fX : A → X and fY : A → Y are maps from some ﬁxed space A. The corresponding relative mapping space is E◦ (n). ∂ × Id) and (c × C(n). fZ ). here ∂ denotes the combined boundary map.4 we prove the propositions. . h In Sections 10 and 11 we study the pairs (Rf D◦ (n). . The program outlined above extends without modiﬁcation to the ﬁbrewise setting. We then introduce two mapping spaces. The generator of the ith cyclic factor is written as ζi . fX ) or (Y. . Then in §8.5. and that homotopy type of the space of homotopyrelative maps does not change when we replace (X.1 The pure ribbon braid group This subsection collects some facts about the pure ribbon braid group that will be useful in the subsections to follow.4 and 7. fX ) and (Y. xn . . ∂) respectively. . There is an embedding Bn ֒→ Aut(Fn ) sending σi to the transformation xi → xi xi+1 x−1 . ∂) and (RC(n). ∂) over C(n).3 and §8. or in other words satisfying g ◦ fX = fY . Artin’s braid group Bn on n strands is the group on generators σ1 . The pure ribbon braid group is the direct product P RBn = P Bn × Zn . Finally. i + 1.proof of Theorem C will follow from general results on spaces of maps satisfying boundary conditions.1 and §8. fY ) with a homotopy equivalent pair (Z. together with a study of the local ﬁbrewise structure of Rf D◦ (n) and RC(n). In Section 9 we consider pairs (X. Let Fn denote the free group on generators x1 . where we can consider the pairs (Rf D◦ (n). for example from [7]. 8 Computing the long exact sequence This section is given to the proof of Propositions 7.2 we recall the pure ribbon braid group and its relationship to framed discs and cacti. these pairs are homotopy equivalent to trivial ﬁbred pairs of the form (a × f D◦ (n). . σn−1 subject to the relations σi σj = σj σi for i − j 2 and σi σi+1 σi = σi+1 σi σi+1 . one consisting of maps g : X → Y relative to A. We show that. ∂ × Id) respectively. Recall. We show that under certain conditions the two mapping spaces are homotopy equivalent. which are equipped with homotopies g ◦ fX ≃ fY . . in Section 12 we combine the results of the preceding three sections to prove Theorem C. 24 xl → xl for l = i. . In §8. locally over f D◦ (n) and C(n). Its kernel is the pure braid group P Bn . i xi+1 → xi . ∂) over f D◦ (n) and (RC(n). There is a homomorphism Bn → Σn sending σi to the transposition of i and i + 1. . fY ). which has generators −1 −1 2 −1 αij = σj−1 σj−2 · · · σi+1 σi σi+1 · · · σj−2 σj−1 for 1 i < j n. and a second consisting of maps g : X → Y homotopyrelative to A. that E. 8.
the second component is (m1 . . n − 1. 1 in turn at constant speed. . Deﬁnition 8. Let cn denote the cactus in which all lobes have equal length and meet at a unique point. . Thus the outgoing boundary map ∂out : S 1 → c traverses the lobes n. where mi is the sum of the exponents of the letter xi in the word wi . The intersection point coincides with the global marked point and also the local marked point on each lobe. The isomorphism Λ translates the group structure on P RBn into the operation (v1 . . . . vn ) · (w1 . The fundamental group π1 (an . . .1. wn ) = (α(v)(w1 )v1 . the ith generator xi is represented by the loop which travels along li . . . n. . •) can be identiﬁed with Fn . which is more directly related to the topological situation at hand. For i = 1. . 1 at the intersection point. In later subsections we will use the following alternative description of the pure ribbon braid group. then around the ith circle and back down li . are arranged horizontally across the big disc in order 1. There is a bijection Λ : Wn − P RBn . . ∼ = 8. . . . l1 ··· ln 1 n an We will sometimes regard the li as paths li : [0. . . 1). . wn ) ∈ (Fn )n let α(w) denote the homomor−1 phism Fn → Fn that sends xi to wi xi wi . so that the local marked point lies at the top of each little disc. and whose outgoing boundary ∂out : S 1 → cn  has constant speed. Given w = (w1 . The ﬁrst component of Λ(w) is α(w) ∈ → P Bn . . 1] → an . . . . and the global marked point is chosen to lie on the nth lobe. Deﬁnition 8. n let li denote the line segment in an  passing from the basepoint • of the big disc to the basepoint ∂i (∗) of the ith little disc.This embedding identiﬁes Bn with the subgroup consisting of those automorphisms f : Fn → Fn for which f (x1 · · · xn ) = x1 · · · xn and for which there is a permutation π of {1. mn ) ∈ Zn . . .2. . . Set Wn = {w ∈ (Fn )n  α(w) ∈ Aut(Fn ). . The lobes are cyclically ordered n. . The same embedding identiﬁes P Bn with the analogous subgroup where the permutation π is the identity map. α(w)(x1 · · · xn ) = x1 · · · xn }. . . We begin by choosing basepoints. . . Let an ∈ f D◦ (n) be an element whose little discs all have the same (small) radius. and sometimes as subsets li ⊂ an . cn n 1 25 . . cacti and the pure ribbon braid group This subsection recalls how both f D(n) (or equivalently f D◦ (n)) and C(n) are EilenbergMacLane spaces K(P RBn . α(v)(wn )vn ). . . . .2 Discs. and which are embedded without rotation. n} so that each f (xi ) is conjugate to xπ(i) . . .
The sections of the covering space correspond to the ﬁxed points of this action. which describes the homotopy type of the ﬁbre of Π. xn .1 of [7].The fundamental group π1 (cn . Deﬁnition 8. . Proposition 3. .4 by showing that this map Fn → Wn is a homotopy equivalence. Ωf cn  is ﬁbre homotopy equivalent to a covering space with ﬁbre π1 (cn . Deﬁne a map Fn → Wn as follows. and are then rotated around one another anticlockwise before returning to their original position. .4. •) is obtained by applying f to the line segment li .4. where wi ∈ Fn = π1 (cn . (This description of the generator is taken from the proof of Kaufmann’s result quoted above. (j − 1)th lobes so that it is adjacent to the ith lobe.) The generator ζi corresponds to the loop in which the parameterization of the ith little disc is rotated through one full turn. 1).5. It is well known that f D(n) (or f D◦ (n)) is a K(P RBn . Let Fn denote the ﬁbre of Π : E◦ (n) → f D◦ (n) × C(n) over (an . A point f ∈ Fn is sent to (w1 . wn ). Kaufmann [8. Lemma 8. We must show that this covering = space admits a unique section. 1). equipped with the compactopen topology. . Throughout this maneouver the framing of the little discs does not change. •) over •. R2 ) of n ordered points in R2 . The generator αij corresponds to the loop based at an in which the ith and jth little discs are brought side by side beneath the other little discs. . A result of R. Deﬁnition 8. since the braid group admits an automorphism inverting all of its standard generators. . We will prove Proposition 7. Since cn  is a K(Fn . for example). Note that Ωf (Fn ) is the space of sections of the ﬁbration Ωf cn  → an  obtained by pulling back the free loopspace ﬁbration Ωcn  → Lcn  → cn  along f . The generator αij corresponds to the loop based at cn in which the jth lobe travels across the (i + 1). Proof. The covering space is determined by an action of the fundamental group π1 (an . this does not aﬀect the outcome of that proposition. . Here xi is represented by the loop ∂i : S 1 → c. for any f ∈ Fn . It will suﬃce to prove that. We must show that the space of sections of this ﬁbration is contractible. The ﬁbre Fn has weakly contractible components. and Conf(n. 1). The boundary conditions guarantee that the image of f ◦ li is a loop in cn . The proof is given in the next three lemmas. For instance. 26 . .19] states that C(n) is a K(P RBn . cn ). In order to obtain this description we have reversed the path αi in the proof there. The generator ζi corresponds to the loop in which the framing of the ith little disc is rotated through one full turn. . . . the ith and jth lobes then rotate around one another anticlockwise before the jth retraces its steps to its original position. the loopspace Ωf (Fn ) is contractible. •) can be identiﬁed with the free group Fn on generators x1 . R2 ) is a K(P Bn . f (x)) ∼ Fn over x.4. Its ﬁbre over x is Ωf (x) cn . . 8. •) of the base on the ﬁbre π1 (cn . It is the space of homotopy equivalences f : an  → cn  such that f ◦ ∂i = ∂i and f ◦ ∂out = ∂out .3.3 The ﬁbre of Π : E◦ (n) → f D◦ (n) × C(n) The purpose of this subsection is to prove Proposition 7. f D(n) is homotopy equivalent to the product of n circles with the conﬁguration space Conf(n.3. 1) (see section 1.
4 The action of π1 (f D◦ (n) × C(n)) on Fn This subsection will give the proof of Proposition 7. γ) · 1 = 1. Now f1 and f2 coincide on the image of every boundary map and are homotopic rel. Then γ can be lifted to a path in E with initial point f ∈ F and ﬁnal point g ∈ F . Lemma 8. By construction h is an element of Fn and its image in Wn is exactly w. Proof. then the action of π1 (B) on π0 (F ) is deﬁned as follows. and the eﬀect of [γ] on [f ] is given by [γ] · [f ] = [g]. This completes the proof. and Lemma 8. . But since cn  is a K(Fn . then it still leads to a description of the action of π1 (B) on π0 (F ). where we show that (γ. then they lie in the same component of Fn . 1) · φ = φ · γ −1 . Let an  denote the space obtained from an  × I by identifying Kn × I with a single copy of Kn . and that g ◦ ∂out is based homotopic to ∂out . through elements of Fn . δ) · φ = δ · φ · γ −1 . Write Kn ⊂ an  for the union of Jn with the boundary of the big disc D2 . when it is possible. f2 ∈ Fn . Proof. and that is given by the γi on li . Choose an identiﬁcation map D2 → an  that is a bijection away from the boundary S 1 . Then we obtain identiﬁcation maps S 2 → an  ∪Kn an  and B 3 → an . Then we can form a homotopy between gKn and a map Kn → cn  that commutes with the ∂i and ∂out . ω is ﬁxed by the action if and only if it commutes with every element of Fn . If F → E → B is only a quasiﬁbration then the lifting process just described may not be possible.8. . Let Kn ⊂ an  be as in the preceding proof. Let us therefore assume that f1 and f2 coincide on Kn . if F → E → B is a ﬁbration. and so we may extend f to a map g : an  → cn . •) = Fn on π1 (cn . The map Fn → Wn is surjective. endpoints on each li . 27 . In general. The proof consists of Lemma 8. we may extend the homotopy to one between g and some h : an  → cn . Let f : Jn → cn  be the map that satisﬁes f ◦ ∂i = ∂i for each i and that is given by γi on li . Take a loop γ in B and a point f in F . Now Jn is a strong deformation retract of an . . Since Kn ֒→ an  is a coﬁbration. which states that the fundamental group P RBn × P RBn of f D◦ (n) × C(n) acts on π0 (Fn ) ∼ P RBn = by the rule (γ. Write Jn ⊂ an  for the union of the boundaries of the little discs with the line segments li . f 8. We may now combine f1 and f2 to obtain a map f : an ∪Kn an  → cn . Let f1 . •) = Fn is given by xi (ω) = φ(xi )ωφ(xi )−1 . This map satisﬁes g ◦ ∂i = ∂i .9. Choose w = (w1 . If the images of these elements in Wn coincide. .If φ ∈ Aut(Fn ) is the automorphism determined by f . Since Kn ֒→ an  is a coﬁbration.5. 1) this is always possible.6.7. we may therefore apply a homotopy to f2 . until it coincides with f1 on all of Kn . where we show that (γ. It has ‘boundary’ an  ∪Kn an . since w ∈ Wn it follows that g is a homotopy equivalence. However. Since φ is an automorphism. To extend f to an  it will suﬃce to extend the composite S 2 → → an  ∪Kn an  − cn  to B 3 . Also. Lemma 8. The identity element is the unique element with this property. wn ) ∈ Wn and represent each wi by a based loop γi : I → cn . then the action of π1 (an . and this completes the proof. To produce a homotopy between f1 and f2 it will suﬃce to extend f to all of an .
Then an (t) and cn (t) are depicted here: 2 1 2 1 2 1 1 2 1 2 t=0 0<t<π t=π π < t < 2π t = 2π 1 2 1 2 2 1 1 2 1 2 And the family of embeddings cn (t) → an (t) is depicted here: Lemma 8. t ∈ [0. It will suﬃce to prove this for γ = ζi and for γ = αij . Represent φ ∈ Wn by (an . so that (γ.5 satisﬁes (γ. Loops in f D◦ (n) and C(n) representing these elements were described in §8. and such that the highest point each lobe is its local marked point. and inwards from the boundary of the big disc onto the cactus.2. The action of Proposition 7.5 satisﬁes (γ. Then for each t ∈ S 1 we can ﬁnd maps fn (t) : an (t) → cn (t) such that t → (an (t). Then we can form a 28 .8. cn . We identify P RBn with Wn . fn (t)) deﬁnes a loop in E◦ (n). t → cn (t) be the loops in f D◦ (n) and C(n) respectively that represent αij . γ) · 1 = 1 for all γ ∈ P RBn .9. Then t → (an (t). fn ) determines a loop in E◦ (n) and our claim follows.Choose a basepoint fn ∈ Fn . Proof. Now ﬁx γ = αij and let t → an (t). Lemma 8. Represent γ by a loop t → an (t) in f D◦ (n) based at an . Let us begin with the case γ = ζi . cn ). such that the image of the ith lobe separates the ith little disc from the others. f ) ∈ Fn . j = 2. cn (t). The representative of γ can be chosen so that there is a continuous family of homeomorphisms hn (t) : an  → an (t). 1) · φ = φ · γ −1 for all γ ∈ P RBn . Let t → an (t) and t → cn (t) be the loops in f D◦ (n) and C(n) that represent ζi . For at all times t ∈ S 1 we can ﬁnd an oriented embedding of cn (t) into an (t) that sends the basepoint to the basepoint. satisfying hn (t) ◦ ∂i = ∂i and hn (t) ◦ ∂out = ∂out for all t. Proof. For example take n = 2. The action of Proposition 7. 1) is represented by t → (an (t). where fn is any function sending each li to the constant path. Then the required fn (t) can be constructed by projecting out from the centre of the ith little disc onto the ith lobe. i = 1. 1]. cn (t).
2. cn . 9 Relative mapping spaces This section begins our work towards a proof of Theorem C. this section will study spaces of maps satisfying a condition of the form f ◦ ∂ = ∂. we will consider spaces of maps g : X → Y satisfying the strict condition g ◦ fX = fY . . Here we will prove nonﬁbrewise versions of all results. Here we have written ∂ for the maps ∂out ⊔ ∂1 ⊔ · · · ⊔ ∂n : ∂out ⊔ ∂1 ⊔ · · · ⊔ ∂n : n i=0 n i=0 S 1 → a. Homotopies will always be parameterized by [0. In order to understand E◦ (n). Using the description of the group operation on Wn . fX ) and (Y. We must show that f ◦ hn (1)−1 ∈ Fn represents φ · γ −1 . we write H2 ·H1 : g0 ⇒ g2 for their concatenation. and another space of maps homotopyrelative to A. Note 9. which we call maps homotopyrelative to A. endpoints to the concatenation of a loop ωi based at • with the arc li . fY ) under A. . which we call maps relative to A. by which we mean pairs (X. c) consists of homotopy equivalences f : a → c with the property that f ◦ ∂ = ∂. There are no further assumptions on fX . In fact the assignment hn (1) → w is a homomorphism from the group of boundaryﬁxing homeomorphisms of an  into Wn . Given maps g. and the identities w = γ follow.path in E◦ (n) given by t → (an (t). However. Then wi = [ωi ]. wn ) ∈ Wn as follows. and also maps g : X → Y equipped with a homotopy g ◦ fX ≃ fY . We will consider spaces under A. it follows that f ◦ hn (1)−1 represents precisely φ · w−1 . Recall that Theorem C describes the ﬁbrewise homotopy type of E◦ (n) locally over f D◦ (n)×C(n). In these cases the family of homeomorphisms hn (t) can be seen explicity.1. f ◦ hn (t)−1 ). The ﬁrst part introduces what it means for two spaces under A to be homotopy equivalent. To hn (1) we assign an element w = (w1 . f ◦ hn (1)−1 ) ∈ Fn . and so is homotopic rel. We compare the two notions. and we show that the homotopy type of the homotopyrelative mapping space is preserved if one of the spaces under A is replaced by a homotopyequivalent space under A. Given homotopies H1 : g0 ⇒ g1 and H2 : g1 ⇒ g2 . ‘Hausdorﬀ’ and ‘compactopen’ with their ﬁbrewise analogues. So we must show that w = γ. 1] = I. whose ﬁbre over (a. . The arc hn (1)(li ) in an  has the same endpoints as li . The section has two parts. Given spaces (X.h k l A − B −→ C − D → − → 29 . Fix a compact Hausdorﬀ space A. Recall also that E◦ (n) is a ﬁbrewise mapping space. and so it suﬃces to prove w = γ in the special cases γ = αij and γ = ζi . The second part introduces two mapping spaces between spaces under A. all results immediately generalize to the setting of spaces ﬁbred over B by replacing ‘compact’. S 1 → c. The endpoint of this path is (an . cn . fX ) consisting of a compact Hausdorﬀ space X and a continuous map fX : A → X. and a homotopy H from f to g will be written H : f ⇒ g. one space of maps relative to A. Notation 9. .
fX ) and (Y. Nevertheless this is the notion best suited to our purposes. K. fX ) ≃ (Y. gX ) are homotopy equivalent. fY ) be spaces under A. Deﬁnition 9. L : ψ ◦ fY ⇒ fX such that the two homotopies L · (ψ ◦ K) and G ◦ fX from ψ ◦ φ ◦ fX to fX are homotopic relative to their endpoints. fZ ). fX/T ) be the space under A in which fX/T is the composite of fX with X → X/T . fX ) to (Y. ψ. then (X. H. Thus the relation is transitive. Proof.3. K. homotopies G : ψ ◦ φ ⇒ IdX . G · (ψG′ φ). If X and Y are homotopy equivalent relative to A. ψψ ′ .and a homotopy H : g ⇒ h. K ′ · (φ′ K). Let (X. 3. H ′ .4. L) is a homotopy equivalence from (X. fY ) and (φ′ .5. gX ) are spaces under A with fX ≃ gX . 1. Let (X/T. ∂) are homotopy equivalent in the present sense.6. given spaces (X. See Section 11. ∂) and (c2 . fY ). K ′ . If (φ. but only satisfy a homotopy version of the condition. fZ ). fY ) under A. Lemma 9. 30 H : φ ◦ ψ ⇒ IdY . A homotopy equivalence from (X. Example 9. ψ ′ . and such that the two homotopies K · (φ ◦ L) and H ◦ fY from φ ◦ ψ ◦ fY to fY are homotopic relative to their endpoints. fX ) to (Y. even though there are no maps between them n relative to A = S 1 ⊔ i=1 S 1 . See Lemma 9. fX ) to (Y. Suppose given a space (X. fY ) are homotopy equivalent in the sense above. G. fX ) to (Z. fY ) to (Z.1 Homotopy equivalences of spaces under A In this subsection we will study the following notion of homotopy equivalence between spaces under A. fX ) and (Y. . Then (X. G. If (X. fX/T ) are homotopy equivalent. fX ) under A and a coﬁbrant embedding i : T ֒→ X of a contractible space. fX ) and (X. H ′ · (φ′ Hψ ′ ). we write H ◦ k : g ◦ k ⇒ h ◦ k and l ◦ H : l ◦ g ⇒ l ◦ h and for the induced homotopies between the composites. We write (X. L′ ) is a homotopy equivalence from (Y. fY ) is a 6tuple (φ. G′ . ψ : Y → X. c2 ∈ C(n) the pairs (c1 . L · (ψL′ )) is a homotopy equivalence from (X. For any c1 . and homotopies K : φ ◦ fX ⇒ fY . then (φ′ φ. 9. fX ) and (Y. we will consider maps g : X → Y that do not necessarily satisfy g ◦ fX = fY . fX ) and (X. Lemma 9. L) consisting of maps φ : X → Y. fX ) and (X/T. H. It is somewhat nonstandard in the sense that. 2. Homotopy equivalence is an equivalence relation on spaces under A. then (X. ψ.5 below. The relation is evidently reﬂexive and symmetric. fY ) if there is a homotopy equivalence from (X.
but we give a proof here in order to indicate just how the given assumptions are used. Y ) depends only f on the homotopy equivalence class of (X. By choosing an appropriate f retraction we can form a map γ : (A × I) × I × Maph (X. both equipped with the compactopen topology. m). G. Let G be the homotopy given by f . Thus the evaluation map α : (X ∪fX A × I) × Maph (X. Set Mapf (X. and such that if m ∈ Mapf (X. This restricts to a homotopy equivalence between the subspaces consisting of those elements for which the map g is a homotopy equivalence. t. this H exists because f (I × T ) ⊂ T . Y ) given by taking the constant homotopy. Y ) = {(g. f Proposition 9. 31 . Y ) → f Y that has an adjoint φ : I × Maph (X. H : g ◦ fX ⇒ fY }. Proof. Y ) and Map(X ∪fX A × I. 9. f f By construction. Y ) has image contained in Mapf (X.8. Now β and γ combine to give a map ω : (X ∪fX A×I)×I ×Maph (X. Since X and Y are compact Hausdorﬀ. α(a. ψ. m) = γ((a. 0). Y ) ֒→ Maph (X. H)  g : X → Y. Then (φ. Y ). Y ).Proof. m). Y ) f f has image contained in Mapf (X. fX ) and (Y. Y ) → Y is continuous and extends f to a continuous map β : X × I × Maph (X. Y ) → Y . fY ). Y ). and φI × Mapf (X. 1. The next result shows that the homotopy type of Maph (X. Let f : I × X → X extend the identity map {0} × X → X and a null homotopy I × T → T . Y ) respectively. t). Let (X. Let H be the homotopy whose composite with φ is given by f . Deﬁnition 9. φ{0} × Maph (X. Y ) ֒→ Maph (X. The inclusion Mapf (X. Y ) is the required homotopy f inverse. α(a. Y ) is the identity. t). Let φ : X → X/T be the collapse map and let ψ : X/T → X be induced by f1 = f {1} × X. There is an inclusion Mapf (X. continuous maps (X ∪fX A×I)×K → Y are in bijection with continuous maps K → Map((X ∪fX A × I).2 Spaces of maps between spaces under A Now we introduce two mapping spaces between spaces under A. m) = m(a) is independent of t and s. Y ) → Maph (X. m) = α(a. f These are topologized as subspaces of Map(X. 1. Y ).7. s. Y ) → Y such that f γ((a. K. H. fY ) be spaces under A. fX ) and (Y. t). This is a standard result. φ{1} × Maph (X. m) = γ((a. Y ) = {g : X → Y  g ◦ fX = fY } and Maph (X. Let K to be the constant homotopy and let L be the homotopy obtained from G. L) is the required homotopy equivalence. s. Thus φ{1} × Maph (X. 0. Y ) then γ((a. Y ) is a homotopy f equivalence so long as for any space K the map fX × IdK : A × K → X × K has the homotopy extension property. s. m).
F ) = (φ ◦ g. K. We have the following homotopies relative to endpoints: L · [(ψ ◦ K) · (ψ ◦ φ ◦ F )] ≃ [L · (ψ ◦ K)] · (ψ ◦ φ ◦ F ) ≃ (G ◦ fY ) · (ψ ◦ φ ◦ F ) ≃ F · (G ◦ g ◦ fX ). L) be a homotopy equivalence from (Y. L · (ψ ◦ F )) are g ˜ ˜ inverse homotopy equivalences. K · (φ ◦ F )) and Ψ(˜. 1. ∂ × Id). H. To be precise. using the language of Section 9. F ) → (ψ ◦ φ ◦ g. relative to ∂. Y ) f f ˜ given by Φ(g. Moreover. fY ) ˜ ˜ to (Y .Proposition 9. For Ψ ◦ Φ is the map (g. local trivializations ∼ Rf D◦ (n)U = U × a can be chosen compatible with the boundary maps.1. as does the second part of the proposition.9. ψ. the second comes from our assumption on the two homotopies φ ◦ ψ ◦ fY ⇒ fY . We prove that each a ∈ ◦ f D (n) has a neighbourhood U over which (Rf D◦ (n). This restricts to a homotopy equivalence between the subspaces consisting of homotopy equivalences. Y ). and the third is obtained by using G and F simultaneously. as required. ∂) is isomorphic to (a × f D◦ (n).8. 10 Let The ﬁbrewise structure of Rf D◦ (n) n ∂: i=0 S 1 × f D◦ (n) → Rf D◦ (n) n denote the combined boundary map ∂out ⊔ i=1 ∂i . It follows that Ψ ◦ Φ is homotopic to (g. 2. F · (G ◦ g ◦ fX )). In this section we will study the ﬁbrewise properties of Rf D◦ (n) over f D◦ (n). Y ) → Maph (X. The ﬁrst of these is a reparametrization. G. We will show that Ψ ◦ Φ ≃ Id. fX ). fY ). and in this setting we will consider n (Rf D◦ (n). Proof. Let (φ. we prove that (Rf D◦ (n). This is homotopic to the identity. we will work in the setting of spaces ﬁbred over f D◦ (n) (and its open subsets). ∂) satisﬁes the hypothesis of Proposition 9. 1) × f D◦ (n) ֒→ Rf D◦ (n). Moreover. F ) = (ψ ◦ g . Proposition 10. fX ) to (X. Then the two maps h ˜ Φ : Mapf (X. Y ) → Maph (X. Rf D◦ (n) → f D◦ (n) is a ﬁbre bundle. The analogous result holds ˜ ˜ for homotopy equivalences from (X. f ˜ Ψ : Maph (X. 32 . That Φ ◦ Ψ is homotopic to the identity follows from the analogous argument. In particular it is a ﬁbrewise coﬁbration. ∂) as a space under i=0 S 1 × f D◦ (n). The combined boundary map ∂ extends to a ﬁbrewise n open embedding i=0 S 1 × [0. L · [(ψ ◦ K) · (ψ ◦ φ ◦ F )]). F ) → (ψ ◦ φ ◦ g.
Our result for cacti is much more involved than its counterpart for framed little discs. and such open subsets cover f D◦ (n). 1) → Rf D◦ (n) i=0 extending ∂. . For each r ∈ {1. beginning with a special case. ∂ × Id) under n S 1 × C(n). In Proposition 11. It is therefore locally a quasiﬁbration. 11 The ﬁbrewise structure of RC(n) The last section studied the ﬁbrewise properties of realizations of framed little discs. We will construct a ﬁbrewise open embedding n S 1 ×[0. r) → ai ([1 + d(a)r]θ) on the ith cofactor. The ﬁrst claim follows. We now turn to the second claim. In a neighbourhood U of a the map (θ. Fix a ∈ f D◦ (1). using the language of Section 9. r) → (1 − r)∂1 (θ) + rφ(θ)) is a homeomorphism S 1 × [0. By combining the two homeomorphisms we obtain a trivialization of Rf D◦ (n) over U . show that RC(n) → C(n) is locally ﬁbrewise homotopy equivalent to a product. . relative to ∂. There is a homeomorphism S 1 × [0. and the homotopies G and H. 1] → a given by (θ. and in this setting we will consider (RC(n). ∂) and (c × C(n). Let n ∂: i=0 S 1 × C(n) → RC(n) n denote the combined boundary map ∂out ⊔ i=1 ∂i . for each a ∈ ◦ f D (n). n} choose closed intervals Ir ⊂ Ir ⊂ S 1 ¯r . such that ∂r (Ir ) does not meet the image ¯ such that Ir lies in the interior of I 33 . Then there is an open neighbourhood i=0 U of c over which (RC(n). 1) such that.1. 11. We will study the ﬁbrewise properties of RC(n) over C(n).2. for i > 0. ∂) and (c × C(n). The required embedding is then given in the ﬁbre over a by (θ. Proof. .1 ¯ Fix c ∈ C(n). Let a ∈ f D◦ (n).1 the local existence of the maps φ and ψ. This is an open subset of f D◦ (n). To be precise. we will work in the setting of spaces ﬁbred over C(n) (and its open subsets). the claim follows. Extend the ith little disc of a to an aﬃne linear map ai : R2 → R2 . 1] → b. Since maps that are locally quasiﬁbrations are themselves quasiﬁbrations. the little discs of a can be dilated by a factor (1 + d(a)) yet still be disjoint and lie within the interior of the disc of radius (1 − d(a)). Let A ⊂ f D◦ (n) consist of those b for which each little disc of b lies in the interior of the corresponding little disc of a. In this section we turn to the analogous issue for cacti.1 Proof of Proposition 11. RC(n) → C(n) is a quasiﬁbration. The required local trivialization of Rf D◦ (n) near a can now be formed by restricting attention to each little disc of a and applying the result for the case n = 1 given above. r) → [1 − d(a)r]θ on the 0th cofactor and by (θ. Corollary 11. Let φ : S 1 → S 1 denote the framing of a. . ∂) as a space under n 1 i=0 S × C(n). ∂ × Id) are homotopy equivalent. Fix c ∈ C(n) and form the ﬁbrewise spaces (RC(n). Proposition 11. r) → (1 − r)∂1 (θ) + rφ(θ). Find a continuous function d : f D◦ (n) → (0.Proof.
(10) r r In both cases the horizontal maps are the restrictions of r=1 ∂r and the righthand map is the inclusion. The left hand maps of these squares are coﬁbrations. The diagrams (10) also show that the ﬁbrewise spaces (RC(n)U )/P and (c/Pc )× U are ﬁbrewise isomorphic. It is therefore natural to expect that RC(n)U → (RC(n)U )/P is a ﬁbrewise homotopy equivalence with the product ( Ir )/( ∂Ir ) × U . in other words. From the claim and from Lemma 9. ∂1 ) and ((c/Pc ) × U. so the same is true for the right hand maps.1 will proceed along these lines. here the ﬁrst map is the inclusion of the 0th cofactor. Form the ﬁbrewise spaces ((RC(n)U )/P. The composites S 1 × U ֒→ S 1 × U ֒→ n i=0 n i=0 ∂ →(RC(n)U )/P ∼ (c/Pc ) × U. The ﬁbrewise quotient RC(n)/P is given in each ﬁbre by the bouquet of circles ( Ir )/( ∂Ir ). r U × ∂Ir U × Ir n / Pc × U / c × U.6 we obtain the ﬁrst and last homotopy equivalence of (9). where ∂1 and ∂2 are the composites ∂1 : ∂2 : n i=0 n i=0 S 1 × U − RC(n)U → (RC(n)U )/P. It remains to ﬁnd the middle homotopy equivalence. → S 1 × U − − c × U → (c/Pc ) × U. Set P = r ∂r (S 1 \ int(Ir )) ⊂ RC(n)U . ∂2 ) under n 1 i=0 S × U . −→ (9) ∂×Id ∂ We will construct three homotopy equivalences. Thus P is a ﬁbrewise subspace of RC(n)U whose ﬁbre over d is a copy of d with an interval removed from each lobe. Now we make a further claim. and moreover the isomorphism is such that the triangle n i=1 n ∂1  nnnn nn nn wnnn (RC(n)U )/P o S1 × U PPP PPP2  ∂ PPP PP' / (c/Pc ) × U ∼ = commutes. Since homotopy equivalence is a transitive relation. S 1 × U −1 = →(c/Pc ) × U S 1 × U −2 ∂ are homotopic. 34 . ∂) ≃ ((RC(n)U )/P. and such that ∂out (•) does not lie ¯ in any Ir . Our proof of Proposition 11.of any incoming boundary maps besides ∂r . ∂) . P is ﬁbrewise contractible. this will suﬃce to prove the proposition. ∂2 ) ≃ (c × U. ∂1 ) ≃ ((c/Pc ) × U. The triangle says nothing about the value of ∂1 and ∂2 on the ﬁrst n cofactor of i=0 S 1 × U . we may choose a neighbourhood U of c over which the above properties ¯ of Ir ⊂ Ir still hold. Since the lobe coordinates of a cactus depend continuously on the cactus. There are ﬁbrewise pushout squares r U × ∂Ir U × Ir /P / RC(n)U. We make the following claim: 1. 2. each ﬁbre is a tree. (RC(n)U.
Proposition 11. . Q is ﬁbrewise contractible. 1] → Q satisfying h((p. The second claim then follows from Proposition 11. i We say that i ∈ {1. 1] ⊂ S 1 . Thus ((RC(n)U )/P.From this claim and from the commutative triangle it follows that ((RC(n)U )/P. . i−1 i+1 n Assume that for each a ∈ A the space i ψ i ({a} × I) is connected. Deﬁne Q → A to be the ﬁbrewise space Q = i ψ i (A × I) ⊂ A × Rn . xn−1 coincide. . . . ∂2 ) is itself homotopy equivalent to ((c/Pc ) × U. . Leaves exist so long as n 2. 35 . t0 + 1 ǫi vt. . For. . given d ∈ U . so ′ that ((c/Pc ) × U. For the beginning of this proof let us assume that A is a single point and omit it from the notation. Note that the ﬁbres of P are connected. . xi ) for some value of i. 1/2]n−1 and the ∂r Ir are maps [0. a n−1 leaf exists. . where ∂2 is some map homotopic to ∂2 . and it follows that n is a leaf.3. xi (a). . or in other words that there is a ﬁbrewise h : Q ×A Q × [0. . 0) = p and h((p. ∂1 ) is homotopy equivalent to ((c/Pc ) × U. ∂1 ) ′ ′ is isomorphic to ((c/Pc ) × U. . Let π : Rn−1 → Rn−2 be the map that forgets the ﬁnal coordinate. . β] on which γ has the form t → (xi . Then the n−2 x1 . n. . t. Then the lobe coordinates are maps U → [0. some t0 . . q). Then πx1 . . 1] meets exactly one other ψ j [0. . . The second claim states that two maps S 1 × U → (c/Pc ) × U are homotopic. . . as required. xi . any two points of d can be joined by a path that does not pass through any of the Ir . . and some ǫi ∈ {±1} depending only n on i. Since Q obviously admits sections. ∂2 ). xi ). πxn−1 deﬁne a conﬁguration of line segments in Rn−1 . 1] are exactly the xj . 1) = q. Write these as xi = (xi . 1] into intervals [α. . . . . . Note that the ith coordinates of the points where ψ i [0. . n} is a leaf if ψ i [0. since Q is connected. . . . 11. xi (a)).3 below. By construction. We refer to Q as a conﬁguration of orthogonal line segments in Rn . and so have a leaf which without loss is (n − 1). We obtain 1 n i−1 i+1 functions ψ i : A × I → Rn given by i (a. . The ﬁrst claim states that P is ﬁbrewise contractible. 1] → Q is one for which there is v 0 and a decomposition of [0. Proof. ∂2 ). . So if (n − 1) is not a leaf then xn must diﬀer from n−1 n−1 the common value of the xi . . . Suppose we are given functions xi : A → Rn−1 for i = 1. . xi . 1]. . We prove the general case by induction. Rescaling [0. 1/2] to [0. q). . ∂2 ). 1] we see that the ﬁbred space P → U is a conﬁguration of orthogonal line segments in the sense described in the next section. This is clear when n = 2. Let us assume without loss that each Ir ⊂ S 1 is equal to [1/2. It remains to prove our claims.2 Conﬁgurations of orthogonal line segments Let A be a topological space. In either case. 1/2] → [0. 1/2]n. lying in the same homotopy class. The claim follows since the space of based maps S 1 → S 1 in a ﬁxed homotopy class is contractible. . An adapted path γ : [0. both are families (parameterized by U ) of based maps from S 1 to a bouquet of circles. . it will suﬃce to show that it is ﬁbrewise convex. t) → (x1 (a). 1] meets other ψ j [0. xi (a).
1]. We must show that it is continuous. This is an application of the ﬁbrewise version of Proposition 9. Proposition 12. H) where h n a ∈ f D◦ (n). c). 1] × i=0 S 1 → c a homotopy from f ◦ ∂ to ∂. . The inclusion E◦ (n) ֒→ E◦ (n) is a ﬁbrewise homotopyh equivalence of spaces over f D◦ (n) × C(n). c). For ﬁxed i ∈ {1. It is equipped with the compact open topology. we obtain a function h : Q ×A Q × [0. Write x′ for the point where ψ n [0. Uniqueness is proved by a similar induction. h ∂ Proposition 12. y of Q can be joined by a unique adapted path. This is clear when n = 1 and n = 2. c) ֒→ Maph (a. By applying the results just obtained in each ﬁbre. f : a → c is a homotopy equivalence. 12 Proof of Theorem C We can now complete the proof of Theorem C. In the ﬁnal case assume without loss that x lies in ψ n [0. c ∈ C(n). c) ∈ f D◦ (n) × C(n) has a neighbourhood W over which there is a ﬁbrewise homotopy equivalence E◦ (n) ≃ Maph (a. Recall that Map∂ (a. This completes the proof. In other words. Deﬁnition 12. c) ֒→ Maph (a.1. Proposition 12.3. 1] and that y lies in i n−1 ψ i [0. Now remove the restriction that A is a single point. c) denote the ∂ ﬁbre of E◦ (n) over (a. 1]. c) is a homotopy ∂ equivalence. H) with n f : a → c a homotopy equivalence and H : [0. Deﬁnition 12.2. let Maph (a. Every pair (a.Any two points x. Taking the constant homotopy gives an inclusion E◦ (n) ֒→ E◦ (n).5. We topologize E◦ (n) as a subspace of h the ﬁbrewise mapping space. The claim is then immediate if x and y both lie in ψ n [0. y. it is the space of homotopy equivalences f : a → c satisfying f ◦ ∂ = ∂. c) × W. Let E◦ (n) denote the space of quadruples (a. and follows by induction if they both lie in i n−1 ψ i [0. c) denotes the ﬁbre of E(n) over (a. 1]. the assumption holds by the ﬁrst part of Proposition 10.4. it is the space of pairs (f. 1] meets i n−1 ψ i [0. so it suﬃces to prove continuity over Ai . 1]. 1] → c is a homotopy from f ◦ ∂ to ∂. n} the subset Ai of A on which i is a leaf is closed. f. Then an adapted path from x′ to y exists by induction. h Fix a ∈ f D(n) and c ∈ C(n). Continuity over Ai follows from an argument similar to the construction of adapted paths. .1. and H : i=0 S 1 × [0. 1] → Q determined by the fact that each h(x. Given a ∈ f D◦ (n) and c ∈ C(n). For we can assume that n is a leaf. and can easily be modiﬁed to produce an adapted path from x to y. In other words. ∂ The proof of Theorem A is completed by the following three propositions. Proof of Proposition 12. .3. c. . 36 . c). The inclusion Map∂ (a.8. and in general is proved by induction. −) is the adapted path from x to y. Taking the constant homotopy gives an inclusion Map∂ (a. We continue to write ∂ for the n combined boundary map ∂out ⊔ i=1 ∂i .
When A ֒→ B is the inclusion of a subspace we will write the pullback as XA or XA .1. this is always the case if X itself is Hausdorﬀ. It is ﬁbrewise Hausdorﬀ if distinct points in the same ﬁbre can be separated by open sets of X.3. f : a → c is a homotopy equivalence.5 ([6. as in Proposition 11. ∂ × Id) are homotopy equivalent as in Proposition 10. and establish some facts regarding the ﬁbrewise mapping space. Let U be a neighbourhood of a over which (Rf D◦ (n).1. We refer almost entirely to the book of James [6]. A continuous bijection φ : X → Y is in fact a homeomorphism.5 is also proved by an application of Proposition 9. Deﬁnition A.f A with its natural projection. d). ∂) and (a × f D◦ (n).4.2. Given X. This completes the proof.7]). Deﬁnition A. and H : f ◦ ∂ ⇒ ∂ is a homotopy. The ﬁbre of X over b ∈ B is Xb = p−1 (b). f. The ﬁbrewise pushout of a diagram Z ← X → Y of spaces and maps ﬁbred over B is the ordinary pushout Y ∪X Z. A square X Z /Y /W of spaces and maps ﬁbred over B is a ﬁbrewise pushout square if the induced ﬁbrewise Y ∪X Z → W is a homeomorphism. A ﬁbrewise map φ : X → Y between spaces ﬁbred over B is a map X → Y that commutes with the projections. Let X and Y be ﬁbrewise compact Hausdorﬀ spaces. or in other words if p is closed and has compact ﬁbres. Proposition 9. Topologize E′ as a subspace of the ﬁbrewise mapping space. and let V be a neighbourhood of c over which (RC(n). The ﬁbrewise space X is ﬁbrewise compact if p : X → B is proper. ∂) and (c × C(n). together with the previous paragraph. Write E′ for the space of quadruples ((b. Deﬁnition A. Then a ﬁbred space or ﬁbrewise space over B is a space X with a map p : X → B called the projection. Let B be a space. If φ : X → Y is a continuous ﬁbrewise surjection. ∂ × Id) are homotopy equivalent.4. Proposition A. Y ﬁbred over B.Proposition 12. H). 3. where (b. Proof of Proposition 12. A Fibrewise topology This appendix will recall some basic notions of ﬁbrewise topology. 37 . the pullback f ∗ X is the ﬁbred space f ∗ X → A given by X ×p. Given a map f : A → B and a space X ﬁbred over B. Part 8 of Proposition A. then X ⊔ Y and X ×B Y become spaces ﬁbred over B in the obvious way. Set W = U × V .8 shows that E′ is ﬁbrewise homeomorphic to the space given in the statement.9. d) ∈ W . then a ﬁbrewise function ψ : Y → Z is continuous if and only if the composite ψ ◦ φ is continuous.1. Deﬁnition A. regarded as a space ﬁbred over B. implies that E◦ (n)W is ﬁbre homotopy equivalent to h E′ .8.
Then the map MapB (X. The ﬁbrewise mapping space MapB (X. f ). and suppose that in each ﬁbre the square is a pushout.63]). Y. Here W is an open subset of B.6. It is equipped with the topology generated by basic open sets (K. X2 be ﬁbrewise compact Hausdorﬀ spaces over B. 1. W ). Then (K. V . Proposition A. The bijection MapB (X1 ⊔ X2 . 6. Y ) → MapB (X1 . Y × C) sending ((b. and V is an open subspace of YW . Then the square is a ﬁbrewise pushout. c). Z and X1 . Let B be a topological space and let X. f ) to (F (a). 2. Z) → MapB (X. Y ) × C → MapB×C (X × C. f ) for which b ∈ W and f (Kb ) ⊂ Vb .7 ([6. f ) is continuous. If θ above is a ﬁbrewise surjection then θ∗ is an embedding. Y ) sending (a. Y × C as ﬁbred over B × C. f ). Let X and Y be spaces ﬁbred over B. Y ) is the space of pairs (b. Z) is continuous. θ∗ : MapB (Y. Let C be a further space and regard X × C. f ). c) to ((b.Corollary A.8 ([6]). W ) consists of all pairs (b. Y ) ×B MapB (X2 . 5. 3. Deﬁnition A. A ﬁbrewise map h : X ×B Y → Z is continuous if and only if its adjoint ˆ h : X → MapB (Y. a ﬁbrewise function W → V is continuous if and only if the composites Y → V and Z → V are continuous. Take a map F : A → B. Moreover. Then there is a continuous map MapA (F ∗ X. Z). K is a ﬁbrewise compact subspace of XW . V . Let θ : X → Y and φ : Y → Z be ﬁbrewise maps. Y ) is a ﬁbrewise homeomorphism. F ∗ Y ) → MapB (X. Then there are ﬁbrewise maps φ∗ : MapB (X. Suppose given a commutative square X Z /Y /W of ﬁbrewise compact Hausdorﬀ spaces ﬁbred over B. 38 . Z). where b ∈ B and f : Xb → Yb is continuous. p. 4. Y ) → MapB (X.
With the graphical assistance of Olivier Dodane. Thus KW1 ×W2 \ J × W2 is ﬁbrewise compact over W1 ×W2 and in particular its image in W1 ×W2 is closed. [7] Christian Kassel and Vladimir Turaev. New York. volume 91 of Cambridge Tracts in Mathematics. Constructions of En operads. Cambridge. CRM Barcelona. Basel. In String topology and cyclic homology. All references in this proof are to [6].. 1998. Cohen and Alexander A. 159(2):265–285. Bargheer. Courses Math. 2005. 2. Getzler. Topol. Geom. Now using 3. James. 1989. 39 . But this image does not contain (b. Part 4 is an immediate consequence of the deﬁnitions. Preprint. Let P and Q be topological spaces. 9.7. Cambridge University Press. a [4] Z. Kaufmann. pages 1–95. 2008. [5] E. Braid groups. Birkh¨user.c) ⊂ J and J b ⊂ f −1 (V1 ). B × Q) ∼ B × Map(P. Fiedorowicz. with P compact Hausdorﬀ. W ) in MapB×C (X × C. [6] I. String topology.12 guarantee that a ﬁbrewise compact ﬁbrewise Hausdorﬀ space is also ﬁbrewise locally compact and ﬁbrewise regular. Note that 3. Algebr. Springer. Now the open set (J. W ) as required.7 and 9. c). 1994. assume that W = W1 × W2 and V = V1 × W2 . c) whose image is contained in (K. We will ﬁnd a neighbourhood of ((b.. and continuity of its inverse follows from part 7. Voronov. 5 and 6 are found in p. References [1] T. University of Copenhagen. [3] Ralph L. available from arXiv:9911159. [8] Ralph M. c) and its image is contained in (K.14 and reducing W1 if needed. To do this we will reduce (K. For part 6 take a standard open set (K. V . Notes on string topology. Q) = is a homeomorphism. Phys. By reducing W and V . f ). B × Q) → B × Map(P. V1 . Proof. where J is ﬁbrewise compact over W1 . Adv. 9. BatalinVilkovisky algebras and twodimensional topological ﬁeld theories. W1 ) × W2 contains ((b. and so by reducing W1 and W2 we may assume that KW1 ×W2 ⊂ J × W2 ⊂ J × W2 . V . f ). M. f ). W ) many times. V . Math. continuity of f : MapB (B × P. 2008. Cultivating operads in string topology. Then parts 1.c) is compact. 5:237–300 (electronic). Preprint. V . ﬁnd an open subset J of XW1 such that K(b. c) of its preimage. this can be done because K(b. Comm. Y × C) and an element ((b.69. 3.13 respectively. On several varieties of cacti and their relations. volume 247 of Graduate Texts in Mathematics. [2] Moira Chas and Dennis Sullivan. Then the isomorphism of sets MapB (B × P. available from arXiv:9808089.6. 2006. 1999.22 and the comments after 3. 9.4. Master’s Thesis. Fibrewise topology. For the ﬁnal part. Q) follows from part 6. W ).
[13] Alexander A. [10] James E. [12] Paolo Salvatore. Geom. Math. Pure Math. pages 81–103. Notes on universal algebra. PQ. 2009.[9] J. Cosimplicial objects and little ncubes. Algebr. pages 91–101... Soc. 126(5):1109–1153. 9(1):237–264. 1990. 2009.1883. To appear in Trans. volume 73 of Proc. Weak equivalences and quasiﬁbrations. Math. Providence. Amer. I. 2004. Sympos. P. McClure and Jeﬀrey H. Springer. [11] Luc Menichi. In Graphs and patterns in mathematics and theoretical physics. A BatalinVilkovisky algebra morphism from double loop spaces to free loops. In Groups of selfequivalences and related topics (Montreal. Berlin. The topological cyclic Deligne conjecture. J. Math. Smith. Soc.. Amer. see also arXiv:0908. RI. Topol. 40 . volume 1425 of Lecture Notes in Math. 2005.. 1988). Voronov. Amer... May.
This action might not be possible to undo. Are you sure you want to continue?