Little Cube Algebras & Factorization Homology
Little Cube Algebras & Factorization Homology
COURSE NOTES
YONATAN HARPAZ
Contents
1 Introduction 2
2 Preliminaries on ∞-categories 4
2.1 Two models for (∞, 1)-categories 5
2.2 Constructions of ∞-categories 10
2.3 Left and right fibrations 16
2.4 Cartesian and cocartesian fibrations 23
2.5 The relative nerve 30
2.6 Limits and colimits in ∞-categories 33
2.7 Kan extensions 36
3 Symmetric monoidal ∞-categories 39
3.1 Introduction 39
3.2 Examples and constructions 43
3.3 cartesian and cocartesian symmetric monoidal structures 46
4 ∞-Operads and their algebras 50
4.1 From colored operads to ∞-operads 50
4.2 Weak ∞-operads and approximations 58
4.3 Tensor products of ∞-operads 65
5 The little cube ∞-operads 69
5.1 Definitions and basic properties 69
5.2 Dunn’s additivity theorem 72
5.3 May’s recognition principle 79
6 Factorization homology 88
6.1 Manifolds and framings 88
6.2 Little cube algebras with tangent structures 96
6.3 Little cube algebras over manifolds 103
6.4 Factorization homology 107
6.5 Axiomatic characterization of factorization homology 113
7 Poincaré-Koszul duality 115
7.1 Factorization homology with support 115
7.2 En -suspension and En -loops 115
7.3 Nonabelian Poincaré duality (after Lurie) 116
7.4 Poincaré-Koszul duality for dg-algebras (after Francis and Ayala) 116
References 116
1
2 YONATAN HARPAZ
1. Introduction
In the field of algebra we often encounter objects described as sets, equipped
with certain operations, which are required in turn to satisfy certain axioms. For
example, a monoid is given by a set with one binary operations, which is associative
and unital. When every element acts invertibly we say that the monoid is a group.
In modern mathematics it is very useful to consider such algebraic structures on
objects more general than sets. For example, if we replace sets by smooth manifolds
we get the notion of a Lie group. If instead we consider algebraic varieties we get
algebraic groups. In order to properly define such notions it is useful to describe
the algebraic structure diagrammatically, that is, as certain arrows in the category
we choose to work in. For example, a binary structure on a object X can be
represented by an arrow X × X Ð→ X. This makes sense whether X is a set, an
algebraic variety, or something more exotic, as long as we know what the product
of two objects are. This product can be something that behaves quite differently
from the usual cartesian product of sets. For example, if we work in the category
of abelian groups, and take the product to be the tensor product operation, then a
monoid object will now be a ring.
To make this approach work we need a convenient way to encode algebraic
structures that is independent of the specific type of objects in which we want to
realize it. A successful framework for doing so is by using the notion of an operad.
An operad encodes the information of an algebraic structure by specifying in a
suitable way what are the operations, and what rules they are required to follow.
For example, the operad that encodes the structure of a monoid is usually called
the associative operad. The operad that encodes the structure of a commutative
monoid is called the commutative operad.
In the realm of homotopy theory this approach encounters a new subtlety. This
is because operads encode the algebraic structure using strict rules that the ope-
rations must obey to (such as associativity), but when dealing with objects with
a homotopical nature (such as spaces, chain complexes, or even categories), we
actually need to let our algebraic axioms hold only up to homotopy, in a suitable
delicate sense where all homotopies need to be specified compatibly (this is usu-
ally called up to coherent homotopy). For example, consider a topological space
X and let x ∈ X be a point. The fundamental group of X is defined to be the
group whose elements are homotopy classes of paths in X from x to itself, where
the group operation is given by concatenation. This invariant is very important in
algebraic topology. However, to some extent what is more interesting is the object
that you get without identifying homotopical paths, that is, the topological space of
all paths from x to x. We would like to say that this is a topological group with the
operation of concatenation. Alas, a direct examination shows that concatenation
is not, strictly speaking, associative. It turns out, however, that it is associative up
to (coherent) homotopy.
In order to encode algebraic structures in which the rules only hold up to cohe-
rent homotopy one can use the homotopy theoretical avatar of operads, which are
called ∞-operads. In the passage from operads to ∞-operads some new phenomena
arises. One such phenomenon, which is the center of the course, is the fact that
commutativity now comes in many flavors. More precisely, between the classical
associative and commutative operads there is now an infinite tower of intermediate
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 3
operads
Ass = E1 , E2 , E3 , ....
which ”converge” in a suitable sense, to the commutative operad E∞ = Com. The
operad En is known as the little n-cube operad. This hierarchy of commutativity
levels can already be seen when considering monoid structures on categories (which
have a certain limited level of homotopyness): between the notion of a monoidal
category (which is an E1 -structure) and a symmetric monoidal category (which is
an E∞ -structure) there is an an intermediate option known as a braided monoidal
structure, which corresponds to an E2 -structure. In spaces, the typical examples of
En -monoids are the n-fold loop spaces. May’s recognition principle (see also work
of Boardman and Vogt) states that these are exactly the group-like En -monoids in
spaces.
We will start the course by recalling some homotopy theoretical preliminaries
surrounding the notion of ∞-categories. We will then discuss symmetric monoi-
dal ∞-categories and ∞-operads, setting up the stage for the study of algebraic
structures in a homotopy theoretical setting. We will then introduce and study the
main object of interest in this course, the little n-cube operad. We will prove a key
result, known as Dunn’s additivity theorem, which roughly states that specifying an
En -structure is equivalent to specifying n commuting E1 -structures.
One of the interesting features of En -algebras are their relations to the topology
of n-manifolds. In particular, given an En -algebra A with a suitable equivariance
structure (taking values in a sufficiently nice ∞-category) and an n-manifold M
with a corresponding tangential structure, we may integrate A along M to obtain
an object ∫M A, known as the factorization homology of M with coefficients in A.
We may think of these construction as either producing homological invariants of
manifolds out of En -algebras, or as producing homological invariants of En -algebras
out of manifolds: both point of views yield interesting insights. In addition, in
the former point of view the resulting homology theories for manifolds can be
characterized axiomatically, as we will describe towards the end of the course,
following the approach of D. Ayala and J. Francis.
We note that any abelian group A can be considered as an En -monoid in spaces.
The factorization homology ∫M A is then simply the chain complex C∗ (M, A) with
coefficients in A, whose homologies are H∗ (M, A). If we replace A by a general
En -monoid is spaces X then we get a nonabelian generalization of classical homo-
logy. When X is group-like, May’s recognition principle gives that X ≃ Ωn Y for
some space Y . In this case, Lurie’s nonabelian Poincaré duality states that ∫M X is
naturally equivalent to the mapping space from M to Y , that which can be consi-
dered as a generalization of cohomology to nonabelian coefficients. We will outline
the mathematics needed to establish this theorem in the final parts of the course,
following the ideas of Lurie.
The passage from X to its n-fold delooping Y can itself be considered as a
variant of factorization homology: it is obtained by taking the reduced factorization
homology of X along the n-sphere. This operation can be done for En -algebras
taking values in any nice ∞-category C, as long as they admit an augmentation
(which is what makes reduced homology make sense). This procedure takes an
augmented En -algebra and yields an augmented En -coalgebra. On the other hand,
given an En -coalgebra, we may take its reduced factorization cohomology along
the n-sphere and obtain again an En -algebra. In spaces this reproduces the looping
4 YONATAN HARPAZ
2. Preliminaries on ∞-categories
The point of departure of modern algebraic topology from traditional point-set
topology is arguably the introduction of homotopies between maps. This has some
remarkable consequences: it yields a notion of a homotopy equivalence between to-
pological spaces, leads to the focus on invariants which respect this new equivalence
relation, such as homotopy groups and homology groups, and demands modifying
various non-homotopical construction to accommodate homotopy. This last point
may seem surprising. If the notion of a homotopy between maps is natural from a
topological point of view, how come many natural topological constructions do not
respect homotopy equivalence? for example, if f ∶ X Ð→ Y is a continuous map
and y ∈ Y is a point, then the fiber f −1 (y) ⊂ X is not a homotopy invariant notion:
it is not stable under a continuous deformation of f or y, nor under replacing X
or Y by homotopy equivalent spaces. A modern answer to this question is that
the introduction of homotopies between maps means that we are no longer working
in the category of topological spaces we thought we were working in. Indeed, this
would explain why some constructions don’t work anymore: these constructions
are usually categorical, given by various limits and colimits, and hence should not
be expected to make sense once the category we are working with changes.
But then in what category are we working in? One potential answer is that
we should consider the homotopy category of spaces. This is the category whose
objects are spaces and whose morphisms are homotopy classes of maps. This point
of view is powerful, but has some serious drawbacks. Besides the general feeling
that some crucial information got lost (what about homotopies between homotopies,
homotopies between homotopies between homotopies, and so on?), working with
the homotopy category does not explain the modifications that various categories
constructions need in order to become homotopical. For example, the notion of
the fiber f −1 (y) of a map f ∶ X Ð→ Y is replaced in algebraic topology with
the notion of the homotopy fiber of f , which is the space of pairs (x, η) where
x is a point of X and η is a continuous path in Y from f (x) to y. This is an
example of switching from a limit to its corresponding homotopy limit, and yields a
satisfyingly homotopy invariant replacement. It is not, however, the corresponding
limit in the homotopy category of spaces. In fact, the homotopy category of spaces
does not admit limits and colimits in general! If we choose the homotopy category
as our categorical framework when studying spaces, then we will need to accept
that the vast majority of our homotopically modified constructions, including all
homotopy limits and colimits, are not determined by the underlying categorical
structure, and thus need to be considered as additional structure. This is of course
a valid approach, which in fact leads to Grothendieck’s theory of derivators, though
conceptually it is not very satisfying, at least if we wish to hold on to the idea that
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 5
Exercise 2.1.6. Prove that the nerve functor N ∶ Cat Ð→ Set∆ is fully-faitful and
its essential image consists of those simplicial sets for which every diagram as (2.1)
admits a unique dotted extension.
The above exercise tells us that we can think of ordinary categories as a par-
ticular type of simplicial sets, and furthermore that this type of simplicial sets is
characterized by a certain unique lifting condition. The notion of an ∞-category is
then obtained by removing the uniqueness part of this condition.
Given an ∞-category C we will call the vertices of C its objects, and if we have
an edge e ∶ ∆1 Ð→ C then we will call it a morphism from x ∶= e∣∆{0} to y ∶= e∣∆{1} ,
in which case we will also use diagramatic notation and write this edge as an arrow
x Ð→ y. For a subset S ⊆ {0, ..., n} we use the notation ∆S ⊆ ∆n to denote the ∣S∣-
dimensional face of ∆n whose vertices are S. Given an object x ∈ C we will consider
the degenerate edge s(x) on x as the identity Idx ∶ x Ð→ x. Given a 2-simplex
σ ∶ ∆2 Ð→ C we will depict it diagramatically as
(2.3) y
? ??? g
f
??
??
?
x
h /z
where x, y, z are the vertices obtained by restricting σ to ∆{0} , ∆{1} and ∆{2}
respectively, and f, g, h are the morphisms obtained by restricting σ to ∆{0,1} , ∆{1,2}
and ∆{0,2} respectively. We can think of σ as designating a homotopy from g ○ f
to h. However, it is important to note that in an ∞-category there isn’t a specific
edge which is g ○ f . Instead, we may consider any triangle of the form (2.3) to
be exhibiting h as the composition of f and g. In particular, there could be two
different triangles σ, σ ′ ∶ ∆2 Ð→ C of the form (2.3), with different edges σ∣∆{0,2} =
h ≠ h′ = σ∆ ′ ′
{0,2} . However, in this case the edges h and h will be homotopic in
(2.4) x@
~~~~~~ @@@h′
~~~~
Id
@@
~~~~ @
x
h /z
which we will consider as a homotopy from h′ to h. In this sense, while the compo-
sition of two morphisms in an ∞-category is not strictly speaking uniquely defined,
it is uniquely defined up to homotopy. Elaborating on this argument one can show
that the collection of compositions of f and g can be organized into a simplicial set
which is a contractible Kan complex (see Remark 2.3.11 below). We may thus say
that composition of two morphisms in an ∞-category is essentially defined. This
essentially defined composition is associative in the following sense: if f ∶ x Ð→ y,
g ∶ y Ð→ z and h ∶ z Ð→ w are three arrows in C then we can compose the three of
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 9
them in two different ways. We can either choose triangles of the form
(2.5) >y@
~~~ @@@ g
f
~ @@
~~ @@
~~ t / z
xA
AA }
AAs h }}}
AA }
A ~}}}
w
exhibiting t as the composition of f and g and s as the composition of t and h, or
choose triangles
(2.6) >y@
~~~ @@@ g
f
~ @@
~~ @@
~~
xA t′ z
AA ′ }
AAs h }}}
AA }
A ~}}}
w
exhibiting t′ as the composition of g and h and s′ as the composition of f and
t′ . We claim that in this case s is homotopic to s′ . To see this, it will suffice
by the argument before to show that there exists a triangle in C exhibiting s as
the composition of f and t′ . To find such a triangle we note that the two triangles
in (2.5) together with the right triangle in (2.6) together determine a map ρ ∶ Λ32 Ð→
C. Since C is an ∞-category we can extend ρ to ρ ∶ ∆3 Ð→ C. The triangle ρ∣∆{0,1,3}
then exhibits s as the composition of f and t′ .
The above considerations concerning the essential well-definition and associati-
vity up to homotopy of composition of arrows in an ∞-category have in particular
the following outcome: we can define an ordinary category Ho(C) whose objects
are the vertices of C and such that HomHo(C) (x, y) is the set of homotopy classes of
arrows from x to y. This category is known as the homotopy category of C. There
is an evident map C Ð→ N(Ho(C)) which sends each arrow to its homotopy class.
Definition 2.1.7. Let C be an ∞-category and f ∶ x Ð→ y an arrow in C. We will
say that f is invertible if there exists an arrow g ∶ y Ð→ x and triangles as in (2.3)
exhibiting Idx as the composition of f and g and Idy as the composition of g and f .
In this case we will also say that f is an equivalence in C from x to y. We will say
that an object x ∈ C is equivalent to y ∈ C if there exists an invertible arrow from
x to y. In this case we will write x ∼ y. We note that, essentially by the definition
of Ho(C), an arrow is invertible if and only if it maps to an isomorphism in Ho(C),
and two objects are equivalent if and only if they become isomorphic in Ho(C).
Definition 2.1.8. Let X, Y be two simplicial sets. We will denote by Y X the
simplicial set such that
(Y X )n = HomSet∆ (∆n × X, Y ).
We will call Y X the mapping simplicial set from X to Y .
Proposition 2.1.9 (Joyal). If C is an ∞-category then for every K ∈ Set∆ the
mapping simplicial set CK is again an ∞-category.
10 YONATAN HARPAZ
by MapC(∆n ) (i, j) = N(P(i, j)) where P(i, j) ⊆ Sub({0, ..., n}) is the poset of sub-
sets of {0, ..., n} whose minimal element is i and whose maximal element is j (in
particular P(i, j) = ∅ if i > j). The composition rule is induced by the poset map
P(i, j) × P(j, k) Ð→ P(i, k) which takes (A, B) to A ∪ B.
The association [n] ↦ C(∆n ) determines a functor from the category ∆op to the
category Cat∆ of simplicial categories. Since Set∆ = Fun(∆op , Set) is a category of
presheaves (of sets), the functor C ∶ ∆ Ð→ Cat∆ admits a unique colimit preserving
extension
C ∶ Set∆ Ð→ Cat∆ C(X) = colim C(∆n )
n∆ →X
We may hence choose any one of these models in order to construct our ∞-category
of spaces. If we start from Top then we can do this by defining the simplicial
category CW whose
- objects are the CW-complexes;
- for two CW-complexes X, Y the simplicial mapping set MapCW (X, Y ) is given
by MapCW (X, Y )n = HomTop (∣∆n ∣ × X, Y ).
On the other hand, if we start from the Kan-Quillen model structure on simplcial
sets then we can consider instead the simplicial category Kan whose
- objects are the Kan complexes;
- for two Kan complexes X, Y the simplicial mapping set MapKan (X, Y ) is given
by MapKan (X, Y )n = HomSet∆ (∆n × X, Y ).
Both CW and Kan are locally Kan simplicial categories. Furthermore, the functors
Sing and ∣ − ∣ are both Dwyer-Kan equivalence
≃ ≃
Sing ∶ CW Ð→ Kan ∣ − ∣ ∶ Kan Ð→ CW,
and so it makes essentially no difference which option we choose for our model. In
modern homotopy theory, it is often customary to take Kan.
Definition 2.2.10. We define the ∞-category of spaces
S ∶= N(Kan)
to be the coherent nerve of the simplicial category of Kan complexes.
Remark 2.2.11. Technically speaking, S is a large ∞-category, that is the set of
n-simplices Sn is a proper class. However, this ∞-category is locally small, that is,
its mapping spaces are all small.
In a similar fashion we can now define the ∞-category Cat∞ of (small) ∞-
categories. Let QC be the simplicial category whose objects are the ∞-categories
and such that for each C, D ∈ QC we have MapQC (X, Y ) = Fun(X, Y )≃ , the maximal
∞-groupoid of Fun(X, Y ). Then QC is locally Kan by Theorem 2.1.12, and we
define
Cat∞ ∶= N(QC)
to be its coherent nerve. We will refer to Cat∞ as the ∞-category of ∞-categories. If
C, D are ∞-categories then we will usually denote MapQC (C, D) simply by Map(C, D).
We will now discuss a second major source of ∞-categories, which arise as locali-
zations of ordinary categories by a collection of weak equivalences. More generally,
one can localize ∞-categories by a collection of edges. Let us recall the definition.
Definition 2.2.12. Let C be an ∞-category and let W be a collection of arrows
in C. Let f ∶ C Ð→ D be a functor which sends every arrow in W to an invertible
edge in D. We will say that f exhibits D as the localization of C with respect to W
if for every ∞-category E the induced functor
Map(D, E) Ð→ Map(C, E)
identifies Map(D, E) with the subspace of Map(C, E) consisting of those functors
C Ð→ E which send every edge in W to an equivalence in E. In this case we will
also write D ≃ C[W −1 ].
The following variant of Definition 2.2.12 is also terminologically useful:
14 YONATAN HARPAZ
2.3. Left and right fibrations. Let X be a nice topological space. A classical
result in algebraic topology asserts an equivalence between the category of covering
spaces Y Ð→ X and the category of functors Π1 (X) Ð→ Set, where Π1 (X) is the
fundamental groupoid of X, i.e., the groupoid whose objects are the points in X and
whose morphisms are homotopy classes of paths. A similar phenomenon happens
in ordinary category theory. Recall that a functor π ∶ C Ð→ D of ordinary categories
is said to be left fibered in sets if for every x ∈ C and every morphism f ∶ π(x) Ð→ y
in D there exists a unique arrow f̃ ∶ x Ð→ ỹ in C which maps to f . The notion of a
functor right fibered in sets is similarly defined using lifts of morphisms y Ð→ π(x).
One then has a classification of functors left fibered in sets over a fixed category
D, analogous to the topological story of covering spaces: they correspond exactly
to functors D Ð→ Set from D to sets. In particular, given π ∶ C Ð→ D we may
construct the corresponding functor D Ð→ Set by associating to each y ∈ D the
set π −1 (y) of objects of C lying above y. If f ∶ y Ð→ y ′ is map in D then for each
x ∈ π −1 (y) there exists a unique lift f̃ ∶ x Ð→ x′ of f starting from x. We may hence
associate to f the map of sets π −1 (y) Ð→ π −1 (y ′ ) given by x ↦ x′ . This association
preserves composition by the uniqueness of lifts. In the other direction, if we start
from a functor F ∶ D Ð→ Set then we can associate to F a functor π ∶ ∫D F Ð→ D left
fibered in sets where ∫D F is the category whose objects are pairs (y, a) where y is
an object of D and a is an element of F(y). A morphism from (y, a) to (y ′ , a′ ) is a
map f ∶ y Ð→ y ′ in D such that f! (a) = a′ (where we denoted by f! ∶ F(y) Ð→ F(y ′ )
the map associated to f by F). The category ∫D F is also known as the category of
elements of F, and in a more general context as the Grothendieck construction of
F. In a dual manner, the notion of a functor right fibered in sets will correspond
to contravariant functors from D to Set.
In this section we will discuss the ∞-categorical avatar of this story, which are
known as left and right fibrations. We first note that since we are replacing cate-
gories with ∞-categories, we will naturally want to replace the notion of sets with
that of spaces, or more precisely, ∞-groupoids. We hence cannot expect to have a
definition in which we require a lift to be unique, but only unique up to equivalence.
Even more, we may want something such as unique up to a contractible space of
choices. Let us see how such a definition can be made in the setting of ∞-categories.
of the form
(2.9) Λni /X
_ |>
|
| π
|
∆n /Y
with 0 ≤ i < n. Dually, we will say that π is a right fibration if is has the right lifting
property with respect to the horn inclusions Λin ⊆ ∆n such that 0 < i ≤ n.
We would like to explain how one can think of the lifting condition in 2.9 with
i = 0 as a kind of “unique lifting property” for arrows with a fixed domain. For
this, we will need a construction of the ∞-category of arrows with a fixed domain.
Definition 2.3.2. Let X, Y be two simplicial sets. The join X ⋆ Y of X and Y is
the simplicial set given by
(X ⋆ Y )n = ∐ Xi × Yn−1−i ,
i=−1,...,n
Example 2.3.6 (over ∞-categories). Given an object x ∈ C, the ∞-category C/x can
be described as follows: the objects of C/x are maps f ∶ z Ð→ x from some z ∈ C to
x. A map in C/x from f ∶ z Ð→ x to f ′ ∶ z ′ Ð→ x is given by a triangle σ ∶ ∆2 Ð→ C
of the form
(2.10) z?
g
/ z′
?? ~ ~
?? ~~
f ??
? ~~ f ′
~~
x
which we consider as a map g ∶ z Ð→ z ′ together with a homotopy exhibiting f as
the composition of g and f ′ . We will call C/x the ∞-category of objects over x.
Similarly, we will call the ∞-category Cx/ the ∞-category of objects under x.
Remark 2.3.7. If C is an ordinary category and x ∈ C is an object then N(C)/x
coincides with the nerve of the ordinary category C/x of objects over x.
We are now ready to explain how Definition 2.3.1 can be considered as an essen-
tially unique lifting property for arrows with a fixed domain. Let π ∶ C Ð→ D be a
map of ∞-categories. Identifying (∂∆n )◁ ≅ Λn+1 0 and (∆n )◁ ≅ ∆n+1 we see that
π satisfies the right lifting property with respect to Λn+1 0 ⊆ ∆n+1 if and only if for
every x ∈ C, the dotted lift exists in any diagram of the form
(2.11) ∂∆ n_ / Cx/
w;
w
w π
w
w
∆n / Dπ(x)/
In other words, if and only if the map Cx/ Ð→ Dπ(x)/ is a trivial Kan fibration for
every x. We interpret this as the ∞-categorical analogue of the unique arrow lifting
property: it says in particular that the fiber Cx/ Ð→ Dπ(x)/ over a fixed arrow
f ∶ π(x) Ð→ y is a contractible Kan complex.
We will see in §2.4 that, in analogy with the situation in ordinary category theory,
left fibrations over a fixed ∞-category D essentially correspond to functors from D
to ∞-groupoids, where the functor corresponding to a left fibration π ∶ C Ð→ D is
given informally by the “formula” y ↦ π −1 (y). Similarly, right fibrations correspond
to contravariant functors from D to ∞-groupoids. For now, let us focus on concrete
constructions and examples of left (and right) fibrations. We note that in ordinary
categories, a canonical example of a functor D Ð→ Set is given by the functor
corepresented by an object x ∈ D, that is, the functor y ↦ HomD (x, y). The
category of elements of this functor is simply the category ∫D HomD (x, −) ≅ Dx/
of objects under x, where the projection to D given by [x Ð→ y] ↦ y. If D is
an ∞-category then we have the ∞-categorical construction of the under category
Dx/ described in Example 2.3.6. We would like to show that this indeed results
in a left fibration Dx/ Ð→ D. More generally, we will show that for every diagram
p ∶ K Ð→ C the projection Cp/ Ð→ C is a left fibration (and similarly that C/p Ð→ C
is a right fibration).
Recall the following terminology: we will say that a class of maps in Set∆ is we-
akly saturated if it closed under pushouts (along any map), transfinite compositions
and retracts.
Definition 2.3.8. We will say that a map of simplicial sets X Ð→ Y is
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 19
(1) inner anodyne if it belongs to the smallest weakly saturated class of maps
generated by Λin ↪ ∆n for 0 < i < n.
(2) left anodyne if it belongs to the smallest weakly saturated class of maps gene-
rated by Λin ↪ ∆n for 0 ≤ i < n.
(3) right anodyne if it belongs to the smallest weakly saturated class of maps ge-
nerated by Λin ↪ ∆n for 0 < i ≤ n.
We note that the condition of being a left (resp. right) fibration is equivalent to
the condition of having the right lifting property with respect to all left (resp. right)
anodyne maps. Similarly, the condition of being an inner fibration is equivalent to
having the right lifting property with respect to all inner anodyne maps.
Lemma 2.3.9. Let f ∶ A0 ↪ A and g ∶ B0 ↪ B be inclusions of simplicial sets and
consider the map
h ∶ A0 ⋆ B ∐ A ⋆ B0 Ð→ A ⋆ B.
A0 ⋆B0
and
(2.13) ∆m ⋆ Λnj ∐ ∂∆m ⋆ ∆n Ð→ ∆m ⋆ ∆n
∂∆m ⋆Λn
j
are inner anodyne when 0 < i ≤ n and 0 ≤ j < n. But (2.12) is just the horn inclusion
Λn+m+1
i Ð→ ∆n+m+1 and (2.13) is the horn inclusion Λm+n+1m+j+1 Ð→ ∆
m+n+1
. Since
both i and m+j +1 are strictly between 0 and m+n+1 we get that (2.12) and (2.13)
are inner anodyne, as desired.
Since π is a left fibration we can lift this triangle to a triangle in C of the form
?y?
??? ̃g
f
??
??
Id /
x x
for some ̃ g ) = g is
g . It follows that f admits a left homotopy inverse in C. Since π(̃
an equivalence in D the same argument shows that ̃ g has a left homotopy inverse
f ′ in C. In particular, the image of ̃ g in Ho(C) has both a left and a right inverse
and is hence an isomorphism there. It follows that the image of f in Ho(C) is an
isomorphism and so f is invertible in C.
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 21
We will hence often simplify notation and denote either one of the above spaces
simply as MapC (x, y).
The definition of mapping spaces also allows for the following definition:
Definition 2.3.18. We will say that a map of ∞-categories ϕ ∶ C Ð→ D is fully-
faithful if it induces an equivalence MapC (x, y) Ð→ MapD (ϕ(x), ϕ(y)) for every
x, y ∈ C.
Remark 2.3.19. The condition that ϕ ∶ C Ð→ D is an equivalence of ∞-categories
(Definition 2.1.10) is in fact equivalent to the condition that ϕ is fully-faithful
and essentially surjective (that is, every object in D is equivalent to an obejct
in the image of ϕ). This is not evident from Definition 2.1.10, which is phrased
in terms of the existence of an inverse functor. To show this one can use the
Quillen equivalence between the categorical model structure and the Dwyer-Kan
model structure on simplicial categories. In particular, ϕ is an equivalence in the
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 23
(2.20) x
g
/z
?
f
y
of g through f whose image in D is (2.19).
Exercise 2.4.2. Let F ∶ D Ð→ Cat be an D-indexed diagram of categories and let
π ∶ ∫D F Ð→ D be the projection from the Grothendieck construction as above.
Show that an arrow (f, ϕ) ∶ (x, a) Ð→ (y, b) in ∫D F is π-cocartesian if and only if
ϕ ∶ f! a Ð→ b is an isomorphism.
Definition 2.4.3. Let π ∶ C Ð→ D be a functor of ordinary categories. We will say
that π is a cocartesian fibration if for every object x ∈ C and morphism f ∶ π(x) Ð→
y, there exists a π-cocartesian morphism f̃ ∶ x Ð→ ỹ lying above f .
Remark 2.4.4. The definitions of cocartesian edges and cocartesian fibrations can
be dualized in an obvious manner. In particular, given a functor π ∶ C Ð→ D with
opposite π op ∶ Cop Ð→ Dop , an arrow f ∶ x Ð→ y in C is π-cartesian if and only if it
is π op -cocartesian when considered as an arrow in Cop , and π is a cartesian fibration
exactly when π op is a cocartesian fibration.
Example 2.4.5. If F ∶ D Ð→ Cat is a D-diagram of categories then the projection
π ∶ ∫D F Ð→ D is a cocartesian fibration: indeed, for every (x, a) ∈ ∫D F and
f ∶ x Ð→ y we have the π-cocartesian lift (f, Idf! a ) ∶ (x, a) Ð→ (y, f! a).
Definition 2.4.6. Let D be a small category. We define Fibcoc (D) to be the 2-
category whose objects are the cocartesian fibrations π ∶ C Ð→ D, whose morphisms
are functors
C? / C′
?? ~~
?? ~~
π ?? ~~ ′
~~~ π
D
over D which send π-cocartesian edges to π ′ -cocartesian edges, and whose 2-morphisms
are the natural transformations which are compatible with the projection to D.
The precise way in which the Grothendieck construction does not loose any
information is summarized in the following folk theorem:
Theorem 2.4.7 (Grothendieck’s correspondence, see, e.g, [4, Theorem 1.3.6]). The
Grothendieck construction determines an equivalence of 2-categories
≃
∫ ∶ PsFun(D, Cat) Ð→ Fib (D)
coc
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 25
f ∨g
Λ20 / N(D)
z<
z
z π
z
∆2 / N(C)
Somewhat surprisingly, the uniqueness can also be phrased as a similar lifting con-
dition using the horn inclusion Λ30 ↪ ∆3 .
Λ30
σ / N(D)
z<
z
z π
z
∆3 / N(C)
∂∆n
ρ
/ Xx/
∆n / Sπ(x)/
(2.23) K × ∆{0}
p
/X
v:
̃ v
H v
v π
v
K × ∆1
H /S
Assume that for every vertex v ∈ K there exists a π-cocartesian edge f ∶ p(v) Ð→ x
̃ ∶ K × ∆1 Ð→ X in (2.23) such that
lifting H({v} × ∆1 ). Then there exists a lift H
̃
H({v} × ∆ ) is π-cocartesian for every v ∈ K.
1
28 YONATAN HARPAZ
6/ X
ρ
(2.24) ∂∆n × ∆1 ∐ ∆n × ∆{0}
∂∆n ×∆{0} n n n
̃ n
ρ
n n π
n n
∆n × ∆1
σ /S
for n ≥ 1. If ρ(∆{0} × ∆1 ) is π-coCatesian then the dotted lift exists.
Proof. For i = 0, ..., n let σi ∶ ∆n+1 Ð→ ∆n ×∆1 be the n+1 simplex given on vertices
by the formula
(j, 0) j≤i
σi (j) = { .
(j − 1, 1) j < i
Consider the filtration Zn+1 ⊆ Zn ⊆ ... ⊆ Z0 = ∆n × ∆1 such that Zn+1 is the top left
corner of (2.24) and for i = 0, ..., n the subsimplicial set Zi is obtained from Zi+1
by adding the (n + 1)-simplex σi . Let us construct the dotted lift ρ̃ inductively on
each Zi . For this, we observe that the n-faces of σi are all contained in Zi+1 except
the face across from the i’th vertex. It follows that we have a pushout square of
simplicial sets
Λn+1 / Zi+1
i
∆n+1 / Zi
Since 0 ≤ i < n + 1 we see that any partial extension ρ̃i+1 ∶ Zi+1 Ð→ X can be
extended to ρ̃i ∶ Zi Ð→ X, either because 0 < i < n + 1 or, when i = 0, by the
assumption that ρ sends ∆{0} × ∆1 to a π-cocartesian edge.
Proof of Proposition 2.4.17. We argue inductively on the skeleton of K. We first
define ρ̃ on {v} × ∆1 for every vertex v ∈ K by choosing π-cocartesian lifts. Given
n ≥ 1 we then extend ρ̃ from the (n − 1)th skeleton to the nth skeleton simplex by
simplex using Lemma 2.4.18.
Construction 2.4.19. Let π ∶ X Ð→ S be an inner fibration and f ∶ s Ð→ s′ an
edge in S such that for every x ∈ Xs there exists a π-cocartesian edge of X lifting
f . Consider the commutative diagram
(2.25) Xs
ι /X
w;
w
w
w π
w
Xs × ∆1 /S
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 29
[f ]
where the bottom horizontal map is the composition Xs × ∆1 Ð→ ∆1 Ð→ S. By
Proposition 2.4.17 there exists a dotted lift in (2.25) which sends {x} × ∆1 to a
π-cocartesian edge for every x ∈ Xs . This can be interpreted as a natural trans-
formation from the fiber inclusion ι ∶ Xs ↪ X to some other functor ι′ ∶ Xs Ð→ X
whose image is contained in the fiber over s′ . In particular, ι′ determines a map
f! ∶ Xs Ð→ Xs′ . It can be shown that this functor is equivalent to the functor
Xs Ð→ Xs′ associated to f by the functor χ ∶ S Ð→ Cat∞ classifying π, and in
particular does not depend on the choice of lift (as long as it is “pointwise” π-
cocartesian). We will refer to the f! ∶ Xs Ð→ Xs′ above as the transition functor
associated to f .
An important particular case of cocartesian fibrations is the following:
Proposition 2.4.20. Let π ∶ X Ð→ S be a cocartesian fibration. Then the following
conditions are equivalent:
(1) π is a left fibration.
(2) Every edge of X is π-cocartesian.
(3) For every s ∈ S the ∞-category π −1 (s) is an ∞-groupoid.
(4) The functor classifying π takes values in the full subcategory Grp∞ ⊆ Cat∞
spanned by ∞-groupoids.
In particular, the Lurie-Grothendieck correspondence descends to an equivalence
≃
Fun(S, S) Ð→ Fibleft (S)
between the ∞-category of functors S Ð→ S (resp. S op Ð→ S) and the ∞-category
of left (resp. right) fibrations over S.
Proof. The equivalence of (1) and (2) is by definition, and the equivalence of (3) and
(4) is by the base change compatibility of the ∞-unstraightening functor (2.22). The
implication (2) ⇒ (3) is obtained by restricting attention to π-cocartesian edges
contained in a fiber and using Example 2.4.12. To see that (3) ⇒ (2) observe that
every arrow in C factors as a composition of a π-cocartesian arrow followed by an
arrow contained in a fiber, and so if (3) holds then every arrow is equivalent to a
π-cocartesian arrow, and is therefore π-cocartesian (see Exercise 2.4.10(3)).
Let us now say a few words about the proof of the Lurie-Grothendieck corre-
spondence (see [5, §3]). The main idea consists of finding suitable model categories
which model the ∞-categories on both sides and then constructing explicit Quil-
len equivalence between them. This is done by using the category Set+∆ of marked
simplicial sets. Recall from Theorem 2.2.16 that there exists a model structure on
Set+∆ whose underlying ∞-category is Cat∞ . The path to the proof of the Lurie-
Grothendieck correspondence passes through the following steps:
(1) Let C = C[S] be the simplicial category generated from S. Then the model
structure of Theorem 2.2.16 induces a model structure on the functor category
(Set+∆ )C whose underlying ∞-category is Fun(S, Cat∞ ).
(2) The category (Set+∆ )/S ♯ of marked simplicial set over S ♯ can be endowed with
a model structure whose underlying ∞-category is Fibcoc (S).
(3) There exists a Quillen equivalence
/
St+S ∶ (Set+∆ )/S ♯ o ⊥ (Set+∆ )C ∶ Un+S
30 YONATAN HARPAZ
which is suitably compatible with base change, and such that for S = ∗ the
resulting Quillen equivalence St+∗ ⊣ Un+∗ is naturally equivalent to the identity.
Furthermore, if S = N(D) is the nerve of a discrete category and F ∶ D Ð→ Cat∞
factors through Cat ⊆ Cat∞ , then Un∞ S (F) is naturally equivalent to the nerve
of ∫D F (with the marked edges being the cocartesian ones).
The Quillen functors St+S and Un+S are known as the straightening and unstraigh-
tening functors. The ∞-unstraightening functor Un∞ S of Theorem 2.4.15 is the one
induced on underlying ∞-categories by Un+S .
2.5. The relative nerve. In this section we will describe an alternative and equi-
valent construction for the ∞-unstraightening which is valid when S is the nerve
of an ordinary category. This construction is combinatorially simpler than that of
St+S and reflects what happens in the ∞-categorical analogue of the Grothendieck
construction more transparently. In [5, §3.2.5] this construction is described under
the name the relative nerve construction.
Definition 2.5.1. Let D be an ordinary category and F ∶ D Ð→ Set∆ a functor.
Define a simplicial set NF (D) as follows. An n-simplex of NF (D) consists of
(1) A functor σ ∶ [n] Ð→ D.
(2) A collection of simplices τS ∶ ∆S Ð→ F(σ(max(S))) for every non-empty subset
S ⊆ [n] such that for every S ′ ⊆ S ⊆ [n] the diagram
(2.26) ∆S
′ τS
/ F(σ(max(S ′ )))
∆S
τS
/ F(σ(max(S)))
commutes.
Forgetting the collection (τS ) we obtain a natural map π ∶ NF (D) Ð→ N(D).
Remark 2.5.2. The compatibility condition (2.26) implies in particular that the
collection τS ∶ ∆S Ð→ F(σ(max(S))) is completely determined by the collection
τ{0,..,i} for i = 0, ...n, and so we could have replaced in Definition 2.5.1 the compatible
collection {τS } by a compatible collection {τ{0,...,i} }. However, choosing all the τS
makes the simplicial structure on NF (D) more evident.
Example 2.5.3. If F ∶ D Ð→ Set∆ is such that F(x) is the nerve of an ordinary
category G(x) for every x ∈ D then NF (D) is naturally isomorphic to the nerve of
the Grothendieck construction of G. In fact, a more general variant of this claim is
true, see Proposition 2.5.9 below.
We note that the the fiber of NF (D) Ð→ N(D) over x ∈ D is canonically isomor-
phic to F(x). In particular, a vertex of NF (D) can be identifies with a pair (x, a)
where x is an object of D and a is an object of F(x). Similarly, an edge from (x, a)
to (y, b) is given by definition by a pair (f, α) where f ∶ x Ð→ y is an arrow in D
and α ∶ f! a Ð→ b is an arrow in F(y), where we have denoted by f! ∶ F(x) Ð→ F(y)
the map associated to f by F.
Proposition 2.5.4. Let D be an ordinary category and F ∶ D Ð→ Set∆ a functor
such that F(x) is an ∞-category for every x ∈ D. Then π ∶ NF (D) Ð→ N(D) is a
cocartesian fibration. Furthermore, an edge (f, α) ∶ (x, a) Ð→ (x, b) is π-cocartesian
if and only if α ∶ f! a Ð→ b is an equivalence in F(y).
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 31
Proof. Let us first show that π is an inner fibration. Consider a lifting problem of
the form
(2.27) Λni
ρ
/ NF (D)
x;
x
x
x
x
∆n
σ / N(D)
with 0 < i < n. For every S ⊆ [n] other than [n] and [n] ∖ {i} the map ρ provides
us with an S-simplex τS ∶ ∆S Ð→ F(σ(max(S))) which satisfies the compatibility
condition 2.26. The collection of composed maps
τS
∆S Ð→ F(σ(max(S))) Ð→ F(σ(n))
then determines a map ρ′ ∶ Λni Ð→ F(σ(n)). The data of the dotted lift is than
equivalent to an extension of ρ′ to a map ρ′ ∶ ∆n Ð→ F(σ(n)). This extension exists
by our assumption that F takes values in ∞-categories.
Let us now consider a lifting problem as in (2.27) with i = 0 and let ρ′ ∶ Λni Ð→
F(σ(n)) be as above. By Corollary 2.3.14 the desired extension exists if ρ′ ∣∆{0,1}
is an invertible edge of F(σ(n)). This will indeed by the case if the edge τ{0,1} ∶
∆{0,1} Ð→ F(σ(1)) is invertible. We may hence conclude that a general edge
(f, α) ∶ (x, a) Ð→ (y, b) as above is π-cocartesian if the edge α ∶ f! a Ð→ b is an
equivalence in F(y). Conversely, if (f, α) is π-cocartesian then we can deduce that
α ∶ f! x Ð→ y is an equivalence from Corollary 2.3.14 by considering the lifting
problem (2.27) as above when σ ∶ ∆n Ð→ N(D) factors through the surjective map
∆n Ð→ ∆1 which sends ∆{0,1} isomorphicaly to ∆1 . To show that π is a cocartesian
fibration it will hence suffice to show that if f ∶ x Ð→ y is an arrow in D and a ∈ F(x)
is an object then there exist b ∈ F(y) and an equivalence α ∶ f! a Ð→ b. But this is
clear: just take b = f! a and α the identity on f! a.
where the coproduct ranges over all maps f ∶ x Ð→ y in D. Taking the fibers over
a particular b ∈ F(y) we may conclude that
MapNF (D) ((x, a), (y, b)) ≃ ∐ MapF(y) (f! a, b).
f ∶x→y
σf
where the bottom horizontal map is the composition F(x) × ∆1 Ð→ ∆1 Ð→ N(D).
Unwinding the definitions we see that we can construct a lift in (2.29) by sending
an n-simplex in F(x) × ∆1 given by a pair (η, ρ) ∈ F(x)n × (∆1 )n to the n-simplex
of NF (D) given by the data (σ, {τS }S⊆[n] ) where σ = σf ○ ρ ∶ ∆n Ð→ N(D) and
τS ∶ ∆S Ð→ F(σ(max(S)) is given by the restriction of ρ ∶ ∆n Ð→ F(x) to ∆S if
f!
ρ(max(S)) = 0 and the composition ρ∣∆S ∶ ∆S Ð→ F(x) Ð→ F(y) if ρ(max(S)) = 1.
We then see that the resulting natural transformation H ∶ F(x) × ∆1 Ð→ NF (D)
is point-wise π-cocartesian by the description of π-cocartesian edges in Proposi-
tion 2.5.4, and that H∣F(x)×∆{1} maps F(x) to the fiber F(y) of π over y via f! . We
may hence conclude that the transition functor F(x) Ð→ F(y) associated to π (see
Construction 2.4.19) is exactly f! .
We finish this section with another version of the Grothendieck construction,
this time for a family of simplicial categories indexed by an ordinary category D.
Given a functor F ∶ D Ð→ Cat∆ , let ∫D F be the simplicial category whose
- objects are pairs (x, a) where x is an object of D and a is an object of F(x);
- The mapping simplicial set from (x, a) to (y, b) is given by
Map∫D F ((x, a), (y, b)) = ∐ MapF(y) (f! x, y)
f ∶x→y
N ( ∫ F) ≅ NN F (D)
D
Proof. Let Θ ∶ [n] Ð→ Set∆ be the functor i ↦ ∆{0,...,i} . In view of Remark 2.5.2
we see that given an n-simplex σ ∶ ∆n Ð→ N(D), the lifts ̃ σ ∶ ∆n Ð→ NN F (D) of σ
are in bijection with natural transformations
(2.30) Θ ⇒ σ∗ N F
of functors from [n] to Set∆ , where we have identified σ with a functor [n] Ð→
D. By adjunction, a natural transformation as in (2.30) is the same as a natural
transformation
(2.31) CΘ ⇒ σ ∗ F,
where we denote by CΘ ∶ [n] Ð→ Cat∆ the composition of Θ and C ∶ Set∆ Ð→ Cat∆ .
Explicitly, a natural transformation as in (2.31) is given by a collection of simplicial
functors ρi ∶ C(∆{0,...,i} ) Ð→ F(i) for i = 0, ..., n, such that for every i ≤ j the diagram
C(∆{0,...,i} ) / F(i)
C(∆{0,...,j} ) / F(j)
commutes. We claim that this is the same data as a map η ∶ C(∆n ) Ð→ ∫[n] σ ∗ F
over [n] (and hence the same data as an n-simplex of N(∫D F) lying above σ).
Indeed, for each i = 0, ..., n let Fi ∶ [i] Ð→ Cat∆ be the restriction of σ ∗ F to
[i] = {0, ..., i} ⊆ [n]. Given an η as above we may restrict it to a map
over [i] for every i = 0, ..., n. We now observe that since i is terminal in [i] we
have a natural functor ∫[i] Fi Ð→ F(i) (which maps each F(i′ ) ⊆ ∫[i] Fi for i′ ≤ i to
F(i) via the map F(i′ ) Ð→ F(i) determined by the functor F). Composing these
functors with ηi we get a compatible collection of functors
η i ∶ C(∆{0,...,i} ) Ð→ F(i).
The association η ↦ {η i } then gives the desired bijection.
2.6. Limits and colimits in ∞-categories. A major advantage of the model of
∞-categories over that of simplicial categories is that it allows for a straightforward
definition of limits and colimits.
Definition 2.6.1. Let C be an ∞-category and x ∈ C an object. We will say that
x is final if MapC (y, x) is contractible for every y ∈ C. Dually, we will say that x is
initial if MapC (x, y) ≃ ∗ for every y ∈ C.
Remark 2.6.2. It follows from Remark 2.3.17 that if C is an ordinary category and
x ∈ C is an object then x is final (resp. initial) in C in the usual sense if and only
if x is final (resp. initial) in N(C) in the sense of Definition 2.6.1. More generally,
if C is a locally Kan simplicial category then x is final in N(C) if and only if it is
homotopy final in C in the sense that MapC (y, x) is weakly contractible for every
y ∈ C.
Definition 2.6.3. Let C be an ∞-category and p ∶ K Ð→ C a diagram in C. We
will say that a left cone p ∶ K ◁ Ð→ C is a limit cone if it is terminal as an object
34 YONATAN HARPAZ
Remark 2.6.4. Given a simplicial set K, the embedding of the cone point {∗} ⊆ K ◁
is left anodyne. It then follows from Proposition 2.6.8 below and Corollary 2.3.10
that if p ∶ K ◁ Ð→ C is a limit cone with p ∶= p∣K and x ∶= p(∗) then in the diagram
C/p
BB
}} BB≃
≃
}}} BB
} BB
}~ }
C/p C/x
CC {
CC {
CC {{
CC {{
! {} {
C
both top projections are trivial Kan fibrations. This expresses the idea that if p is
a limit cone then the object x = p(∗) represents the functor classifying the right
fibration C/p Ð→ C.
Theorem 2.6.5. The classical notion of homotopy limits and colimits in spaces,
coincides with the notion of limits and colimits in the ∞-category S. More generally,
in any model category M the construction of homotopy limits and colimits, as given,
for example, by the explicit construction of Bousfield and Kan, coincides with limits
and colimit in the underlying ∞-category M∞ .
Our next goal is to show that limits and colimits are essentially unique once
they exist. Given their definition via initial and final objects we may formulate this
uniqueness as follows:
Proof. We first note that (2) ⇒ (1) since the base change of a trivial Kan fibration
is a trivial Kan fibration. Now assume that the fibers of π are contractible Kan
complexes. We need to show that a dotted lift exists in any square of the form
∂∆n
σ /C
z<
(2.32)
z
z π
z
∆n
τ /D
If n = 0 then this follows from the fact that the fibers of π are not empty. We may
hence suppose that n ≥ 1. Consider the composed map H ∶ ∆n ×∆1 Ð→ ∆n Ð→ C in
which the first map is given on vertices by the rule (i, 1) ↦ i and (i, 0) ↦ 0. Consider
the restriction H ′ ∶= H∣∂∆n ×∆1 , which we can consider as a natural transformation
from the constant map ∂∆n Ð→ D with value τ (0) to πσ ∶ ∂∆n Ð→ D. Since π is
a right fibration we can use the (dual version of the) lifting property for natural
transformations given in Proposition 2.4.17 to lift H ′ to a natural transformation
′
H ∶ ∂∆n × ∆1 Ð→ C from some σ ′ ∶ ∂∆n Ð→ C to σ, and such that πσ ′ ∶ ∂∆n Ð→ D
is constant with value τ (0). In particular the image of σ ′ is concentrated in the
fiber over τ (0). Since this fiber is a contractible Kan complex by assumption we
can extend σ ′ to a map τ ′ ∶ ∆n Ð→ C such that πτ ′ ∶ ∆n Ð→ D is constant with
′
value τ (0). The maps H and τ ′ together then determine a diagram of the form
∂∆n × ∆1 ∐ ∆n × ∆{0} /6 C
nn
(2.33)
∂∆n ×∆{0} n n
n nn π
nn
∆n × ∆1 /D
Note that we assumed that n ≥ 1. The top horizontal map in (2.33) then sends
the edge ∆{0} × ∆1 to an edge which lies over the identity τ (0) Ð→ τ (0) in D,
and which is hence invertible by Lemma 2.3.12(1). Since any invertible edge is π-
cocartesian (Example 2.4.12) we can apply Lemma 2.4.18 to deduce that the dotted
lift exists in (2.33). Restricting this lift to ∆n ×∆{1} then yields a lift in the original
diagram (2.32).
where the desired lift exists by Proposition 2.6.8 since the projection C/σ(n) Ð→ C
is a right fibrations whose fibers are contractible Kan complexes.
∂∆n
ρ
/C
z<
z
z π
z
∆n
σ /D
in which case we can identify right extensions with objects in the fiber of the
categorical fibration
(2.36) ϕ , E) Ð→ Fun(C, E)
Fun(ConeR
over ψ.
Definition 2.7.1. We will say that τ ∶ ψ ⇒ ϕ∗ ψ ′ exhibits ψ ′ as a left Kan extension
of ψ if (ψ, ψ ′ , τ ) is initial in the fiber of (2.35) over ψ. Dually, we will say that
τ ∶ ϕ∗ ψ ′ ⇒ ψ exhibits ψ ′ as a right Kan extension of ψ if (ψ, ψ ′ , τ ) is final in the
fiber of (2.36) over ψ.
Remark 2.7.2. One can show that the restriction
Fun(ConeLϕ , E) Ð→ Fun(C, E)
is a cartesian fibration. It then follows from (the dual of) Exercise 2.6.10 that if
every ψ ∶ C Ð→ E admits a left Kan extension and we denote by FunL (ConeLϕ , E) ⊆
Fun(ConeLϕ , E) the full subcategory spanned by those triples (ψ, ψ ′ , δ) which corre-
spond to left Kan extensions of ψ along ϕ then the projection
FunL (ConeLϕ , E) Ð→ Fun(C, E)
is a trivial Kan fibration. More generally, if we do not assume that every ψ admits
a left Kan extension but instead we take some full subcategory X ⊆ Fun(C, E) such
that every ψ ∈ X admits a left Kan extension then the projection
FunL (ConeLϕ , E) ×Fun(C,E) X Ð→ X
is a trivial Kan fibration.
The main result that we will need about left and right Kan extensions is their
relation to the notions of limits and colimits described in 2.6. We will describe
this result without giving the proof, and refer the interested reader to [5, §4.3]. To
38 YONATAN HARPAZ
phrase this result, we first note that the cones ConeLϕ and ConeRϕ are simplicial sets
which are generally not ∞-categories. To fix this, let us define
MLϕ ∶= N[ϕ] ([1]) Ð→ ∆1
to be the cocartesian fibration obtained by applying the relative nerve construction
of Definition 2.5.1 to the diagram [ϕ] ∶ [1] Ð→ Set∆ determined by the arrow
ϕ ∶ C Ð→ D in Set∆ . We have inclusions
C ↪ MLϕ ↩ D
identifying C with MLϕ ×∆1 ∆{0} and D with MLϕ ×∆1 ∆{1} . Consider the commutative
diagram
(2.37) C × ∆{1} / C × ∆1
D × ∆{1} / ML
ϕ
(2.38) ̃
C × {1} /̃
C × [1]
̃
ϕ
̃ × {1} / ∫ [ϕ]
D [1] ̃
where we have denoted by [ϕ]̃ ∶ [1] Ð→ Cat∆ the functor corresponding to the arrow
̃ ̃
̃ ∶ C Ð→ D. Inspecting the square 2.38 we now see that it is an actual pushout
ϕ
square in Cat∆ . Since ϕ̃ was chosen to be a cofibration in Cat∆ and Cat∆ is left
proper the square (2.38) is also a homotopy pushout square. The claim now follows
from the fact that N is a right Quillen equivalence and hence preserves homotopy
pushout squares consisting of fibrant objects.
By Lemma 2.7.3 we have that Fun(MLϕ , E) ≃ Fun(ConeLϕ , E) is equivalent to the
category of triples (ψ, ψ ′ , τ ) as above. We will then say that a functor ψ ∶ MLϕ Ð→ E
is a left Kan extension if in the corresponding triple τ exhibits ψ ′ as the left Kan
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 39
commutes, and that the associator and the unitors satisfy the triangle identity,
which says that the diagram
ax,1,y
(x ⊗ 1) ⊗ y / x ⊗ (1 ⊗ y)
LLL
LLL rr
LL rrrrr
LL
yrrr
ρx ⊗Id Id ⊗λy
%
x⊗y
commutes.
≅
Bx,y ∶ x ⊗ y Ð→ y ⊗ x such that Bx,y ○ By,x = Idx⊗y and such that the diagram
ax,y,z
(x ⊗ y) ⊗ z / x ⊗ (y ⊗ z) Bx,y⊗z / (y ⊗ z) ⊗ x
We will denote by π ∶ C⊗ Ð→ Fin∗ the natural projection (n, (x1 , ..., xn )) ↦ ⟨n⟩.
We now claim that, similarly to the case of the Grothendieck construction des-
cribed in §2.4, the entire symmetric monoidal structure on C can be completely
reconstructed from C⊗ together with the functors π ∶ C⊗ Ð→ Fin∗ and the identifi-
cation of the fiber C⊗
⟨1⟩
with C. To see this, we first observe the following:
Lemma 3.1.4. An edge (α, {fj }) ∶ (n, (x1 , ..., xn )) Ð→ (m, (y1 , ..., ym )) in C⊗ is
π-cocartesian if and only if each fj ∶ ⊗i∈α−1 (j) xi Ð→ yj is an isomorphism. In
particular, π is a cocartesian fibration.
42 YONATAN HARPAZ
⟨3⟩
β1
/ ⟨2⟩
AA
AA γ
AA
β2
AA α
α /
⟨2⟩ ⟨1⟩
where for j = 1, 2 we denote by βj ∶ ⟨3⟩ Ð→ ⟨2⟩ is the unique pointed surjective map
such that βj−1 (j) = {j, j + 1}.
We may consider the association C ↦ C⊗ as an operation which transforms a
symmetric monoidal category to a cocartesian fibrations over Fin∗ . However, not
every cocartesian fibration over Fin∗ can be obtained in this way: indeed, the fiber
C⊗
⟨n⟩
over ⟨n⟩ (equivalently, the category associated to ⟨n⟩ by the corresponding
pseudofunctor) is always the n-fold cartesian product of the fiber C⊗ ⟨1⟩
. To phrase
this condition more precisely let us introduce some terminology.
Definition 3.1.6. We will say that a morphism α ∶ ⟨n⟩ Ð→ ⟨m⟩ is active if α−1 (0) =
{0} and that it is inert if α−1 (j) ⊆ ⟨n⟩ contains exactly one element for every
j ∈ {1, ...n}.
Remark 3.1.7. Every morphism in Fin∗ can be factored as an inert morphism
followed by an active morphism in an essentially unique way. In addition, this
factorization is initial in the category of factorizations whose second map is active
and terminal in the category of factorizations whose first map is inert.
Notation 3.1.8. For an ⟨n⟩ ∈ Fin∗ and an i = 1, ..., n let us denote by ρi ∶ ⟨n⟩ Ð→ ⟨1⟩
the inert morphism such that (ρi )−1 (1) = {i}.
We then have the following folk theorem:
Theorem 3.1.9. The association C ↦ C⊗ induces an equivalence between the 2-
category of symmetric monoidal categories, symmetric monoidal functors and sym-
metric monoidal natural transformations and the full sub-2-category of Funcoc (Fin∗ )
spanned by the cocartesian fibrations D Ð→ Fin∗ with the following property: for
every ⟨n⟩ ∈ Fin∗ the functor
n
(ρ1! , ..., ρn! ) ∶ D⟨n⟩ Ð→ ∏ D⟨1⟩
i=1
Remark 3.1.10. We have not defined terms “symmetric monoidal functor” appea-
ring in Theorem 3.1.9. It consists of a functor ϕ ∶ C Ð→ D between symmetric
≅
monoidal categories together with a specified isomorphism 1D Ð→ F(1C ) and a
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 43
specified natural isomorphism µx,y ∶ F(x) ⊗ F(y) Ð→ F(x ⊗ y) which satisfy several
compatibility conditions with respect to the associators, unitors and symmetry iso-
morphisms of C and D. Theorem 3.1.9 tells us that this data is essentially the same
as the data of a functor C⊗ Ð→ D⊗ which preserves cocartesian edges, a notion
which is considerably simpler in comparison. This reflects the fact that the “fibra-
tion” picture encodes homotopy coherences in a much more efficient manner than
the “algebraic” picture. When passing to symmetric monoidal ∞-categories this
improvement makes a critical difference.
Theorem 3.1.9 suggests the following way to define the notion of a symmetric
monoidal ∞-category:
Definition 3.1.11. A symmetric monoidal ∞-category is a cocartesian fibration
π ∶ C⊗ Ð→ N(Fin∗ ) such that for every n the map
n
(ρ1! , ..., ρn! ) ∶ C⊗ ⊗
⟨n⟩ Ð→ ∏ C⟨1⟩
i=1
determined by the transition functors of ρ1 , ..., ρn ∶ ⟨n⟩ Ð→ ⟨1⟩ (see Construction 2.4.19)
is an equivalence of ∞-categories. We will denote by C ∶= C⊗ ⟨1⟩
the fiber over 1, and
⊗
will refer to it as the underlying ∞-category of C . In this situation we will also
say that π ∶ C⊗ Ð→ Fin∗ exhibits C as a symmetric monoidal ∞-category.
If π ∶ C⊗ Ð→ N(Fin∗ ) and π ′ ∶ D⊗ Ð→ N(Fin∗ ) are two symmetric monoidal
∞-categories then we define a symmetric monoidal functor from C⊗ to D⊗ to be a
functor C⊗ Ð→ D⊗ over N(Fin∗ ) which sends π-cocartesian edges to π ′ -cocartesian
edges.
Remark 3.1.12. The notation C⊗ in Definition 3.1.11 is merely suggestive in light
of the discussion in §3.1, we do not mean that C⊗ is obtained by applying some
construction to C.
Remark 3.1.13. Let C⊗ Ð→ N(Fin∗ ) be a symmetric monoidal ∞-category with
underlying ∞-category C = C⊗
⟨1⟩
. Using Construction 2.4.19 we see that the active
morphisms ⟨0⟩ Ð→ ⟨1⟩ and ⟨2⟩ Ð→ ⟨1⟩ determine functors
∆0 Ð→ C and ⊗ ∶ C × C Ð→ C
which are well-defined up to a contractible space of choices. The first of these
functors determines an object of C which we will denote by 1 and refer to as the
unit object of C. One can then check that the unit object 1 and the product ⊗ satisfy
the axioms of a symmetric monoidal category up to homotopy. In particular, the
homotopy category Ho(C) inherits a canonical symmetric monoidal structure.
3.2. Examples and constructions. In this section we will review some examples
and constructions of symmetric monoidal ∞-categories. We first note the following
immediate example:
Example 3.2.1. If C is an ordinary category then the nerve of the category C⊗
constructed in §3.1 is a symmetric monoidal ∞-category.
The following proposition provides a large source of symmetric monoidal ∞-
categories:
44 YONATAN HARPAZ
on S = N(Kan) and Cat∞ = N(QC). These structures agree with those of Exam-
ple 3.2.4 (see [6, Corollary [Link]]) and both coincide with the respective cartesian
symmetric monoidal structures (see §3.3).
A large supply of symmetric monoidal simplicial categories comes from symme-
tric monoidal simplicial model categories. These are the symmetric monoidal model
categories M which admit a symmetric monoidal left Quillen functor ι ∶ Set∆ Ð→ M,
in which case they automatically acquire the structure of a simplicial model cate-
gory (see Definition 2.2.21). In this case the simplicial enrichment associated to the
simplicial structure will endow M with the structure of a (symmetric monoida) sim-
plicial category. However, the resulting simplicial category will not be locally Kan
in general, and so we cannot apply Proposition 3.2.8 directly. On the other hand, we
have the full simplicial subcategory M○ ⊆ M spanned by fibrant-cofibrant objects,
which is locally Kan and satisfies further that N(M○ ) ≃ M∞ . Unfortunately, M○ will
generally not be closed under ⊗: the tensor product of two fibrant-cofibrant objects
will generally not be fibrant anymore. The solution to this problem is however not
complicated:
Proposition 3.2.10. Let M be a simplicial symmetric monoidal model category,
and let M⊗,○ ⊆ M⊗ denote the full subcategory spanned by those (n, (x1 , ..., xn ))
such that x1 , ..., xn ∈ M○ . Then the map
N(M⊗,○ ) Ð→ N(Fin∗ )
is a cocartesian fibration which exhibits N(M○ ) as a symmetric monoidal ∞-category.
Remark 3.2.11. If M is a simplicial symmetric monoidal model category then
N(M○ ) ≃ M∞ ≃ Mc [(W c )−1 ] can be endowed with a symmetric monoidal struc-
ture by either Proposition 3.2.2 or by Proposition 3.2.10. These two symmetric
monoidal structure are however equivalent to each other, see [6, Corollary [Link]].
3.3. cartesian and cocartesian symmetric monoidal structures. Let C be an
∞-category which admits finite coproducts, that is colimits for diagrams indexed by
finite sets. Then we can expect that C can be endowed with a symmetric monoidal
structure in which the monoidal product is given by the categorical coproduct.
Similarly, if C admits finite products then we can expect to have a symmetric
monoidal structure with the operation of cartesian products. In this section we will
see how to express these symmetric monoidal structures explicitly in the formalism
of symmetric monoidal ∞-categories.
We begin with the case of coproducts. Define a category Γ∗ as follows:
○
- The objects of Γ∗ are pairs (⟨n⟩ , i) where ⟨n⟩ ∈ Fin∗ and i ∈ ⟨n⟩ .
- A morphism in Γ∗ from (⟨n⟩ , i) to (⟨m⟩ , j) is a map of pointed sets α ∶ ⟨n⟩ Ð→
⟨m⟩ such that α(i) = j.
The category Γ∗ admits an obvious forgetful functor Γ∗ Ð→ Fin∗ sending (⟨n⟩ , i)
to ⟨n⟩.
Definition 3.3.1. Let C be an ∞-category. We define C∐ to be the simplicial set
whose n-simplices are pairs (σ, ρ) where σ ∶ ∆n Ð→ N(Fin∗ ) is an n-simplex of
N(Fin∗ ) and ρ ∶ ∆n ×N(Fin∗ ) N(Γ∗ ) Ð→ C is a map of simplicial sets, where the fiber
product is taken with respect to σ. By construction the simplicial set C∐ comes
equipped with a map C∐ Ð→ N(Fin∗ ) sending (σ, ρ) to σ.
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 47
We note that the vertices of C∐ can be identified with pairs (⟨n⟩ , x) where
○
⟨n⟩ ∈ Fin∗ and x ∶ {⟨n⟩} ×N(Fin∗ ) Γ∗ ≅ ⟨n⟩ Ð→ C is a map, which we can identify
with a tuple x = (x1 , ..., xn ) of objects in C. Similarly, an arrow of C∐ consists of a
tuple (α, f ) where α ∶ ⟨n⟩ Ð→ ⟨m⟩ is an arrow in N(Fin∗ ) and
(3.2) f ∶ ∆1 ×N(Fin∗ ) Γ∗ = N(Iα ) Ð→ C
is a map, a data which we may identify with s triple (x, y, {fi }) where x = (x1 , ..., xn )
is an n-tuple of objects in C, y = (y1 , ..., ym ) is an m-tuple of objects in C and
○ ○
fi ∶ xi Ð→ yα(i) is a map for each i ∈ α−1 (⟨m⟩ ) ⊆ ⟨n⟩ .
Exercise 3.3.2. Suppose that C is a locally Kan simplicial category. Define a sim-
plicial category C∐ as follows: the objects of C∐ are pairs (n, (x1 , ..., xn )) as in
Construction 3.2.7 and the simplicial mapping sets are given by
m
Map((n, (x1 , ..., xn ), (m, (y1 , ..., ym )) = ∐ ∏ ∏ MapC (xi , yj ),
α∶⟨n⟩→⟨m⟩ j=1 i∈α−1 (j)
where α ranges over all maps from ⟨n⟩ to ⟨m⟩ in Fin∗ . Show that there is a canonical
isomorphism of simplicial sets
N(C∐ ) ≅ N(C)∐
where the left hand side is constructed as in Definition 3.3.1.
Proposition 3.3.3 ([6, Proposition [Link]]). Let C be an ∞-category. Then π ∶
C∐ Ð→ N(Fin∗ ) is an inner fibration and an arrow
(α, {fi }) ∶ (⟨n⟩ , (x1 , ..., xn )) Ð→ (⟨m⟩ , (y1 , ..., ym ))
as above is π-cocartesian if and only if for every j = 1, ..., m the collection of maps
fi ∶ xi Ð→ yj for i ∈ α−1 (j) exhibit yj as the coproduct of {xi }i∈α−1 (j) . In addition,
if C admits finite coproducts then π is a symmetric monoidal structure on C, which
we call the cocartesian symmetric monoidal structure.
Proof. Let us show that π is an inner fibration. Consider a lifting problem of the
form
(3.3) Λni
ρ
/ C∐
w;
w
w
w
w
∆n
σ / N(Fin∗ )
We note that ∆n ×N(Fin∗ ) Γ∗ can be identified with the nerve of a poset Iσ whose
○
objects are pairs (j, a) with j ∈ {0, ..., n} and a ∈ ⟨kj ⟩ and such that (j, a) ≤ (j ′ , a′ ) if
′ ′
and only if j ≤ j and αj,j ′ (a) = a . We then observe that m-simplices ∆m Ð→ N(Iσ )
are in bijections with pairs (τ, a) where τ ∶ ∆m Ð→ ∆n is an m-simplex in ∆n and
○ ○
a ∈ ⟨kτ (0) ⟩ is such that ατ (0),τ (m) (a) ∈ ⟨kτ (m) ⟩ . Furthermore, such a simplex is
non-degenerate if and only if τ is a non-degenerate m-simplex of ∆n .
○
Let A ⊆ ⟨k0 ⟩ be the subset containing those elements a ∈ ⟨k0 ⟩ such that α0,n (a) ∈
○
⟨kn ⟩ . Using this explcit description we see that the non-degenerate simplices of
N(Iσ ) that are not in Λni ×N(Fin∗ ) Γ∗ are either of the form (∆{0,...,n} , a) for a ∈ A or
of the form (∆{0,...,î,...,n} , a) for a ∈ A, for a fixed a the latter simplex is the i’th face
of the former, while all the other faces of the former are contained in Λni ×N(Fin∗ ) Γ∗ .
It follows that the simplicial set N(In ) is obtained from Λni ×N(Fin∗ ) Γ∗ by performing
a pushout along Λin ⊆ ∆n for each a ∈ A. Since C is an ∞-category it follows that
the dotted extension exists in (3.4), and so the dotted lift exists in (3.3). We may
hence conclude that C∐ Ð→ N(Fin∗ ) is an inner fibration.
Let us now consider a lifting problem as in (3.3) with i = 0 and such that ρ
sends the edge ∆{0,1} to an arrow given by (α, f ) ∶ (⟨n⟩ , x) Ð→ (⟨m⟩ , y) as above,
○ ○
where x ∶ ⟨n⟩ Ð→ C can be identified with a tuple (x1 , ..., xn ), y ∶ ⟨m⟩ Ð→ C can
be idetified with a tuple (y1 , ..., ym ) and
(3.5) f ∶ ∆1 ×N(Fin∗ ) Γ∗ = N(Iα ) Ð→ C
○
can be identified with a collection of maps fi ∶ xi Ð→ yα(i) for each i ∈ α−1 (⟨m⟩ ) ⊆
○
⟨n⟩ . Such a lifting problem is then equivalent to an extension problem as in (3.4)
○
with i = 0. Let B ⊆ ⟨k1 ⟩ be the subset containing those elements b ∈ ⟨k1 ⟩ such that
○
α1,n (b) ∈ ⟨kn ⟩ . The non-degenerate simplices of N(Iσ ) that are not in Λn0 ×N(Fin∗ ) Γ∗
are either of the form (∆{0,...,n} , a) for a ∈ A or of the form (∆{1,...,n} , b) for b ∈ B.
In addition, all the faces of (∆{0,...,n} , a) are contained in Λn0 ×N(Fin∗ ) Γ∗ except
the 0’th face which is (∆{1,...,n} , α0 (a)). We then see that N(Iσ ) is obtained from
Λn0 ×N(Fin∗ ) Γ∗ by performing, for each b ∈ B, a pushout along a map of the form
(3.6) Sb ⋆ ∂∆{1,...,n} Ð→ Sb ⋆ ∆{1,...,n}
where Sb ⊆ A is the preimage of b. Hence to solve the extension problem (3.4) we
need to solve a series of extension problems of the form
Sb ⋆ ∂∆{1,...,n} /
(3.7)
s s9 C
s s
ss
Sb ⋆ ∆{1,...,n}
Let xb ∶ Sb Ð→ C be the restriction of x ∶ ⟨n⟩ Ð→ C. The extension problem 3.7 is
equivalent to the extension problem
(3.8) ∂∆{1,...,n} / Cx /
u:
b
u
u
u
u
∆{1,...,n}
Let f b ∶ Sb ⋆ ∆{1} Ð→ C be the restriction of (3.2) to Sb ⋆ ∆{1} ⊆ ∆1 ×N(Fin∗ ) Γ∗ . By
Corollary 2.6.9 we see that the extension problem 3.8 is solvable as soon as f b is
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 49
initial in Cxb , that is, as soon as f b is a colimit diagram. We may thus conclude that
○
the edge (x, y, {fi }) is π-cocartesian if (and in fact only if) for each b ∈ ⟨m⟩ the
collection of maps fa ∶ xa Ð→ yb for a ∈ Sb exhibit yb is the coproduct of {xa }a∈Sb .
Now assume that C has coproducts. In this case we have by the above that for
every (⟨n⟩ , (x1 , ..., xn )) in C∐ and every α ∶ ⟨n⟩ Ð→ ⟨m⟩ there exists a cocartesian
edge (⟨n⟩ , (x1 , ..., xn )) Ð→ (⟨m⟩ , (y1 , ..., ym )) lying above α, and so π is a cocar-
tesian fibration. To finish the proof we need to check that the transition functors
along the inert maps ρi ∶ ⟨n⟩ Ð→ ⟨1⟩ induce an equivalence
(3.9) C∐
⟨n⟩
≃ ∏ C∐
⟨1⟩
.
i
Inspecting the definition 3.3.1 we now observe that base change
C∐ ×N(Fin∗ ) N(Fin∗ )in Ð→ N(Fin∗ )in
is canonically isomorphic to the relative nerve construction (see Definition 2.5.1)
of the functor (Fin∗ )in Ð→ Set∆ which sends ⟨n⟩ to C⟨n⟩ and each inert map
○
○ ○
α ∶ ⟨n⟩ Ð→ ⟨m⟩ corresponding to an injection α′ ∶ ⟨m⟩ Ð→ ⟨n⟩ to the restriction
(α′ )∗ ∶ C⟨n⟩ Ð→ C⟨m⟩
○ ○
⟨n⟩ ⟨1⟩
is equivalent to the projection to
the ith factor C⟨n⟩ Ð→ C, hence the equivalence (3.9).
○
Remark 3.3.4. Proposition 3.3.3 implies that C∐ is an ∞-category as soon as C is an
∞-category (even if C does not admit finite coproducts). The mapping spaces in C∐
can then be explicitly identified: given x = (x1 , ..., xn ) ∈ C∐
⟨n⟩
and y = (y1 , ..., ym ) ∈
C∐
⟨m⟩
we have a canonical isomorphism of simplicial sets
∐
Cx/ ×N(Fin∗ ) C∐
⟨m⟩
≅ ∐ ∏ Cxj / ,
α∶⟨n⟩→⟨m⟩ j∈⟨m⟩○
○
N(Pn )op Ð→ C where Pn denote the poset of subsets of ⟨n⟩ . We then define
C× ⊆ ̃
C× be the subsimplicial set spanned by those vertices whose corresponding
○
functors ϕ ∶ N(Pn )op Ð→ C satisfy the following property: for every I ⊆ ⟨n⟩ the
maps ϕ(I) Ð→ ϕ({i}) for i ∈ I exhibit ϕ(I) as the product in C of the objects
{ϕ({i})}.
(4.1) [ ∏ MulO ({xi }i∈α−1 (j) , yj )] × MulO ({yj }j∈J , z) Ð→ MulO ({xi }i∈I , z).
j∈J
- for every object x ∈ Ob(O) a designated identity multimap Idx ∈ MulO ({x}, x)
which is (left and right) neutral with respect to the composition map above.
If I has cardinality n then we will say that {xi }i∈I Ð→ y is a multimap of arity
n. The composition rule is required to be associative in the following sense: for
α β
every sequence of maps I Ð→ J Ð→ K of finite sets with γ = β ○ α, every triple of
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 51
collections {xi }i∈I , {yj }j∈J , {zk }k∈K and every object w ∈ Ob(O), the diagram
ffff fffff
ffff
fffff
∏ Mul({xi }i∈γ −1 (k) , zk ) × Mul({zk }k∈K , w)
k∈K
commutes.
A map of operads ϕ ∶ O Ð→ P consists of a map of sets Ob(O) Ð→ Ob(P) together
with maps MulO ({xi }i∈I , y) Ð→ MulP ({ϕ(xi )}i∈I , ϕ(y)) for every finite collection
{xi }i∈I and object y, which are compatible with the composition operation and the
identity multimaps.
Remark 4.1.2. Every operad O has an underlying category whose objects are the
objects of O and whose morphism sets are given by Mul({x}, y). We may hence
consider an operad as a category with additional structure, namely, the multimaps
{xi }i∈I Ð→ y. Some authors hence use the term multicategory to talk about these
kinds of operads.
Example 4.1.3. Let C be a symmetric monoidal category. Then we may associate
to C an operad as follows: the objects of O are the objects of C and for every finite
collection {xi }i∈I and object y ∈ C we set
MulC ({xi }i∈I , y) = HomC (⊗i xi , y).
We will refer to this operad as the underlying operad of C.
Warning 4.1.4. If C is a symmetric monoidal category then C can be completely
reconstructed from its underlying operad by the Yoneda lemma. However, if C
and D are two symmetric monoidal categories then a map of operads C Ð→ D is
generally not the same as a map of symmetric monoidal categories. The latter
is given by a symmetric monoidal functor while the former corresponds to a lax
symmetric monoidal functor from C to D.
The main motivation of the notion of an operad is that it can be used to encode
the information of an algebraic structure.
Definition 4.1.5. Let O be an operad and C a symmetric monoidal category. An
O-algebra in C is a map of operads O Ð→ C, where C is considered as an operad by
Example 4.1.3.
Example 4.1.6. If O has only multimaps of arity 1 then O is simply a category. In
this case an O-algebra in a symmetric monoidal category C is the same as a functor
of ordinary categories O Ð→ C.
Examples 4.1.7.
52 YONATAN HARPAZ
(1) Let Com be the operad with a single object ∗ and such that the set of multimaps
MulCom ({∗}i∈I , ∗) contains a single element for every finite set I (in particular,
Com is terminal in the category of operads). If C is a symmetric monoidal
category then the data of a Com-algebra in C consists of an object A ∈ C (the
image of ∗) together with maps ⊗i∈I A Ð→ A for every finite set I, which satisfy
a compatibility condition for every I Ð→ J. Unwinding the definition we see
that this is exactly the data of a commutative algebra structure on A.
(2) Let Triv be the operad with a single object ∗ and such that the only multimap
is the identity. Then a Triv-operad in a symmetric monoidal category C is
simply an object of C with no additional structure. We will refer to Triv as the
trivial operad.
(3) Let Poi be the operad with a single object ∗ and such that the set of multimaps
MulPoi ({∗}i∈I , ∗) contains a single element if ∣I∣ ≤ 1 and is empty otherwise.
Then the data of a Poi-algebra object in a symmetric monoidal ∞-category C
consists of an object A ∈ C together with a map 1 Ð→ A from the unit of C. We
will refer to Poi as the operad of pointed objects.
(4) Let Ass be the operad with a single object ∗ and such that for a finite set I
the multimaps {∗}i∈I Ð→ ∗ are given by linear orderings on I (composition is
defined by concatenating linear orders). If C is a symmetric monoidal category
then the data of an Ass-algebra in C consists of an object A ∈ C (the image
of ∗) together with maps ⊗i∈I A Ð→ A for every finite linearly ordered set I,
which satisfy a compatibility condition for every order preserving map I Ð→ J.
Unwinding the definition we see that this is exactly the data of an associative
algebra structure on A.
(5) Let AssInv be the operad with a single object ∗ and for a finite set I, the
multimaps {∗}i∈I Ð→ ∗ are given by pairs (≤, ε) where ≤ is a linear order on
I and ε ∶ I Ð→ {−1, 1} is an assignment of signs to each i ∈ I. Composition of
multimaps is given by the concatenation of linear orders, reversal of linear orders
according to signs, and multiplication of signs. For example, the composition
of the multimaps
1<2,ε1 =ε2 =1 ε1 =−1
{∗}i∈{1,2} / {∗}i∈{1} /∗
is the multimap
2<1,ε1 =ε2 =−1
{∗}i∈{1,2} /∗.
- maps from (n, (x1 , ..., xn )) to (m, (y1 , ..., ym )) are given by maps α ∶ ⟨n⟩ Ð→ ⟨m⟩
in Fin∗ , together with, for every j = 1, ..., n, a multimap in O of the form
We will denote by π ∶ O⊗ Ð→ Fin∗ the natural projection (n, (x1 , ..., xn )) ↦ ⟨n⟩.
We emphasize that this time π is not a cocartesian fibration, i.e., not every edge
in Fin∗ admits cocartesian lifts. However, some edges in Fin∗ do have cocartesian
lifts: if α ∶ ⟨n⟩ Ð→ ⟨m⟩ is an inert map and (n, (x1 , ..., xn )) is an object of O⊗ lying
over ⟨n⟩ then maps f ∶ (n, (x1 , ..., xn )) Ð→ (m, (y1 , ..., ym )) lying above α are given
by specifying, for each j = 1, ..., m, a map fj ∶ xi Ð→ yj in the underlying category
of O, where i ∈ {1, ..., n} is the unique element mapping to j by α. It is then not
difficult to check that f is π-cocartesian if and only if each fj is an isomorphism
in the underlying category of O. We then see that each inert map α ∶ ⟨n⟩ Ð→ ⟨m⟩
admits a cocartesian lift starting at an arbitrary object over ⟨n⟩. This means that
π becomes a cocartesian fibration when restricted to the subcategory Finin ∗ ⊆ Fin
consisting of all objects and all inert maps.
Similarly to the case of symmetric monoidal category, the operad O can be
completely reconstructed from the category O⊗ together with the forgetful functor
π ∶ O⊗ Ð→ Fin∗ . For example, the underlying category of O can be identified with
the fiber O⊗ ⟨1⟩
= π −1 (⟨1⟩). More generally, suppose that n ≥ 0 and for 1 ≤ i ≤ n let
ρi ∶ ⟨n⟩ Ð→ ⟨1⟩ be as in Notation 3.1.8. Since ρi is inert it has by the discussion
above an associated transition functor ρi! ∶ O⊗ ⟨n⟩
Ð→ O⊗ ⟨1⟩
(see Construction 2.4.19),
and the collection of transition functors ρ! determines an equivalence O⊗
i
⟨n⟩
≃ (O⊗⟨1⟩
)n .
Given an object in O⊗ ⟨n⟩
we may hence identify it with a sequence x1 , ..., xn ∈ O⊗⟨1⟩
intrinsically. Furthermore, if x is an object of O⊗ ⟨n⟩
which corresponds to the tuple
(x1 , ..., xn ) and y ∈ O⊗ ⟨1⟩
≃ O then maps from x to y in O⊗ which lie above the
unique active map ⟨n⟩ Ð→ ⟨1⟩ are in bijection with multimaps {xi }ni=1 Ð→ y in O.
Elaborating further along these lines one can also reconstruct the composition of
multimaps in O.
The discussion above, together with the definition of a symmetric monoidal ∞-
category (see Definition 3.1.11), suggest that we might attempt to define an ∞-
operad as an ∞-category O⊗ over Fin∗ which satisfies certain conditions. As far as
we know defining ∞-operads in this way was first done in by Lurie (see [6, §2.1.1]).
(3) Let y ∈ O⊗ be an object lying above ⟨m⟩ and for each j = 1, ..., m let fj ∶ y Ð→ yi
be a π-cocartesian edge lying above ρj . Then the maps {fj } exhibit y as the
relative product of y1 , ..., ym in O⊗ over N(Fin∗ ) in the following sense: for
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 55
MapFin∗ (⟨n⟩ , ⟨m⟩) / MapFin (⟨n⟩ , ⟨1⟩)
∗
is a cartesian square in S.
In this case we will also say that π exhibits O⊗ as an ∞-operad.
Example 4.1.11. If O is an ordinary operad and O⊗ is obtained from O via con-
struction 4.1.9 then the map N(O⊗ ) Ð→ N(Fin)) exhibits N(O⊗ ) as an ∞-operad.
In particular, all the examples of (4.1.7) also yield examples of ∞-operads. We note
that Com⊗ is equivalent to Fin∗ itself and Poi⊗ is equivalent to the subcategory of
Fin∗ consisting of all objects and all maps α ∶ ⟨n⟩ Ð→ ⟨m⟩ such that ∣α−1 (i)∣ ≤ 1 for
i = 1, ..., n. Similarly, Triv⊗ is simply the subcategory of Fin∗ consisting of all inert
maps.
Example 4.1.12. Let C be an ∞-category and let π ∶ C∐ Ð→ N(Fin∗ ) be as in
Definition 3.3.1. Then Proposition 3.3.3 implies that C∐ satisfies Condition (1) of
Definition 4.1.10. Arguing as in the final part of the proof of Proposition 3.3.3
we see that π ∶ C∐ Ð→ N(Fina ∗) is in fact an ∞-operad (even if C does not have
coproducts). We call C∐ the cocartesian ∞-operad of C.
Generalizing Example 4.1.11 we may also construct ∞-operads starting from
simplicial operads. Recall that a simplicial operad O is given by the simplicially
enriched version of Definition 4.1.1: we have a set of objects Ob(O) and for every
collection of objects {xi }i∈I indexed by a finite set I and every object y ∈ Ob(O), a
simplicial set MapO ({xi }, y) of multimaps from {xi }i∈I to y. The composition of
multimaps is then given as in (4.1) by replacing sets with simplicial sets. Given a
simplicial operad O we may perform the analogue of Construction 4.1.9 to obtain
a simplicial category O⊗ whose
- objects are pairs (n, (x1 , ..., xn )) consisting of a non-negative integer n and a
family of objects in O parameterized by {1, ..., n}.
- the simplicial set of maps from (n, (x1 , ..., xn )) to (m, (y1 , ..., ym )) is given by
n
∐ ∏ MulO ({xi }i∈α−1 (j) , yj )
α∶⟨n⟩→⟨m⟩ j=1
where (x, ..., x) denotes an object of O which corresponds to the tuple (x, ..., x)
under the (canonical) equivalence O⟨n⟩ ≃ On⟨1⟩ , and Σn is the permutation group
on n elements which naturally act on MapO ((x, ..., x), y) ⋅ Z n . In particular, if ϕ
induces an equivalence MonP (S) Ð→ MonO (S) then it also induces an equivalence
Fx (Z) ≃ ϕ∗ Fϕ(x) (Z) for every space Z. Setting Z = ∗ we obtain that ϕ induces an
equivalence
MapO ((x, ..., x), y)hΣn Ð→ MapP ((ϕ(x), ..., ϕ(x)), ϕ(y))hΣn
for every n and hence an equivalence
MapO ((x, ...., x), y) ≃ MapO ((x, ..., x), y)hΣn ×(∗)hΣn {∗} Ð→
MapO ((ϕ(x), ..., ϕ(x)), ϕ(y))hΣn ×(∗)hΣn {∗} ≃ MapP ((ϕ(x), ..., ϕ(x)), ϕ(y))
for every n. This finishes the proof if the underlying ∞-category O is a connected
Kan complex. In general, to handle non-equivalent x1 , ..., xn one can do a similar
argument using the left adjoint of the forgetful functor
evx1 ,...,xn ∶ MonO (S) Ð→ Sn
which sends an O-monoid in spaces to the n-tuple of spaces associated to the objects
x1 , ..., xn .
4.2. Weak ∞-operads and approximations. The notion of an ∞-operad defi-
ned above is somewhat rigid. It is hence useful sometimes to consider mild genera-
lizations, which are more amenable to various constructions. In this section we will
focus on one such notion, which we call a weak ∞-operad. This notion is somewhat
add-hoc (we are not aware of any literature using this particular definition), and
for the purposes of these notes it should be considered as merely a tool in order
to prove results on ∞-operads. However, we note that it does admits quite a few
interesting examples which arise naturally, and can be considered as a variant on
the theory of operator categories.
Recall that a map ι ∶ C0 Ð→ C of ∞-categories is called a subcategory of C if
the induced map MapC0 (x, y) Ð→ MapC (ι(x), ι(y)) has (−1)-truncated homotopy
fibers (that is, it is equivalent to an inclusion of a set of components). In this case
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 59
we will often omit ι and simply write C0 ⊆ C (though we do not mean by this that
ι is injective on the level of simplicial sets). A subcategory C0 ⊆ C is called wide if
the induced map C≃0 Ð→ C≃ is an equivalence of ∞-groupoids. In this case we will
1
say that an arrow of C belongs to C0 if is equivalent in C∆ to an arrow which is
in the image of C0 . In particular, when C0 ⊆ C is a wide subcategory then every
equivalence in C belongs to C0 .
Definition 4.2.1. Let C be an ∞-category. By a factorization system on C we will
mean two wide subcategories Cin , Cact ⊆ C, whose maps we call the inert and active
maps respectively, such that the following two conditions hold:
f g
(1) Every morphism h ∶ x Ð→ z in C can be factored as a composition x Ð→ y Ð→ z
such that f is inert and g is active.
(2) For every commutative square in C of the form
(4.3) x /z
~>
~ g
f
~
~ / w
y
such that f is inert and g is active the space of dotted lifts is contractible.
Remark 4.2.2. The notion of a factorization system given in Definition 4.2.1 is
also sometimes called an orthogonal factorization system or a unique factorization
system. It is stronger than the notion of a weak factorization system commonly
found in litterature on model categories, in which the lifting solutions in (4.3) are
not required to be essentially unique.
Exercise 4.2.3. Let (Cin , Cact ) be a unique factorization system on C.
(1) For a map h ∶ x Ð→ z let us denote by Cx//z ∶= (Cx/ )/h the ∞-category of
factorizations
f g
x Ð→ y Ð→ z
of f . Consider the full subcategory Cx//z ×Cx/ Cin x/ ⊆ Cx//z consisting of those
factorizations h = g ○ f such that f is inert and the full subcategory Cx//z ×Cx/
/z ⊆ Cx//z consisting of those factorizations h = g ○ f such that g is active.
Cact
Show that if h = g ○ f is a factorization such that f is inert and g is active then
(f, g) is final when considered as an object of Cx//z ×Cx/ Cin x/ and initial when
considered as an object of Cx//z ×C/z C/z .
act
(2) Deduce that for every morphism h ∶ x Ð→ z in C the full subcategory Eh ⊆ Cx//y
consisting of those factorizations
f g
x Ð→ y Ð→ z
of h such that f ∈ Cin and g ∈ Cact , is a contractible Kan complex.
f g
(3) Show that if x Ð→ y Ð→ z are two composable arrows such that f is inert then
g is inert if and only if g ○ f is inert. Similarly, if g is active then f is active if
and only if g ○ f is active.
Definition 4.2.4. A weak ∞-operad is an ∞-category C equipped with a facto-
rization system (Cin , Cact ) and a full subcategory C0 ⊆ C such that the following
60 YONATAN HARPAZ
condition hold: for every x ∈ C there exists a finite set Ix and a collection of maps
fi ∶ x Ð→ xi in Cin indexed by i ∈ Ix , such that xi ∈ Cin
0 and the associated functor
≃
Ix Ð→ Cin
x/ ×Cin C0
in
0 = C0 ∩ C
is coinitial (where Cin is the preimage of C0 in Cin ).
in
We will refer to maps of C which belong to Cin and as inert and active respecti-
vely, and to the objects of Cin
0 as the basics.
Remark 4.2.6. Let (C, Cin , Cact , C0 ) be a weak ∞-operad. The condition that
/x ×Cin C0 admits a coinitial map from a finite set is equivalent to the condi-
Cin in
be the full subcategory spanned by those objects which are initial in their compo-
nent. Then Ix is categorically equivalent to a finite set, and the inclusion (4.4) is
coinitial. In particular, up to categorical equivalence we may assume that the Ix in
Definition (4.2.4) are given by the canonical choice of (4.4). If ϕ ∶ C Ð→ D is now a
functor that sends inerts to inerts, actives to actives and basics to basics then the
condition that ϕ is a map of weak ∞-operads is equivalent to the condition that
for every x ∈ C the functor ϕ maps Ix to Iϕ(x) bijectively.
Definition 4.2.7. Let (C, Cin , Cact , Cin
0 ) be a weak ∞-operad and let E be an ∞-
category with finite products. A C-monoid object in E is a functor ϕ ∶ C Ð→ E such
that ϕ∣Cin is a right Kan extension if ϕ∣Cin 0
(see §2.7). By Theorem 2.7.5 this is
equivalent to saying that for every x ∈ C the collection of maps fi ∶ ϕ(x) Ð→ ϕ(xi )
for i ∈ Ix exhibit ϕ(x) as the product in E of the objects {ϕ(xi )}i∈Ix .
Remark 4.2.8. If ϕ ∶ C Ð→ D is a functor of weak ∞-operads in the sense of
Definition 4.2.5 and E is any ∞-category with finite products then restriction along
ϕ sends D-monoids in E to C-monoids in E.
Example 4.2.9. For any factorization system (Cin , Cact ) on C we can choose C0 = C
as our subcategory of basics and get an associated weak ∞-operad. We then have
MonC (E) = Fun(C, E) for every E.
Example 4.2.10. In any ∞-category C we have two canonical factorization systems:
the one where the inerts are the equivalences and the actives are all maps and the
one where the inerts are all maps and the actives are the equivalences. In addition,
any full subcategory C0 ⊆ C can serve as a subcategory of basics for the former, in
which case ψ ∶ C Ð→ E is a monoid object if and only if ψ(c) is terminal in E for
every c ∉ C0 .
Example 4.2.11. If O⊗ Ð→ N(Fin∗ ) is an ∞-operad then the ∞-category O⊗ has an
associated structure of a weak ∞-operad where (O⊗ )in consists of the inert maps,
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 61
(O⊗ )act consists of the active maps, and the subcategory of basics consists of the
objects which lie above ⟨1⟩.
Example 4.2.12. The category ∆op admits a structure of a weak ∞-operad, where
ρ ∶ [n] Ð→ [m] is active if ρ(0) = 0 and ρ(n) = m, is inert if it is of the form
ρ(i) = i + a for some a, and the full subcategory of basics is {[1]}. More generally,
the Leinster category of any perfect operator category is a weak ∞-operad.
There is a map of weak ∞-operads ∆op Ð→ Ass⊗ which sends [n] to the pointed
set {∗, (0, 1), (1, 2), ..., (n − 1, n)} ≅ ⟨n − 1⟩ of consecutive edges in [n] (plus an extra
base point). Restriction along this map induces an equivalence
≃
MonAss⊗ (E) Ð→ Mon∆op (E)
for every ∞-category E with finite products.
Example 4.2.13. If C, D are two weak ∞-operads then the cartesian product C × D
has a naturally associated weak ∞-operad structure, where a map (f, g) ∶ (x, y) Ð→
(x′ , y ′ ) is inert (resp. active) if and only if f is inert (resp. active) in C and g is
inert (resp. active) in D, and the full subcategory of basics is C0 × D0 ⊆ C × D. If E
is an ∞-category with finite products then MonC×D (E) ≃ MonC (MonD (E)).
Definition 4.2.14. Let ϕ ∶ C Ð→ D be a functor of weak ∞-operads. We will say
that ϕ is a weak approximation if the following conditions holds:
(1) For every y ∈ C the homotopy fibers of Cact
/y Ð→ D/ϕ(y) are weakly contractible.
act
Example 4.2.15. The map ϕ ∶ ∆op Ð→ Ass⊗ of Example 4.2.12 is a strong approxi-
mation. This follows from the fact that for every [n] ∈ ∆op the functor (∆op )act
/[n] Ð→
(Ass⊗ )act
/ϕ([n]) is an equivalence of categories and (Ass ⊗ in
⟨1⟩ ) ≃ ϕ−1
(Ass ⊗ in
⟨1⟩ ) ≃ ∗.
Remark 4.2.16. If ϕ ∶ C Ð→ D and ϕ′ ∶ C′ Ð→ D′ are weak approximations of weak
∞-operads then (ϕ, ϕ′ ) ∶ C × C′ Ð→ D × D′ is a weak approximation as well.
Remark 4.2.17. Consider a homotopy pullback square of ∞-categories
ϕ′
P /C
ψ′ ψ
D
ϕ
/E
If C, D and E are given weak ∞-operad structures such that ϕ and ψ are weak
∞-operad maps then P inherits an associated weak ∞-operad structure where a
map in P is inert (resp. active) if and only if its image in C and D is inert (resp.
active) and P0 ∶= D0 ×hE0 C0 . In this case, if ψ is a strong approximation then ψ ′ is
a strong approximation.
Our main interest in weak approximations comes from the following result:
Proposition 4.2.18. Let ϕ ∶ C Ð→ D be a weak approximation map of weak ∞-
operads and let E be an ∞-category which admits small limits. Then the restriction
(4.5) ρ∗ ∶ MonD (E) Ð→ MonC (E)
62 YONATAN HARPAZ
∞-category of factorizations
′′
(4.7) @ y ?? ′′
?? g
??
′
g
??
y
g
/ y′
of g such that g ′ is inert and ϕ(y ′′ ) ∈ D0 . This ∞-category embeds in the ∞-category
′′ ′ ′
Cy//y′ ×Cy/ Cin
y/ of all factorizations g = g ○ g with g inert. By Exercise 4.2.3(2)
the latter has a final object given by any factorzation g = g ′′ ○ g ′ as in (4.7) such
that g ′ is inert and g ′′ is active. In this case, since g and g ′ map to inert maps in
D we have by Exercise 4.2.3(1) that g ′′ must map to a map in D which is both
inert and active, and hence an equivalence. It then follows that ϕ(y ′′ )D0 and so
the factorization (g ′ , g ′′ ) belongs to [Cin
y/ ×C (C0 ) ]/g . We may thus conclude that
ϕ in
the latter ∞-category has final objects and is therefore weakly contractible. The
inclusion (4.6) is consequently coinitial.
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 63
It now follows from Theorem 2.7.5 that ψ∣Cϕ is a right Kan extension of ψ∣Cϕ0
if and only if ψ∣Cin is a right Kan extension of ψ∣(Cϕ0 )in . Since (Cϕ
0)
in
is a full
subcategory of C which contains C0 the latter condition holds whenever ψ is a
in in
gory spanned by the locally constant monoid objects. By definition the restriction
functor (4.5) lands in locally constant C-monoids, thus yielding a functor
(4.8) ρ∗ ∶ MonD (E) Ð→ Monloc
C (E)
The key idea of the proof is that (4.8) admits an inverse which is given by right
Kan extension along ϕ. More precisely, let
π ∶ M ∶= N[ϕ] ([1]) Ð→ ∆1
be the cartesian fibration (2.39) classified by the diagram (∆1 )op Ð→ Cat∞ cor-
ϕ
responding to [D ←Ð C]. As we outlined in §2.7, the ∞-category of functors
M Ð→ E is equivalent to the ∞-category of triples (ψ, ψ ′ , δ) consisting of a functor
ψ ∶ C Ð→ E, a functor ψ ′ ∶ D Ð→ E, and a natural transformation δ ∶ ψ ′ ○ ϕ ⇒ ψ.
Consider the following two properties a functor ψ ∶ M Ð→ E can have:
(1) ψ∣C is a locally constant monoid object and the natural transformation δ ∶
ψ ′ ○ ϕ ⇒ ψ exhibits ψ ′ as a right Kan extension of ψ along ϕ.
(2) ψ∣D is a monoid object and the natural transformation δ ∶ ψ ′ ○ ϕ ⇒ ψ is an
equivalence.
To prove that (4.8) is an equivalence it will suffice to show that for a given ψ ∶
MRϕ Ð→ E, Condition (1) is equivalent to Condition (2). Indeed, assume this claim
for the moment and let X ⊆ Fun(MR ϕ , E) be the full subcategory spanned by those
ψ which satisfy either of those equivalent conditions. Consider the diagram
(4.9) u X HHH
uuu HH
HH
uuu HH
uz u H#
MonC (E)
loc
MonD (E)
Since we assumed that E admits limits we have by Theorem 2.7.5 that every
ϕ ∶ C Ð→ E admits a right Kan extension ψ ∶ M Ð→ E. Since X is determines
by Condition (1) the left diagonal projection in (4.9) is obtained from the cocar-
ϕ , E) ×Fun(C,E) MonC (E) Ð→ MonC (E) by restricting to
loc loc
tesian fibration Fun(MR
those objects which are final in their fiber, and is hence a trivial Kan fibration
by Exercise 2.6.10. On the other hand, since X is determined by Condition (2)
the right diagonal map in 4.9 is obtained by restricting the cartesian fibration
Fun(MR ϕ , E) ×Fun(D,E) MonD (E) Ð→ MonD (E) to the full subcategory spanned by
those objects which are initial in their fiber, and is hence a trivial Kan fibra-
tion as well. In particular, (4.9) determines an equivalence between MonD (E) and
C (E). This implies that (4.8) is an equivalence is well, since we can iden-
Monloc
tify (4.8) up to homotopy with the composition of (any) section MonD (E) Ð→ X
and the projection X Ð→ Monloc C (E).
64 YONATAN HARPAZ
We shall now prove that Condition (1) and Condition (2) are equivalent. Let ψ ∶
M Ð→ E be such that ψ∣C is a locally constant C-monoid object. By Theorem 2.7.5
the condition that ψ is a right Kan extension is equivalent to the condition that for
ψ
every x ∈ D the composed map (Mx/ ×M C)◁ Ð→ M Ð→ E is a limit diagram. Let
Ix ⊆ Mx/ ×M C be the full subcategory spanned by those objects [x Ð→ y] ∈ Mx/ ×M C
whose corresponding arrow f ∶ x Ð→ ϕ(y) in D is inert. We claim that the inclusion
Ix ⊆ Mx/ ×M C is coinitial. Concretely, what we need to check is that for every
arrow in D of the form f ∶ x Ð→ ϕ(y), the comma ∞-category (Ix )/f is weakly
contractible. This comma ∞-category can be identified with the full subcategory
X ⊆ C/y ×D/ϕ(y) Dx//ϕ(y) consisting of the pairs (g ∶ y ′ Ð→ y, h ∶ x Ð→ ϕ(y ′ ))
such that h is inert, and is hence weakly contractible by Lemma 4.2.19 and our
assumption that ϕ is a weak approximation. We then get that the second part
of condition (1) is equivalent to the condition that for every x ∈ D the restricted
ψ
diagram ψ x ∶ I◁
x Ð→ M Ð→ E is a limit diagram. Using again Theorem 2.7.5 this
is equivalent to saying that δ∣Cϕ ∶ ϕ∗ ψ ′ ∣Cϕ ⇒ ψ∣Cϕ exhibits ψ ′ ∣Din as a right Kan
extension of ψ∣Cϕ . Consider the commutative diagram
(4.10) Cϕ / Din
0 0
Cϕ / Din
By Lemma 4.2.20 we have that ψ∣Cϕ is a right Kan extension of ψ∣Cϕ0 . Since Din 0 ⊆
Din is a full inclusion a double application of Proposition 2.7.7 and an application
of Remark 2.7.6 imply that Condition (1) is equivalent to the following condition:
(1’) ψ∣C is a locally constant monoid object, ψ∣D is a monoid object, and the natural
transformation δ∣Cϕ0 ∶ ϕ∗ ψ ′ ∣Cϕ0 ⇒ ψ∣Cϕ0 exhibits ψ ′ ∣Din
0
as a right Kan extension
of ψ∣C0 along ϕ0 ∶ C0 Ð→ D0 .
ϕ
ϕ in
M0 = N[ϕ0 ] ([1]) Ð→ ∆1
be the associated relative nerve, so that we have a natural inclusion of simplicial
sets M0 ⊆ M. Let ψ 0 ∶= ψ∣M0 . Since ψ∣C is a C-monoid object Condition (2)
can be checked only for objects in Cϕ 0 . To finish the proof it will hence suffice to
show that ψ 0 is a right Kan extension if and only if the natural transformation
δ∣Cϕ0 ∶ ϕ∗ ψ ′ ∣Cϕ0 ⇒ ψ∣Cϕ0 is an equivalence. Now since we assumed that ψ∣C is a locally
constant C-monoid object it follows that (ψ 0 )∣Cϕ0 factors through Din 0 . The desired
result now follows from assumption that the restriction functor Fun(Din 0 , E) Ð→
Fun(Cϕ 0 , E) is fully-faithful since C0 Ð→ D0 is assumed to be a localization functor.
ϕ in
We finish this section with the following result which allows one to relate two
different weak ∞-operad structures on the same ∞-category C.
′ ′
Lemma 4.2.21. Let C be an ∞-category and let (Cin , Cact , C0 ) and (Cin , Cact , C1 )
′ ′
two weak ∞-operad structures on C such that Cin ⊆ Cin and Cact ⊆ Cact . Let E be
an ∞-category with finite products. Then the following assertions hold:
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 65
(1) Suppose that C1 = C0 and that for every morphism f ∶ x Ð→ y in Cin such
f′ f ′′ ′
that y ∈ C0 and every factorization x Ð→ y ′ Ð→ y of f such that f ′ is in Cin
we have that y ′ is in C0 . Then ψ ∶ C Ð→ E is a monoid object with respect to
′ ′
(Cin , Cact , C0 ) if and only if it is a monoid object with respect to (Cin , Cact , C0 ).
′ ′
(2) Suppose that Cin = Cin ,Cact = Cact and C0 ⊆ C1 . Then ψ ∶ C Ð→ E is a monoid
object with respect to (Cin , Cact , C0 ) if and only if it is a monoid object with
respect to (Cin , Cact , C1 ) and in addition ψ∣Cin
1
is a right Kan extension of ψ∣Cin 0
.
′ ′
Proof. Let us first prove (1). Let x ∈ C be an object and write Cin 0 = Cin ∩C0 . Then
′ ′
our assumption implies that the inclusion Cin x/ ×C Cin 0 ⊆ Cin
x/ ×C C0 is coinitial. It
in
′ ′
then follows that ψ ∶ C Ð→ E is a monoid object with respect to (Cin , Cact , C0 ) if
and only if it is a monoid object with respect to (Cin , Cact , C0 ).
To prove (2), we first note that the “if” direction follows from the pasting lemma
for right Kan extensions (Proposition 2.7.7). To show the “only if” direction note
1 ⊆ C is a full subcategory which contains C0 we have by Remark 2.7.6
that since Cin in in
(4.11) ∆n × P⊗
τ / C⊗
(σ,π)
∧
N(Fin∗ ) × N(Fin∗ ) / N(Fin∗ )
and such that if f ∶ x Ð→ y is an inert map in P⊗ and i ∈ [n] then τ sends ∆{i} × f
to an inert map in P. One can then check that the projection Alg⊗O (P) Ð→ N(Fin∗ )
given by (σ, τ ) ↦ σ exhibits (AlgO (P))⊗ as an ∞-operad. We will refer to Alg⊗
O (P)
as the ∞-operad of O-algebras in P.
66 YONATAN HARPAZ
(4.12) O⊗ × P⊗
ϕ
/ Q⊗
∧
N(Fin∗ ) × N(Fin∗ ) / N(Fin∗ )
such that for every inert morphism f ∶ x Ð→ y in O⊗ and every inert mor-
phism g ∶ z Ð→ w in P⊗ , the arrow ϕ(f, g) is inert in Q⊗ . We will denote by
BiFun(O⊗ , P⊗ ; Q⊗ ) the full subcategory of FunN(Fin∗ ) (O⊗ × P⊗ , Q⊗ ) spanned by the
bifunctors.
Remark 4.3.4. For ∞-operads O⊗ , P⊗ , Q⊗ we have a canonical isomorphism of sim-
plicial sets
BiFun(O⊗ , P⊗ ; Q⊗ ) ≃ AlgO (AlgP (Q)).
Definition 4.3.5. We will say that a bifunctor ϕ ∶ O⊗ × P⊗ Ð→ Q⊗ exhibits Q⊗
as the tensor product of O and P if for every ∞-operad R⊗ , composition with ϕ
induces an equivalence of ∞-categories
≃
AlgQ (R) Ð→ BiFun(O⊗ , P⊗ ; R⊗ ) ≅ AlgO (AlgP (R)).
It follows by standard arguments that if a tensor product of O⊗ and P⊗ exists
then it is essentially unique. To show that a tensor product always exists it is
convenient to employ the model structure of Theorem 4.1.19. In particular, if O⊗
and P⊗ are two ∞-operads then we may consider the cartesian product of marked
simplicial sets O⊗,♮ × P⊗,♮ as an object of (Set+∆ )/(N(Fin∗ ),Ein ) via the composed map
∧
O⊗,♮ × P⊗,♮ Ð→ (N(Fin∗ ), Ein ) × (N(Fin∗ ), Ein ) Ð→ (N(Fin∗ ), Ein ).
The following is the a formal consequence of the model categorical setup of Theo-
rem 4.1.19:
Proposition 4.3.6. A bifunctor ϕ ∶ O⊗ × P⊗ Ð→ Q⊗ exhibits Q⊗ as the tensor
product of O⊗ and P⊗ if and only if the associated map
O⊗,♮ × P⊗,♮ Ð→ Q⊗,♮
is a weak equivalence in (Set+∆ )/(N(Fin∗ ),Ein ) with respect to the operadic model struc-
ture of Theorem 4.1.19.
We note that Proposition 4.3.6 implies in particular that a tensor product of
O⊗ × P⊗ always exists: just take any fibrant replacement of O⊗,♮ × P⊗,♮ in the
operadic model structure.
Remark 4.3.7. Let ϕ ∶ O⊗ × P⊗ Ð→ Q⊗ be a bifunctor of ∞-operads such that the
induced map (O⊗ )⟨1⟩ ×(P⊗ )⟨1⟩ Ð→ Q⊗⟨1⟩
is essentially surjective. It then follows from
Proposition 4.1.23 that ϕ exhibits Q⊗ as the tensor product of O⊗ and P⊗ if and if
the induced map
MonQ (S) Ð→ MonO (MonP (S))
is an equivalence of ∞-categories.
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 67
Unfortunately, fibrant replacements in the operadic model structure are not very
explicit in general and not easy to compute. One way to overcome this issue is by
using the construction of wreath products.
Construction 4.3.8. Let O⊗ , P⊗ be two ∞-operads. Then P⊗ is in particular
an ∞-category and we may defined the simplicial set (P⊗ )∐ as in definition 3.3.1.
We note that (P⊗ )∐ is now equipped with two different maps to N(Fin∗ ): one is
the projection π ∶ (P⊗ )∐ Ð→ N(Fin∗ ) associated to the construction (−)∐ as in
Definition 3.3.1, and the other is the composed map
ρ ∶ (P⊗ )∐ Ð→ N(Fin)∐ Ð→ N(Fin),
where the first map is induced by the ∞-operad structure P⊗ Ð→ N(Fin∗ ) and
the second is the map N(Fin)∐ Ð→ N(Fin) which sends a tuple (⟨n1 ⟩ , ..., ⟨nk ⟩) to
⟨n1 + ... + nk ⟩. We then define the wreath product of O⊗ and P⊗ to be the fiber
product
O⊗ ≀ P⊗ / (P⊗ )∐
π
O⊗ / N(Fin∗ )
and we equip O⊗ ≀ P⊗ with the map to N(Fin∗ ) determined by the projection to
(P⊗ )∐ followed by ρ ∶ (P⊗ )∐ Ð→ N(Fin∗ ).
We can identify an object of O⊗ ≀ P⊗ with a tuple (x, (y 1 , ..., y n )) where x is
an object of O⊗ lying above ⟨n⟩ ∈ N(Fin∗ ) and y 1 , ..., y n are objects of P⊗ . A
morphism in O⊗ ≀ P⊗ from (x, (y 1 , ..., y n )) to (x′ , (y ′1 , ..., y ′n′ )) is given by a map
f ∶ x Ð→ x′ in O⊗ lying above a map α ∶ ⟨n⟩ Ð→ ⟨n′ ⟩ together with a collection of
○
maps gi ∶ y i Ð→ y ′α(i) in P⊗ for every i ∈ α−1 (⟨n′ ⟩ ). We will say that such a map
is inert if f is inert in O⊗ and each gi is inert in D⊗ .
The projection Γ∗ Ð→ Fin∗ determines a map P⊗ × N(Fin∗ ) Ð→ (P⊗ )∐ given
informally by the formula (y, ⟨n⟩) ↦ (y, ..., y). We then obtain a commutative
diagram
(4.13) O⊗ × PL⊗
ϕ
/ O⊗ ≀ P⊗
LL rr
LL r
LL rrr
LL rr
L& yrr
N(Fin∗ )
Theorem 4.3.9 ([6, Theorem [Link]]). Let O⊗ and P⊗ be two ∞-operads and
let Ein be the set of of inert arrows in O⊗ ≀ P⊗ . Then (4.13) determines a weak
equivalence
O⊗,♮ × P⊗,♮ Ð→ (O⊗ ≀ P⊗ , Ein )
in the operadic model structure of Theorem 4.1.19.
We will not prove Theorem 4.3.9 here. We will however prove the following
variant, which is sufficient for our needs, concerning the closely related notion of
monoid objects. For this, let us consider the ∞-categories O⊗ × P⊗ and O⊗ ≀ P⊗ as
weak ∞-operads (see §4.2). More precisely, we may consider each of O⊗ and P⊗
as a weak ∞-operad by Example 4.2.11, and then consider the cartesian product
O⊗ ×P⊗ as a weak ∞-operad by Example 4.2.13. To put a weak ∞-operad structure
on the wreath product we use the following construction:
68 YONATAN HARPAZ
has weakly contractible homotopy fibers. But this is true because the map (4.14)
is in fact an equivalence of ∞-categories: both the left and right hand side are
equivalent to (O⊗ ⊗
act )/x because the active maps in the coarse structure on P are
only the equivalences. To show that ϕ is in fact a strong approximation we now
note that the subcategory of coarse-basics and coarse-inert maps in O⊗ ≀ P⊗ is
(O⊗⟨1⟩
)≃ ×{⟨1⟩} P⊗ ⊆ O⊗ ×N(Fin) (P⊗ )∐ , and its inverse image in O⊗ × P⊗ is exactly
the same ∞-category, considered as a subcategory of O⊗ × P⊗ in the obvious way.
We thus established that ϕ is a strong approximation. It then follows from
Proposition 4.2.18 that restriction along ϕ induces an equivalence
≃
O⊗ ≀P⊗ (E) Ð→ MonO⊗ ×P⊗ (E) ≃ MonO⊗ (FunP⊗ (E)).
Moncoarse coarse
(4.15)
We now wish to compare coarse O⊗ ≀P⊗ -monoid objects with fine O⊗ ≀P⊗ -monoid
objects. Define the mixed weak ∞-operad structure O⊗ ≀ P⊗ to be the one in which
the active-inert factorization is the one of the fine structure while the subcategory
of basics is the one of the coarse. In particular, the subcategory of mixed basics
≃
with mixed inert maps can be identified with (O⊗ ⟨1⟩
) × (P⊗ )in . We may readily
verify that this is indeed a weak ∞-operad structure. Applying Lemma 4.2.21(1)
we may deduce that a functor ψ ∶ O⊗ ≀ P⊗ Ð→ E is a monoid object with respect to
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 69
the coarse structure if and only if it is a monoid object with respect to the mixed
structure. Applying Lemma 4.2.21(2) we may identify fine monoid objects with the
full subcategory
such that
MulE∆n
({◻n }i∈I , ◻n ) = Sing(Rect(◻n × I, ◻n )),
where Sing denotes the singular simplicial set functor. Composition is given by the
composition of rectilinear embeddings.
We note that since the singular complex of a space is always a Kan the simplicial
operad E∆
n is locally Kan.
Definition 5.1.3. We define the little n-cube ∞-operad to be the operadic nerve
E⊗ ∆ ⊗
n ∶= N((En ) ).
Variant 5.1.4. Definition 5.1.2 could naturally be made (and historically this was
the way it was done) in the setting of topological operads rather than simplicial,
that is, operads in which the sets of multimaps are replaced by topological spaces.
In particular, we may define the topological variant Etop
n of the little n-cube operad
to be the topological operad with a single object ◻n and such that
MulEtop
n
({◻n }i∈I , ◻n ) = Rect(◻n × I, ◻n ).
We then have that E∆ top
n is obtained form En by applying the functor Sing to all
spaces of multimaps.
70 YONATAN HARPAZ
Variant 5.1.5. For n ≥ 1 the little n-cube ∞-operads admit a natural variant where
we also consider cubes with partial boundary. More precisely, let ◻n∂ = [0, −1)×◻n−1
n,,∂ denote the simplicial operad with two objects object ◻ , ◻∂ and such
and let E∆ n n
that
MulE∆n
({Xi }i∈I , Y ) = Sing(Rect(∐ Xi , Y )),
i
for Xi , Y ∈ {◻n , ◻n∂ }. We note that there are no open embeddings from ◻n∂ to ◻n
and that every open embedding from ◻n to ◻n∂ is contained in (0, 1) × ◻n−1 . We
define E⊗ ∆
n,∂ to be the operadic nerve of En,∂ , and refer to it as the operad of little
n-cubes with boundary.
Example 5.1.6. When n = 0 we have ◻0 = (−1, 1)0 ≃ ∗, the space with one point
by convention. For a finite set I the space Rect(◻0 × I, ◻0 ) is empty if ∣I∣ > 1 and
consist of a single point if ∣I∣ ≤ 1. In particular, E∆
0 is isomorphic to the operad Poi
of Example 4.1.7(5), whose algebras are pointed objects, and E⊗ 0 is isomorphic to
the operadic nerve of Poi.
Example 5.1.7. When n = 1 we have ◻1 = (−1, 1). In this case every rectilinear
embedding f ∶ ◻1 × I Ð→ ◻1 determines a unique linear ordering on I by i ≤ i′ ⇔
f (0, i) ≤ f (0, i′ ). This association determines a map of simplicial operads
(5.2) 1 Ð→ Ass,
E∆
which in turn determines a map of simplicial categories
⊗ ⊗
(5.3) (E∆
1 ) Ð→ Ass .
Lemma 5.1.10. Let I be a finite set. Then the map ev0 ∶ Rect(◻n × I, ◻n ) Ð→
Conf(I, M ) obtained by restricting along I = {(0, ..., 0)} × I ↪ ◻n × I is a homotopy
equivalence.
Proof. Given a configuration f ∶ I Ð→ ◻n let us denote by
ε(f ) = min min ∣fj (i) − fj (i′ )∣
i,i ∈I j=1,...,n
′
the minimum distance between the points in f (I) (in the taxi-driver metric) and
by
ε′ (f ) = min min (1 − ∣fj (i′ )∣)
i∈I j=1,...,n
the minimum distance between the points in f (I) and the boundary of the cube.
Let ε′′ (f ) = min(ε(f2),ε (f )) and let
′
ϕ ∶ Conf(I, M ) Ð→ Rect(◻n × I, ◻n )
be the map which sends f ∶ I Ð→ ◻n to the rectilinear embedding
ϕf (x1 , ..., xn , i) = f (i) + ε′′ (f )(x1 , ..., xn ).
Then ϕ is continuous (since min is continuous) and ev0 ϕf = f . To finish the proof it
suffices to note that the map g ↦ ϕev0 (g) from Rect(◻n × I, ◻n ) to Rect(◻n × I, ◻n )
is homotopic to the identity. An explicit homotopy is given by
Ht (g)(x1 , ..., xn , i) = (bi1 + ai1 (t)x1 , ..., bin + ain (t)xn ),
where aij (t) = taij +(1−t)ε′′ (ev0 (g) and aij , bij are the constants determining g∣◻n ×{i} .
Corollary 5.1.11. The space Rect(◻n × I, ◻n ) ≃ Conf(I, ◻n ) is (n − 2)-connected.
Proof. We argue by induction on ∣I∣. If I = ∅ then the space Conf(I, ◻n ) is a
singleton. Otherwise, let i0 ∈ I be an element and consider the projection
(5.4) Conf(I, M ) Ð→ Conf(I ′ , M ).
where I ′ = I ∖ {i0 }. Then (5.4) is a Serre fibration whose base is (n − 2)-connected
by the induction hypothesis. By the long exact sequence in homotopy groups it
will suffice to show that the fibers are (n − 2)-connected. Now the fiber over a
configuration f ∶ I ′ Ð→ ◻n is given by ◻n ∖ f (I ′ ). To finish the proof we hence need
to verify that the complement of a finite number of points in ◻n is (n−2)-connected.
In fact, it is a classical result that the complement of k points in ◻n is homotopy
equivalent to a wedge of k spheres of dimension n − 1. We note that when n ≤ 2
this can be verified by hand. When n ≥ 3 the proof is classical using Van-Kampen’s
theorem and Alexander’s duality.
For every n ≥ 0 the identification ◻n+1 ≅ ◻n × (−1, 1) determines a map of simpli-
n Ð→ En+1 which acts on spaces of operations by taking the product
cial operads E∆ ∆
⊗
with (−1, 1). We then obtain a map of ∞-operads E⊗ n Ð→ En+1 . Using Corol-
lary 5.1.11 and the fact that weak equivalences in the operadic model structure of
Theorem 4.1.19 are closed under filtered colimits we can now deduce the following:
Corollary 5.1.12. The colimit of the sequence
(5.5) E0 Ð→ E1 Ð→ E2 Ð→ ...
in the ∞-category Op∞ is equivalent to the terminal ∞-operad Com⊗ .
72 YONATAN HARPAZ
be the map of simplicial categories which sends the object (⟨m⟩ , ⟨l⟩) to the object
⟨m ⋅ l⟩ and for every pair of maps α ∶ ⟨m⟩ Ð→ ⟨m′ ⟩ and β ∶ ⟨l⟩ Ð→ ⟨l′ ⟩ sends the
component
−1 −1
∏ Sing Rect(◻ × α (i), ◻ ) × ∏ Sing Rect(◻ × β (j), ◻ )
n n k k
i∈⟨m′ ⟩○ j∈⟨l′ ⟩○
′ ′
of Map(E∆ ∆ ⊗ ((⟨m⟩ , ⟨l⟩), (⟨m ⟩ , ⟨l ⟩)) to the component
n ) ×(E )
⊗
k
∧
N(Fin∗ ) × N(Fin∗ ) / N(Fin∗ )
Theorem 5.2.2 (Dunn’s additivity theorem). The bifunctor (5.6) exhibits En+k
as the tensor product of En and Ek .
The remainder of this section is devoted to the proof of Theorem 5.2.2. Let
us consider En+k as a weak ∞-operad with its operadic weak structure of Exam-
ple 4.2.11, and En × Ek as a weak ∞-operad endowed with the product structure of
Example 4.2.13. Combining Proposition 4.1.23 and Proposition 4.1.21 it will suffice
to show that restriction along (5.6) induces an equivalence of ∞-categories
≃
(5.7) MonE⊗ (S) Ð→ MonE⊗n ×E⊗ (S) ≃ MonE⊗n (MonEk (S)),
n+k k
where S denotes the ∞-category of spaces. Our first step is to reduce the higher
categorical complexity of the problem by replacing E⊗
n by a suitable discrete model.
Definition 5.2.3. Let D⊗ n be the operadic nerve of the ordinary colored operad
whose objects are the open subcubes of ◻n (i.e., the images of rectilinear embed-
dings). Given a collection of subcubes U1 , ..., Un ⊆ ◻n and an additional subcube
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 73
It can then be shown that the geometric realization of this simplicial space is a
model for the homotopy colimit 5.10. The following can the be proven using clas-
sical methods (in particular, partitions of unity, which is why the assumption of
paracompactness was needed):
Theorem 5.2.7 (see, e.g.,[3, Corollary 4G.2]). Let X be a paracompact topological
spaces and P ⊆ O(X) an open covering. Then the natural map
∣C● (P )∣ Ð→ X
is a homotopy equivalence.
Remark 5.2.8. In Theorem 5.2.7 we did not assume that P is closed under inter-
section since this was not assumed in the classical setting and is not needed for
the claim of Theorem 5.2.7 to hold. The condition that P is closed under finite
intersections is only needed for the identification of ∣C● (P )∣ with hocolimU ∈P U .
Remark 5.2.9. Suppose that P ⊆ O(X) is an open covering which satisfies the
following property: for every x ∈ X and every U, V ∈ P there exists a W ∈ P
such that x ∈ W ⊆ U ∩ V . Then for every x ∈ X the subposet {U ∈ P ∣x ∈ U } is
filtered (every finite subset has a lower bound) and is hence weakly contractible.
In this particular case one can construct a hypercovering U● Ð→ X such that ∣U● ∣ ≃
hocolimU ∈P U and prove Theorem 5.2.5 using direct arguments (not invoking ∞-
topos machinery), see [2].
Corollary 5.2.10. Let M be a manifold and O(M ) the poset of open subsets of
M . Let I be a finite set and P ⊆ O(M )I be a subposet satisfying the following
properties:
(1) For every (Ui )i∈I ∈ P , we have Ui ∩ Uj = ∅ for i ≠ j.
(2) For each configuration f ∶ I Ð→ M there exists an element (Ui )i∈I ∈ P such
that f (i) ∈ Ui .
(3) If (Ui )i∈I and (Vi )i∈I are two elements in P and f ∶ I Ð→ M is a cofiguration
such that f (i) ∈ Ui ∩ Vj then there exists a (Wi )i∈I ∈ P such that f (i) ∈ Wi .
Then the natural map
hocolim ∏ Conf({i}, Ui ) Ð→ Conf(I, M )
(Ui )i∈I ∈P i∈I
is a homotopy equivalence.
Proof. The association (Ui )i∈I ↦ ∏i∈I Conf({i}, Ui ) determines an injective fully-
faithful map from the poset P to the poset of open subsets of Conf(I, M ). The
desired result is now provided by Theorem 5.2.5 (in the situation of Remark 5.2.9).
has weakly contractible homotopy fibers. We first observe that both the left and
right hand sides are products of the corresponding ∞-categories for the individual
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 75
The right diagonal map is a right fibration whose fiber over the object ⟨l⟩ ∈ (E⊗ n)
act
⊗ act
the homotopy fiber of the left diagonal map over ⟨l⟩ ∈ (En ) . Unwinding the de-
finitions we see that (D⊗ n )/U can be identified with the subposet of O(U ) spanned
act
by those open subsets which are finite disjoint unions of subcubes. Let us call such
open subsets multi-subcubes. Under this identification the left diagonal map can be
written as V ↦ π0 (V ). We may then identify Xl with the ∞-category of pairs (V, ρ)
○
where V is a multi-subcube of U and ρ ∶ ◻n × ⟨l⟩ Ð→ V is a rectilinear embedding
which induces a bijection of π0 . Equivalently, we can describe the objects of Xl as
pairs ((V1 , ..., Vl ), (ρ1 , ..., ρl )) where V1 , ..., Vl are pairwise disjoint open subcubes of
U and ρi ∶ ◻n Ð→ Vi is a rectilinear embedding. Let us denote by X l ⊆ O(U )⟨l⟩ the
○
subposet spanned by the pairwise disjoint tuples of subcubes (V1 , ..., Vl ) ∈ O(U )⟨l⟩ ,
○
is a weak homotopy equivalence. But this now follows from Corollar 5.2.10 since
every intersection of cubes is either empty or a cube, and every cofiguration of
points is covered by at least one tuple of subcubes.
Corollary 5.2.11. The restriction functor
MonE⊗n (S) Ð→ MonD⊗n (S)
is fully-faithful and its essential image is spanned by those D⊗n -monoid objects ψ ∶
D⊗n Ð→ S which are locally constant in the sense that the restriction to (D⊗
n )⟨1⟩
sends every arrow to an equivalence of spaces.
We now note that the bifunctor of ∞-operads of Definition 5.2.1 can also be
defined on the level of D⊗
n . Indeed, it is simply given by the functor of ordinary
categories
(5.13) D⊗ ⊗ ⊗
n × Dk Ð→ Dn+k
which sends ((Ui )i∈⟨m⟩○ , (Vj )j∈⟨l⟩○ to (Ui × Vj )(i,j)∈⟨m⟩○ ×⟨l⟩○ (where we identified
○ ○
⟨m⟩ × ⟨l⟩ with (⟨m⟩ ∧ ⟨l⟩)○ ).
76 YONATAN HARPAZ
Let us now recall from §4.3 the formation of wreath products. We then observe
that (5.13) factors as
D⊗ ⊗ ⊗ ⊗
n × Dk Ð→ Dn ≀ Dk Ð→ Dn+k
⊗
has weakly contractible homotopy fibers. We first observe that both the left and
right hand sides of (5.15) factor as the product over i = 1, ..., m of the respective
∞-categories for Xi ∶= (Ui , {V1i , ..., Vi11 }) on the left and li on the right. We may
hence assume that m = 1. Consider the commutative triangle
(D⊗ ⊗ act / (E⊗ )act
(5.16) n ≀ Dk )/(U,{V1 ,...,Vl }) n+k /⟨l⟩
QQQ rr
QQQ rr
QQQ rrr
QQQ
( xrrr
(E⊗
n+k )
act
The right diagonal map is a right fibration whose fiber over the object ⟨l′ ⟩ ∈
(E⊗n+k )
act
is a Kan complex which is naturally equivalent to the singular complex of
○ ○
Rect(◻n × ⟨l′ ⟩ , ◻n × ⟨l⟩ ). It will hence suffice to show that (5.12) induces a weak
homotopy equivalence between homotopy fibers above every object ⟨l′ ⟩ ∈ (E⊗ n+k )
act
.
⊗ ⊗
Let us denote by Xl′ Ð→ (Dn ≀ Dk )/(U,{V1 ,...,Vl }) the homotopy fiber of the left
act
○ ○
X = U × V ⊆ ◻n × ◻k × ⟨l⟩ = ◻n+k × ⟨l⟩
equipped with its two projections p ∶ X Ð→ U and q ∶ X Ð→ V . Let O(X, U ) denote
the poset of pairs (W, U ′ ) where W is an open subset of X and U ′ is open subset of
U which contains the image of W via the projection p ∶ X Ð→ U , where we consider
O(X, U ) as ordered by inclusion. The association
((U1 , ..., Um ), ({V11 , ..., Vl11 }, ..., {V1m , ..., Vlm
m
})) ↦ (⋃ Ui × Vji , ⋃ Ui )
i,j i
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 77
⊗ act
identifies (D⊗n ≀ Dk )/(U,{V1 ,...,Vl }) with the subposet O
cube
(X, U ) ⊆ O(X, U ) spanned
′
by those pairs (W, U ) which satisfy the following properties:
(1) U ′ is a disjoint union of subcubes U1 , ..., U1 ⊆ U .
(2) For each component Ui ⊆ U ′ the inverse image W ∩ p−1 (Ui ) ⊆ X of Ui in W
factors as a product Ui × V ′ where V ′ ⊆ V is a (possibly empty) union of
subcubes V1i , ..., Vlii .
⊗ act
Under the identification (D⊗ n ≀ Dk )/(U,{V1 ,...,Vl }) ≃ O
cube
(X, U ) the left diagonal map
′
in (5.16) can be written as (W, U ) ↦ π0 (W ). We may then identify Xl′ with the ∞-
○
category of pairs (W, U ′ , ρ) where (W, U ′ ) ∈ Ocube (X, U ) and ρ ∶ ◻n ×⟨l′ ⟩ Ð→ W is a
rectilinear embedding which induces a bijection of π0 . Equivalently, we can describe
the objects of Xl as tuples ((W, U ′ ), η, (ρ1 , ..., ρl )) where (W, U ′ ) ∈ Ocube (X, U ), η ∶
○ ≅
⟨l′ ⟩ Ð→ π0 (W ) is a bijection of sets and ρi ∶ ◻n+k Ð→ Wiη is a rectilinear embedding,
where by Wiη we mean the component of W determined by η(i) ∈ π0 (W ).
Let X0l′ be the full subcategory of Xl′ spanned by those ((W, U ′ ), η, (ρ1 , ..., ρl ))
such that the map W Ð→ U ′ is surjective on π0 . The inclusion X0l′ ⊆ Xl′ admits a
right adjoint Xl′ Ð→ X0l′ which sends ((W, U ′ ), η, (ρ1 , ..., ρl )) to ((W, p(W )), η, (ρ1 , ..., ρl )),
and is hence a weak homotopy equivalence. It will hence suffice to show that the
map
X0l′ Ð→ (E⊗ n+k )/⟨l⟩ ×(E⊗ )act {⟨l ⟩}
act ′
n+k
implicitly. To finish the proof it is left to check that the canonical map
′ ○
∏ Rect(◻ × {i}, Wi ) Ð→ Rect(◻ × ⟨l ⟩ , X)
n n
hocolim 0
(W1 ,...,Wl′ )∈Xl′ i∈⟨l′ ⟩○
0
object (Uα(j) × Vj )j∈⟨l′ ⟩○ ∈ Ocube (X)⟨l ⟩ then belongs to Xl′ and the corresponding
′ ○
○
open subset of Conf(⟨l′ ⟩ , X) contains the configuration f , as desired.
where the right horizontal maps are equivalences by Theorem 4.3.11. By Proposi-
tion 4.2.18 and Proposition 5.2.12 the left diagonal map in (5.17) is fully-faithful
⊗
and its essential image is spanned by those monoid objects ψ ∶ D⊗ n ≀ Dk Ð→ S
⊗ ⊗
which send every arrow in (Dn ≀ Dk )⟨1⟩ to an equivalence. On the other hand, by
Proposition 4.2.18, Proposition 5.2.4 and Remark 4.2.16 the right most vertical
arrow is fully-faithful and its essential image is spanned by those monoid objects
⊗ ⊗
ψ ∶ D⊗ ⊗
n ×Dk Ð→ S which send every arrow in (Dn )⟨1⟩ ×(Dk )⟨1⟩ to an equivalence. To
⊗
finish the proof it will hence suffice to show that if a monoid object ψ ∶ D⊗ n ≀Dk Ð→ S
⊗ ⊗
sends every arrow in (Dn )⟨1⟩ × (Dk )⟨1⟩ to an equivalence then it sends every arrow
⊗ ⊗
in (D⊗ ⊗
n ≀ Dk )⟨1⟩ to an equivalence. Indeed, let ψ ∶ Dn ≀ Dk Ð→ S be such a monoid
⊗
object. Then for every object ((U1 , ..., Um ), ({}, ..., {V }, ..., {})) in (D⊗
n ≀ Dk )⟨1⟩ the
map
≃
ψ((U1 , ..., Um ), ({}, ..., {V }, ..., {})) Ð→ ψ(Ui0 , V ) ∏ ψ(Ui , {})
i0 ≠i∈⟨m⟩○
induced by the inert map (U1 , ..., Um ), ({}, ..., {V }, ..., {}) Ð→ (Ui0 , V ) and the inert
maps (U1 , ..., Um ), ({}, ..., {V }, ..., {}) Ð→ (Ui , {}) for i ≠ i0 , is an equivalence. In
addition, the monoid condition also implies that ψ(Ui , {}) ≃ ∗. We may then
conclude that ψ sends every inert map of the form
(5.18) ((U1 , ..., Um ), ({}, ..., {V }, ..., {})) Ð→ (Ui0 , V )
to an equivalence. By the 2-out-of-3 rule we then get that ψ sends every map
(5.19) (U ′ , V ′ ) Ð→ ((U1 , ..., Um ), ({}, ..., {V }, ..., {}))
⊗
whose domain is in (D⊗ n )⟨1⟩ × (Dk )⟨1⟩ to an equivalence. We now observe that any
⊗ ⊗
object in (Dn ≀ Dk )⟨1⟩ receives at least one map from an object in (D⊗
⊗
n )⟨1⟩ × (Dk )⟨1⟩ .
Applying again the 2-out-of-3 argument we may conclude that ψ sends every map
⊗
in (D⊗
n ≀ Dk )⟨1⟩ to an equivalence, as desired.
Remark 5.2.14. A variant of the argument in the proof of Theorem 5.2.2 shows that
E⊗ ⊗ ⊗
n+k,∂ ≃ En,∂ ⊗ Ek . In particular, by example 5.1.8 we may identify En,∂ -algebras
in C with En -algebras in AlgE1,∂ (C) ≃ AlgPRAss (C).
5.3. May’s recognition principle. Let (X, x0 ) be a pointed connected topologi-
cal space. Then the loop space
ΩX ∶= {p ∶ [−1, 1] Ð→ X∣p(−1) = p(1) = x0 },
equipped with the compact open topology, acquires a natural action of the topologi-
cal little 1-cube operad Etop
1 (see Variant 5.1.4), which can be described as follows:
given a rectilinear embedding f ∶ ◻1 × I Ð→ ◻1 , the corresponding map
f∗ ∶ ∏ ΩX Ð→ ΩX
i∈I
sends {pi }i∈I to the path q ∶ [−1, 1] Ð→ X which is given by q(x) = pI (f −1 (t)) if
t is in the image of f and q(t) = x0 otherwise. Here we consider ◻1 = (−1, 1) as a
subspace of [−1, 1] and pI ∶ ◻1 × I Ð→ X is the map pI (t, i) = pi (t). More generally,
the n-fold loop space
Ωn X ∶= Ω(Ω(⋯(ΩX))) ≅ {p ∶ [−1, 1]n Ð→ X∣p(∂[−1, 1]n ) = x0 }
acquires a similarly defined action of Etop
n . A famous theorem of May (related to a
large body of work of Boardman and Vogt on the topic) identifies the Etop
n -monoid
objects in Top which can be obtained in this way. To formulate May’s result we
will need the following definition:
Definition 5.3.1. Let X be a topological space equipped with an action of Etop
n
for n ≥ 1. We will say that X is group-like if for every rectilinear embedding
f ∶ ◻n × {1, 2} Ð→ ◻n the map
(pr1 ,f∗ )
X ×X / X ×X
is a homotopy equivalence, where pr1 ∶ X × X Ð→ X is the projection to the first
coordinate and f∗ ∶ X × X Ð→ X is the map associated to f by the action of Etop
n .
Theorem 5.3.2 (May [7]). Let X be a space equipped with an action of Etop
n . Then
the following conditions are equivalent:
(1) X is of the form Ωn Y for some pointed (n − 1)-connected topological space Y .
(2) X is group-like.
Theorem 5.3.2 gives an important relation between the theory of Etop n -monoids
in spaces and the theory of iterated loop spaces, and was historically one of the
motivations behind the introduction of the little n-cube operads. It is often referred
to as May’s recognition principle, since it can be considered as giving an internal
characterization of n-fold loop spaces in terms of the structure they acquire, which
then allows one to recognize them without knowing any explicit delooping. We
would like to formulate and prove May’s theorem in the setting of ∞-operads. For
this, let us first define the notion of being group-like in a more general context:
Definition 5.3.3. Let ψ ∶ E⊗ n Ð→ E be an En -monoid object in E for n ≥ 1. We
will say that ψ is group-like if for every active f ∶ ⟨2⟩ Ð→ ⟨1⟩ in E⊗
n the map
(ρ1∗ ,f∗ )
(5.20) ψ(⟨2⟩) / ψ(⟨1⟩) × ψ(⟨1⟩)
is an equivalence.
80 YONATAN HARPAZ
(5.22) {x} × X
fx
/ {x} × X
operad (see Example 4.2.12) and we have a strong approximation map ∆op Ð→ Ass⊗
(see Example 4.2.15). By Proposition 4.2.18 the restriction functor
≃
MonAss (E) Ð→ Mon∆op (E)
is an equivalence of ∞-categories. We note that we may also model En -monoids
in this way using the additivity theorem. More precisely, let us consider the n-
fold product (∆op )n = ∆op × ... × ∆op equipped with the product weak ∞-operad
structure of Example 4.2.13. By Remark 4.2.16 the map (∆op )n Ð→ (Ass⊗ )n is
again a strong approximation and so
(5.23) Mon(∆op )n (E) ≃ Mon(Ass⊗ )n (E) ≃ Mon(E⊗1 )n (E) ≃ MonEn (E)
by Theorem 5.2.2.
We can also construct a similar type of strong approximation for the ∞-operad
E0 . Consider the one-arrow category [1] = ● / ● equipped with the weak ∞-
operad structure in which all maps are active, only the isomorphisms are inert,
and the basics are {1} ⊆ [1]. We have a natural map of weak ∞-operads [1] Ð→
Poi⊗ which sends 0 to ⟨0⟩ and 1 to ⟨1⟩. This map a strong approximation since
it induces an equivalence on active over-categories (and the second condition of
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 81
exhibits y as the loop object of x at p if ∗ is a final object and the square is cartesian
in E. In this case we will also informally write y ≃ Ωp x, or even y ≃ Ωx if the base
point is implied.
Example 5.3.9. Consider the category Top of topological spaces equipped with the
model structure in which the weak equivalences are the weak homotopy equivalences
and the fibrations are Serre fibrations. Given a pointed space (X, x0 ), the inclusion
{x0 } Ð→ X can be replaced with a fibration ev1 ∶ Px0 Ð→ X where Px0 is the space
of paths in X starting from x0 , and ev1 is given by evaluating at the end point.
82 YONATAN HARPAZ
One can then compute the homotopy pullback {x0 } ×hX {x0 } as the fiber product
Px0 ×X {x0 } ≅ ΩX. In particular, Definition 5.3.8 specializes to the classical notion
of a loop space in the case E = S.
Let now E be an ∞-category which admits final objects and pullbacks. We would
like to express the idea that loop objects Ωp x in E carry a canonical structure of
a group-like E1 -monoid object (in particular, this structure has nothing specific to
do with topological spaces). For this it will be convenient to work in the following
setting: let ∆+ denote the enlargement of ∆ obtained by including also the empty
linearly ordered set [−1] ∶= ∅ ∈ ∆op
+ . For −1 ≤ k ≤ m let us denote by ∆[m,k] ⊆ ∆+
the full subcategory spanned by the objects [k], [k + 1], ..., [m], so that ∆[k,m] ⊆ ∆
when k ≥ 0. We note that ∆[−1,0] has two objects and a unique non-zero arrow
[−1] Ð→ [0]. In particular, N(∆[−1,0] ) ≅ ∆1 . We then have the following:
Proposition 5.3.10. Let E be an ∞-category with a final object ∗ ∈ E and let
ψ ∶ N(∆op+ ) Ð→ E be a diagram. Then the following are equivalent:
(1) ψ([0]) ≃ ∗ and ψ is a right Kan extension of ψ∣∆op
[−1,0]
.
op
(2) ψ∣∆op is a group-like ∆ -monoid object and ψ∣(∆[−1,1] )in is a right Kan exten-
op
sion of ψ∣(∆op
[0,1]
)in .
is an isomorphism, and hence the second part of Condition (1) is equivalent to the
condition that ψ ′ ∣(∆op + )
in is a right Kan extension of ψ ∣∆op
′
[−1,0]
. By the pasting lemma
for Kan extensions (Proposition 2.7.7) and Remark 2.7.6 we see that Condition (1)
is equivalent to the conjunction of the following three conditions
(1a) ψ([0]) ≃ ∗;
(1b) ψ∣(∆op
+ )
in is a right Kan extension of ψ∣(∆op
[−1,1]
)in ; and
(1c) ψ∣(∆op
[−1,1]
)in is a right Kan extension of ψ∣∆op [0,1]
.
Similarly, Condition (2) above is equivalent to the conjunction of Condition (1c)
and the following three conditions:
(2a) ψ∣(∆op )in is a right Kan extension of ψ∣(∆op[0,1]
)in ;
(2b) ψ∣(∆[0,1] )in is a right Kan extension of ψ∣{1} ; and
op
(2c) the map ψ([2]) Ð→ ψ({0, 1}) × ψ({0, 2}) of (5.24) is an equivalence.
Now for n ≥ 2 the full subcategory inclusions
(5.25) (∆op )in
[n]/ ×∆op ∆[0,1] ⊆ (∆+ )[n]/ ×∆+ ∆[−1,1]
op op in
op
op
is coinitial: this follows from the fact that the only object in the right hand side
which is not in the left hand side is final and the left hand side is weakly contractible.
It then follows from Theorem 2.7.5 that Condition (1b) is equivalent to Condition
(2a). Furthermore, since there are no inert maps from [0] to [1] in ∆op , the
same theorem implies that Condition (1a) is equivalent to Condition (2b). To
finish the proof it will suffice to show that Condition (1) implies Condition (2c).
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 83
Let I ⊆ (∆op
+ )[2]/ be the full subcategory containing (∆+ )[2]/ ×∆op
in op
+
∆op
[−1,0]
and in
addition the objects corresponding to the inclusions {0, 1}, {0, 2} ⊆ {0, 1, 2}. Let
ψ
ψ ′ ∶ I Ð→ ∆op
+ Ð→ E
be the composed diagram. Condition (1) implies that ψ ′ is a right Kan extension
of its restriction to (∆op
+ )[2]/ ×∆op
+
∆op
[−1,0]
and so by Theorem 2.7.5 and Proposi-
tion 2.7.7 the composed map
ψ
I◁ Ð→ ∆op
+ Ð→ E
is a limit diagram. Since ψ([0]) ≃ ∗ is final this exactly means that the map
ψ([2]) Ð→ ψ({0, 1}) × ψ({0, 2}) ≃ ψ({0, 1}) ×ψ({0}) ψ({0, 2} is an equivalence, as
desired.
where the left diagonal map is given by restriction along ∆op ⊆ ∆op+ (and is well-
defined in light of Proposition 5.3.10), and the right diagonal map is given by
restriction along ∆op
[−1,0]
⊆ ∆op
+ . By Remark 2.7.2 the right diagonal map is a trivial
Kan fibration. Choosing a section for this trivial Kan fibration we hence obtain a
functor
̃ E ∶ E∗ Ð→ MonN(∆op ) (E),
Ω
well-defined up to a contractible space of choices. Let us denote by
̃E
Ω
ΩE ∶ E∗ Ð→ MonN(∆op ) (E) Ð→ E
the composed functor, where the second functor is given by evaluation at [1] ∈ ∆op .
84 YONATAN HARPAZ
τ
∗ / Forget
τ
τx
∗(x) / x
τx
∆op
[−1,0]
, so that we can identify I with the category
●
● / ●.
The adjunction between the cone and slice constructions yields a natural map
σ ∶ I◁ Ð→ ∆op
+ which sends the cone point to [1]. We may then consider the map
ρ ∶ E∗ × I◁ Ð→ E
given by the formula ρ(x, i) = sx (σ(i)). The adjoint map I◁ Ð→ Fun(E∗ , E) can
then be depicted as a square of the form (5.26) which has the desired properties
by Theorem 2.7.5, since sx is a right Kan extension of (sx )∣N(∆op
[−1,0]
) and sx ([0]) is
final in E.
φ(y) / ψ(y)
is cartesian in D.
Proof. This is a classical fact. One way to prove this is by representing p and q
by actual colimit diagrams of simplicial sets and τ by a natural transformation
which is levelwise a minimal Kan fibration. In this case all the squares (5.29) will
necessarily be cartesian squares in simplicial sets, at which point it will be enough
to verify the 1-categorical analogue of the result for the category of simplicial sets.
There we can reduce to checking it for sets, which is clear.
Proof of Proposition 5.3.14. By Remark 2.7.2 it will suffice to show that the follo-
wing conditions are equivalent for a given functor ψ ∶ N(∆op
+ ) Ð→ S:
ψ([0]) / ψ([−1])
ψ([m]) / ψ([n])
is cartesian. We now note that the collection of σ ∶ [n] Ð→ [m] in ∆ for which (5.31)
is cartesian is closed under composition, and if it contains σ and σ ′ ○ σ then it
contains σ ′ . It will hence suffice to show that (5.31) is cartesian for maps of the
form σ ∶ {i} ↪ [n]. In this case, the square (5.31) becomes
ψ({i}) / ψ({0, ..., n})
We note that the monoid condition implies that ψ([m]) ≃ ∏m−1 j=0 ψ({j, j + 1}) and
hence that ψ([n + 1]) ≃ ψ({0, ..., i}) × ψ({i, ..., n + 1}) and ψ([n]) ≃ ψ({0, ..., i}) ×
ψ({i, ..., n}). To show that (5.32) is cartesian it will hence suffice to show that
ψ({i}) / ψ({i, ..., n})
is cartesian. This now follows from Remark 5.3.6, according to which, when ψ is
group-like, ψ([m]) decomposes as ∏j ψ({0, j}) for every [m]. We now note that
ρ∗ ψ is a colimit diagram since it is a split simplicial object. By Proposition 5.3.16
we may thus conclude that ρ∗ ϕ ⇒ ϕ is a cartesian natural transformation. Taking
the edge [0] Ð→ [−1] in ∆op + we recover that (5.30) is cartesian, as desired.
Let us now assume that (i) holds. Then ψ∣∆op is a group-like monoid object by
Proposition 5.3.10. Since S admits colimits it follows from Theorem 2.7.5 that there
exists a functor ψ ′ ∶ ∆op + Ð→ S and a natural transformation δ ∶ ψ∣∆op ⇒ ψ ∣∆op
′
′
exhibits ψ as a left Kan extension of ψ∣∆op along ∆ ⊆ ∆+ . This means, in
op op
right.
Proof. Let us denote by
̃ (∆ )
op n
(5.35) Ω 0 ∶ Mon(∆op )n (S∗ ) Ð→ Mon(∆op )n (Mon∆op (S))
the functor induced by (5.28) upon taking (∆op )n -monoid objects on both sides.
Since S0∗ ⊆ S∗ is a full subcategory closed under finite products Proposition 5.3.14
̃ (∆ ) restricts to an equivalence between Mon(∆op )n (S0 )
op n
implies that the functor Ω 0 ∗
and Mon(∆op )n (Mon∆op (S∗ )). It will hence suffice to show that under the identifi-
gr
cations
Mon(∆op )n (S∗ ) ≃ Mon(∆op )n (S)∗
and
Mon(∆op )n (Mon∆op (S)) ≃ Mon(∆op )n+1 (S) ≃ Mon∆op (Mon(∆op )n (S)),
̃ (∆ ) is homotopic to Ω̃ n , and the full subcategory Mon(∆op )n (Mongrop (S))
op n
the functor Ω 0 ∆
corresponds to the full subcategory Mongr ∆op (Mon(∆op )n (S)). Now the second claim
follows from the fact that for every ∞-category E with finite products the forget-
ful functor Mon(∆op )n (E) Ð→ E preserves products and detects equivalences, and
so in both cases the group-like condition is detected by the forgetful functor to
Mongr∆op (S). To prove the first claim let
Xn ∶= XMon(∆op )n (S) ⊆ Fun(N(∆op
+ ), Mon(∆op )n (S))
the full subcategory Xn on the left hand side corresponds to the full subcate-
gory Mon(∆op )n (X0 ) on the right hand side. Unwinding the definitions, this is
a direct consequence of Theorem 2.7.5 and the fact that the forgetful functor
Mon(∆op )n (S) Ð→ S detects equivalences and preserves limits (because the inclu-
sion Mon(∆op )n (S) ⊆ Fun((∆op )n , S) preserves limits and the evaluation ev[1] ∶
Fun((∆op )n , S) Ð→ S preserves limits).
We will say that an ∞-category E is pointed if it contains an object ∗ ∈ E which
is both initial and final. In this case we will refer to such an object as a zero object.
Now suppose that E is an ∞-category with a final object. Then the forgetful functor
(5.36) E∗ Ð→ E
is a left fibration whose fiber over x ∈ E is equivalent to the full subgroupoid of E/x
spanned by those y Ð→ x such that y is final. This ∞-groupoid is contractible if
88 YONATAN HARPAZ
and only if E is pointed. In this case, (5.36) is even a trivial Kan fibration by (the
dual of) Proposition 2.6.8.
Construction 5.3.18. Let E be a pointed ∞-category. Then (5.36) is a trivial
Kan fibration and so we may fix a section
ρ ∶ E Ð→ E∗ .
The choice of ρ is unique up to a contractible space of choices.
Example 5.3.19. By Remark 5.2.13 it follows that Mon(∆op )n (E) ≃ MonEn (E) is a
pointed ∞-category for every ∞-category with limits E. A zero object is given by
the the terminal object of E equipped with its unique En -monoid structure.
Now for every n ≥ 1 let us denote by
(5.37)
̃n
̃ ∗ ∶ Mon(∆op )n (S)
Ω
ρ
/ Mon(∆op )n (S)∗ Ω
/ Mon∆op (Mon(∆op )n (S)) ≅ Mon(∆op )n+1 (S)
n
Let Sn∗ ⊆ S∗ be the full subcategory spanned by those pointed spaces which are
n-connected (i.e., have no homotopy groups in degree ≤ n). We may now deduce
the following version of May’s recognition principle:
Corollary 5.3.20. For every n ≥ 1 the iterated loop functor (5.38) restricts to
an equivalence between the full subcategory Sn∗ ⊆ S∗ of pointed n-connected spaces
and the full subcategory Mongr
(∆op )n
(S) ⊆ Mon(∆op )n (S) of group-like (∆op )n -monoid
objects.
Proof. For k = 1, ..., n let us denote by Monn−k
(∆op )k (S) ⊆ Mon(∆op )k (S) the full
subcategory spanned by those (∆ ) -monoids whose underlying space is (n − k)-
op k
connected. Combining Proposition 5.3.14 and Lemma 5.3.13 we may conclude that
the functor (5.28) restricts to an equivalence
≃
̃ 0 ∶ Sn Ð→
Ω ∗ ∆op (S),
Monn−1
where we use the fact that any (n − 1)-connected (and so in particular connected)
∆op -monoid in spaces is automatically group-like by Example 5.3.4. Similarly,
by Proposition 5.3.14 for every k = 1, ..., n − 1 the functor (5.37) restricts to an
equivalence
≃
̃ ∗ ∶ Monn−kop k Ð→
Ω k (∆ ) (∆op )k+1 (S).
Monn−k−1
The desired result now follows by composing these equivalences.
6. Factorization homology
6.1. Manifolds and framings.
Definition 6.1.1. Let n ≥ 1 be an integer. By an n-manifold we will mean a
paracompact Hausdorff space M such that each x ∈ M either has a neighborhood
homeomorphic to ◻n or has a neighborhood homeomorphic to ◻n∂ . We will say that
M is open if every point has a neighborhood homeomorphic to ◻n .
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 89
Lemma 6.1.8. Let M be an n-manifold and I be a finite set. Then the natural
map
≃
Sing Emb(◻n × I, M ) Ð→ Germ(I, M ).
is a homotopy equivalence of Kan complexes.
Proof. It will suffice to show that for every k ≥ 0 the map Sing Emb(◻n (1/2k ) ×
I, M ) Ð→ Sing Emb(◻n (1/2k+1 ) × I, M ) is a homotopy equivalence. Indeed, this
follows from the fact that for every ε < ε′ the inclusion i ∶ ◻n (ε) ⊆ ◻n (ε′ ) is an
isotopy equivalence: it admits an inverse p ∶ ◻n (ε′ ) Ð→ ◻n (ε) such that p ○ i is in
the path component of the identity in Emb(◻n (ε), ◻n (ε)) and i ○ p is in the path
component of the identity in Emb(◻n (ε′ ), ◻n (ε′ )).
Remark 6.1.9. Using a suitable variant of the isotopy extension theorem it can be
shown that for every finite set I the evaluation at 0 map
Emb(◻n × I, M ) Ð→ Conf(I, M )
is a Serre fibration, and so the map
Sing Emb(◻n × I, M ) Ð→ Sing Conf(I, M )
is a Kan fibration. Since every n-simplex and every horn have finitely many non-
degenerate simplices it follows that the map
Germ(I, M ) Ð→ Sing Conf(I, M )
is a Kan fibration as well.
Proof of Proposition 6.1.6. Consider the simplicial functor Sing ∶ Mfld∆
n Ð→ Kan
which sends M to Sing(M ) and acts on mapping spaces via the natural map
Sing Emb(M, N ) Ð→ Map(Sing M, Sing N ). Let Sing∞ ∶ Mfldn Ð→ S be the in-
duced functor on coherent nerves. Then Sing∞ induces a map of Kan complexes
M ∶= (Mfldn )/M ×Mfldn BTop(n) Ð→ S/ Sing M ×S B
where B ⊆ S is the full subcategory spanned by the contractible Kan complex
Sing(◻n ).
Let j ∶ ◻n Ð→ M be a map in Mfldn , which we can identify with an object
of (Mfldn )/M ×Mfldn BTop(n). Then the induced map j∗ ∶ MapMfldn (◻n , ◻n ) Ð→
MapMfldn (◻n , M ) can be identified with the inclusion of the homotopy fiber of the
map
MapMfldn (◻n , M ) = (Mfldn )/M ×Mfldn {◻n } Ð→ (Mfldn )/M ×Mfldn BTop(n)
over j. The same statement holds for S with respect the base point determined by
Sing(j) ∶ Sing ◻n Ð→ Sing M . We then obtain a commutative diagram
(6.1)
MapMfldn (◻n , ◻n )
j∗
/ MapMfld (◻n , M ) / (Mfldn )/M ×Mfld BTop(n)
n n
≃
MapS (Sing ◻n , Sing ◻n ) / MapS (Sing ◻n , Sing M ) / S/ Sing M ×S B
≃ ≃
Sing ◻n / Sing M
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 91
in which the first two rows are fiber sequences of spaces and the bottom verti-
cal maps are given by evaluating at 0 ∈ Sing ◻n , and are weak equivalences since
Sing ◻n ≃ ∗ in S. The middle right horizontal map is then an equivalence since
its fibers MapS (Sing ◻n , Sing ◻n ) are contractible. I will now suffice to show that
the right most vertical map is an equivalence. For this, it will suffice to show that
the left square is homotopy cartesian for every j ∶ ◻n ↪ M , or alternatively, that
the left external rectangle is homotopy cartesian. By Lemma 6.1.8 we may instead
prove that the equivalent square
(6.2) Germ(∗, ◻n ) / Germ(∗, M )
Sing ◻n / Sing M
is homotopy cartesian, where the horizontal maps are induced by the fixed embed-
ding j ∶ ◻n Ð→ M . But this is because the right vertical map in (6.2) is a Kan
fibration (see Remark 6.1.9) and the square itself is strictly cartesian.
Definition 6.1.10. By a tangent structure we will mean a Kan complex B equipped
with a Kan fibration π ∶ B Ð→ BTop(n).
Definition 6.1.11. Let B Ð→ BTop(n) be a tangent structure and let M be an
open n-manifold. A B-framing of M is a lift of the form
v:
B
v
v
v
v
M
π / BTop(n)
spanned by the objects ◻n , ◻n∂ . Given an n-manifold M we define the right fibration
π ∶ M ∂ ∶= (Mfldn )/M ×Mfldn BTop∂ (n) Ð→ BTop∂ (n).
To understand this fibration it will be useful to extend the construction of germs to
cubes with boundary. For 0 < ε < 1 let us denote by ◻n∂ (ε) ∶= [0, ε) × (−ε, ε)n−1 ⊆ ◻n∂ .
Definition 6.1.16. Let M be an n-manifold, I a finite set and X ∶ I Ð→ {◻n , ◻n∂ }
an I-tuple of objects of BTop∂ (n). We define
Germ(X, M ) ∶= colim Sing Emb ( ∐ Xi (1/2k ), M ),
k i∈I
and refer to it as the simplicial set of boundary X-germs in M . We will also denote
n n
by Germ∂ (I, M ) ∶= Germ(◻∂ , M ) where ◻∂ ∶ I Ð→ {◻n , ◻n∂ } denotes the constant
tuple with value ◻∂ .
n
Germ∂ (I1 , M ) / Germ(I1 , ∂M ) / Conf(I1 , ∂M )
where the vertical maps are induced by ι. The right square is cartesian on the level
of simplicial sets and is also homotopy cartesian since the right bottom horizontal
map is a Kan fibration (see Remark 6.1.9). For the same reason the external
rectrangle is homotopy cartesian (see Remark 6.1.18). We may hence conclude
that the left square is homotopy cartesian. In particular, the homotopy fibers of
the top left horizontal map are the same as those of the bottom left horizontal map.
Since every vertex of Germ(I1 , ∂M ) is in the image of the vertical map for at least
one ι ∶ ◻n∂ × I1 Ð→ M it will now suffice to show that the top left horizontal map is
a homotopy equivalence of Kan complexes. Since this map breaks as a product of
maps which just need to show that
Germ∂ (∗, ◻n∂ ) Ð→ Germ(∗, ◻n−1 ).
is a homotopy equivalence of Kan complexes. For this, it will suffice to show that
for every ε > 0 the map of topological spaces
(6.7) Emb(◻n∂ (ε), ◻n∂ ) Ð→ Emb(◻n−1 (ε), ◻n−1 ).
We claim that in this case the map
(6.8) [0, ε) × (−) ∶ Emb(◻n−1 , ◻n−1 ) Ð→ Emb([0, ε) × ◻n−1 , [0, ε) × ◻n−1 )
given by taking the product with [0, ε) is a homotopy inverse to (6.3). Indeed,
the composition of (6.3) after (6.8) is the identity, and the composition of (6.8)
after (6.3) is homotopic to the identity via the homotopy f ↦ ft where
(s, f (x)) s≤t
ft (s, x)i ∶= { −1 0 ,
gt f (gt (s, x)) s≥t
Here we use the coordinates s ∈ [0, ε) and x ∈ ◻n−1 , f0 ∶ ◻n−1 Ð→ ◻n−1 the restriction
of f to s = 0 and gt ∶ [t, ε) × ◻n−1 Ð→ [0, ε) × ◻n−1 is the homeomorphism given by
gt (r, x) = ( r−t
ε−t
, x).
Let
(6.9) q ∶ BTop∂ (n) Ð→ ∆1
be the unique map which sends ◻n to 0 and ◻n∂ to 1.
Lemma 6.1.20. The map (6.9) is a right fibration classified by the functor [1] Ð→
Cat∞ depicted by the arrow BTop(n−1) Ð→ BTop(n) induced by taking the product
with (−1, 1).
Proof. Let BTop∆+ (n) ⊆ BTop∂ (n) be the full subcategory spanned by ◻∂ and let
∆ n
BTop+ (n) ∶= N(BTop+ (n)) be its coherent nerve. Consider the functor
∆
(6.10) ι ∶ BTop∆
+ (n) Ð→ BTop (n)
∆
94 YONATAN HARPAZ
which sends ◻n∂ to ◻n and any open embedding f ∶ ◻n∂ Ð→ ◻n∂ to the open em-
≅
bedding ◻n Ð→ (0, 1) × ◻n−1 Ð→ (0, 1) × ◻n−1 where the first arrow is the unique
rectilinear homeomorphism of this form and the second is obtained by restricting
f . We then observe that BTop∂ (n) is naturally equivalent to the Grothendieck
construction ∫[1] [ι] where [ι] ∶ [1] Ð→ Cat∆ is the functor corresponding to the
arrow of simplicial categories ι ∶ BTop∆
+ (n) Ð→ BTop (n). It then follows by Pro-
∆
position 2.5.9 that BTop∂ (n) ×∆1 ∆{0} ≅ BTop(n),BTop∂ (n) ×∆1 ∆{1} ≅ BTop+ (n)
and q ∶ BTop∂ (n) Ð→ ∆1 is a right fibration classified by the arrow
(6.11) N(ι) ∶ BTop+ (n) Ð→ BTop(n).
It will hence suffice to show that ι is weakly equivalent in the arrow category of
Cat∆ to the functor BTop∆ (n − 1) Ð→ BTop∆ (n) induced by taking the product
with (−1, 1). To see this, observe that the latter functor factors as
η ι
BTop∆ (n − 1) Ð→ BTop∆
+ (n) Ð→ BTop (n)
∆
where η is induced by taking the product with [0, 1). It will hence suffice to show
that η is a Dwyer-Kan equivalence, that is that the map
(6.12) [0, 1) × (−) ∶ Emb(◻n−1 , ◻n−1 ) Ð→ Emb([0, 1) × ◻n−1 , [0, 1) × ◻n−1 )
is a weak homotopy equivalence. But this now follows from Lemma 6.1.19 by the
2-out-of-3 property.
By Lemma 6.1.20 the composed map M ∂ Ð→ BTop∂ (n) Ð→ ∆1 is also a right
fibration. We note that the fiber M ∂ ×∆1 ∆{0} is naturally isomorphic to the
simplicial set M . Let us denote by
∂M ∶= M ∂ ×∆1 ∆{1}
the fiber over ∆{1} ⊆ ∆1 . The map (6.11) then naturally extends to a commutative
square of Kan complexes:
(6.13) ∂M /M
BTop+ (n) / BTop(n)
BTop+ (n) / BTop(n)
Remark 6.1.24. If ∂B = ∅ then any B∂ -framed n-manifold is open, and the notion
of a B∂ -framing coincides with a B-framing. In particular, we may consider the
condition of being an open n-manifold as a special case of framing.
Example 6.1.25. If N is an n-manifold then we can take B∂ = N ∂ equipped with
its natural map to BTop∂ (n). Similarly to Example 6.1.15, if U is either ◻n or ◻n∂ ,
then the data of an N ∂ -framing on U is essentially equivalent to the data of an
open embedding U Ð→ N .
We close this section with the following lemma, which we will invoke several
times in the subsequent sections. Upon first contemplating the following lemma,
the reader is encouraged to take M = ∗.
Lemma 6.1.26. Let p ∶ N Ð→ M be a continuous map from an n-manifold N to
an m-manifold M . Let I be a finite set, X ∶ I Ð→ {◻n , ◻n∂ } an I-tuple of objects of
Top∂ (n). Let Disj(M ) ⊆ O(M ) be the full subposet spanned by those open subsets
V ⊆ M which are homeomorphic to disjoint unions of copies of ◻n and ◻n∂ and let
P ⊆ O(N )I × Disj(M ) be the full subposet spanned by those pairs ({Ui }i∈I , V ) such
that Ui ≅ Xi , the Ui ’s are pairwise disjoint, and p(Ui ) ⊆ V . Then the canonical
maps
(6.15)
hocolim ∏ Emb(Xi , Ui ) Ð→ hocolim Emb(∐ Xi , p−1 (V )) Ð→ Sing Emb ( ∐ Xi , N )
({Ui },V )∈P i∈I V ∈Disj(M ) i∈I i
hocolim ∏ Sing Conf({i}, Ui ) × ∏ Sing Conf({i}, ∂Ui ) / Sing Conf(I0 , N ) × Sing Conf(I1 , ∂N )
({Ui },V )∈P i∈I i∈I1
0
such that
MulE∆ ({◻n }i∈I , ◻n ) = Sing Emb(◻n × I, ◻n ).
Top(n)
Remark 6.2.2. We point out that the difference between Definition 6.2.1 and Defi-
nition 5.1.2 is that in the former the embeddings are not assumed to be rectilinear.
Since the singular complex of a space is always a Kan complex the simplicial
operad E∆
Top(n) is locally Kan.
where BTop(n)∐ is defined as in Exercise 3.3.2. Taking coherent nerves and using
Exercise 3.3.2 we obtain a map of ∞-operads
(6.18) E⊗ ∐
Top(n) Ð→ BTop(n) .
(E∆ ⊗ / BTop∆ (n)∐
Top(n) )
(6.20) E⊗ / Rect(n)∐
n
E⊗ / BTop(n)∐
Top(n)
of ∞-operads (see Example 4.1.12), in which the top right corner is equivalent to
the terminal ∞-operad N(Com⊗ ) (since Rect(n) is a contractible Kan complex).
Proof. We first note that all four ∞-operads appearing in (6.20) have underlying
∞-categories which are Kan complexes, and hence have the property that a map
in any of them is inert if and only if its image in N(Fin∗ ) is inert. It follows that if
O⊗ is an ∞-operad then any map of ∞-categories from O⊗ to any of the ∞-operads
in (6.20) will automatically preserve inert maps, and hence will automatically be a
map of ∞-operads. It will hence suffice to show that (6.20) is a homotopy pullback
square of ∞-categories. Since the coherent nerve is a right Quillen functor we may
instead show that (6.19) is a homotopy pullback square of simplicial categories. We
note that all four simplcial categories in (6.19) have the same set of objects as Fin∗ .
Comparing Construction 4.1.9 and the construction in Exercise 3.3.2 it will now
98 YONATAN HARPAZ
suffice to show that for every ⟨m⟩ ∈ Fin∗ the square of Kan complexes
○ / ∏ Sing Rect(◻n × {i}, ◻n )
(6.21) Sing Rect(◻n × ⟨m⟩ , ◻n )
i∈⟨m⟩○
○ / ∏ Sing Emb(◻n × {i}, ◻n )
Sing Emb(◻n × ⟨m⟩ , ◻n )
i∈⟨m⟩○
○ / ∏ Sing Emb(◻n × {i}, ◻n )
Sing Emb(◻n × ⟨m⟩ , ◻n )
i∈⟨m⟩○
≃
≃
○ / ∏ Germ({i}, ◻n )
Germ(⟨m⟩ , ◻n )
i
○ / ∏ Sing Conf({i}, ◻n )
Sing Conf(⟨m⟩ , ◻n )
i
where the vertical maps in the middle square are equivalences by Lemma 6.1.8. In
particular, the middle square is homotopy cartesian. Similarly, the vertical maps
in the external rectangle are equivalences by Lemma 5.1.10, and so the external
rectangle is homotopy cartesian. By the pasting lemma for homotopy cartesian
square, to show that the top square in (6.22) is homotopy cartesian it will suffice
to show that the bottom square in (6.22) is homotopy cartesian. But this is indeed
true since the right vertical map in this right most square is a Kan fibration (see
Remark 6.1.9) and the right most square itself is cartesian on the level of simplicial
sets.
Definition 6.2.6. Let B Ð→ BTop(n) be a tangent structure (Definition 6.1.10).
We define the ∞-operad E⊗
B by the pullback square
E⊗ / B∐
B
E⊗ / BTop(n)∐
Top(n)
We will refer to E⊗
B as the B-framed little cube ∞-operad.
E⊗ / B∐
B
We will consider ∫B E⊗
n as a weak ∞-operad via the pullback structure of Re-
mark 4.2.17.
Warning 6.2.9. The maps E⊗ B Ð→ B
∐ and N(Fin ) × B Ð→ B ∐ are generally not
∗
categorical fibrations. To obtain an explicit model for ∫B E⊗
n one needs to first
replace these maps by categorical fibrations and then take the actual product of
simplicial sets.
⊗
Remark 6.2.10. When B ≃ ∗ we have ∫B E⊗ ⊗
n ≃ EB ≃ En by Proposition 6.2.5.
has weakly contractible homotopy fibers. But this is true because the map (6.25) is
in fact an equivalence of ∞-categories: indeed, we can consider (6.25) as a map over
′
N(Fin∗ )act
/⟨m⟩ , and for a given active α ∶ ⟨m ⟩ Ð→ ⟨m⟩, the map on homotopy fibers
over α induced by (6.25) is given by the diagonal map ∗ ≃ B/b Ð→ ∏i∈m′ B/b ≃ ∗.
100 YONATAN HARPAZ
To show that (6.23) is in fact a strong approximation we simply note that both for
N(Fin∗ ) × B and B ∐ the basics are those objects which lie above ⟨1⟩ ∈ N(Fin∗ ), a
map is inert if and only if its image in N(Fin∗ ) is inert, and the subcategories of
basics and inert maps identifies on both sides with the Kan complex B.
is an equivalence of ∞-categories.
We now unwind the definitions to see what the notion of an ∫B E⊗
n -monoid in E
actually is. Consider the composed map
(6.26) π ∶ ∫ E⊗
n Ð→ N(Fin∗ ) × B Ð→ B.
B
Without loss of generality we may assume that we have chosen ∫B E⊗ n such that
π is a categorical fibration. Since B is a Kan complex the map (6.26) is then
automatically a cocartesian fibration. Let us denote by E⊗
b the fiber of π over
b ∈ B. We then have a commutative diagram
E⊗ / {b} × N(Fin∗ )
b
∫B En
⊗ / B × N(Fin∗ )
E⊗ / B∐
B
E⊗ / BTop(n)∐
Top(n)
in which all squares are homotopy cartesian, and so the external rectangle is ho-
motopy cartesian. This yields an identification E⊗ ⊗
b ≃ En by Proposition 6.2.5. The
cocartesian fibration π then corresponds to a functor χ ∶ B Ð→ Cat∞ , all of whose
fibers are equivalent to E⊗n . Furthermore, since π factors through the projection
N(Fin∗ ) × B Ð→ B we see that the transition functors α! ∶ E⊗ ⊗
b Ð→ Eb′ respect the
projection to N(Fin∗ ). Since they are all equivalences they must also preserve inert
maps. In particular, all the transition functors are maps of ∞-operads E⊗ ⊗
b Ð→ Eb′ .
We may then consider χ as a B-indexed diagram of ∞-operads, all of which are
equivalent to E⊗n.
To translate this into an a description of the notion of an ∫B E⊗ n -monoid object
in E we note first that since (6.26) is a cocartesian fibration the data of a functor
ψ ∶ ∫B E⊗
n Ð→ E can be identified with the data of a compatible B-indexed family
of functors E⊗b Ð→ E. Unwinding the definitions we see that such a ψ is a monoid
object if and only if each ψ∣E⊗ is an E⊗ b -monoid object. We may hence conclude
b
⊗
that ∫B En -monoid objects in E correspond to compatible families ψb ∈ MonE⊗ (E)
b
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 101
of E⊗ ⊗
b -monoid objects. By Proposition (6.2.11) this description is valid for EB as
well.
Remark 6.2.13. Using a suitable generalization of Proposition 4.2.18 one can show
that the same description holds also for E⊗
B -algebra objects in symmetric monoidal
∞-categories which are not necessarily cartesian. More generally, for every ∞-
operad O⊗ the collection of maps E⊗ ⊗
b Ð→ EB induces an equivalence of ∞-categories
We finish this section with a useful variant of the above construction for the
∞-operads E⊗n,∂ of little cubes with boundary (see Variant 5.1.5).
102 YONATAN HARPAZ
for Xi , Y ∈ {◻
n
, ◻n∂ }. We will denote by
E⊗ ∶= N(E∆
Top∂ (n) )
Top ∂ (n)
E⊗ / B∐
B∂ ∂
E⊗ / BTop∂ (n)∐
Top ∂ (n)
then E⊗ ⊗
B∂ ≃ En,∂ . This can be proven using a similar argument to the one used in
the proof of Proposition 6.2.5.
Remark 6.2.21. Arguing using suitable weak ∞-operadic models as above one can
extract a description of E⊗B∂ -monoids in an ∞-category E in the spirit of Re-
mark 6.2.15. More precisely, given a boundary tangent structure B∂ Ð→ Top∂ (n)
the map
χn ∶ B ∶= B∂ ×∆1 ∆{0} Ð→ BTop(n)
determines a family {E⊗ ⊗
b }b∈B of ∞-operads (all equivalent to En ), and the map
Definition 6.3.4. Let Dord M be the ordinary operad whose objects are pairs (U, ρ)
where U ∈ {◻n , ◻n∂ } and ρ ∶ U Ð→ M is an open embedding. Given objects
(X1 , ρ1 ), ..., (Xm , ρm ), (Y, η) ∈ Dord
M , a multimap {(Xi , ρi }} Ð→ (Y, η) is given by a
commutative diagram
(6.27) ∐i XiF /Y
FF ~~
FF ~~
F
(ρ1 ,...,ρm ) FF" ~~~ η
~~
M
where the horizontal arrow is an open embedding. We note that for any given
(X1 , ρ1 ), ..., (Xm , ρm ), (Y, η), there is at most one multimap {(Xi , ρi } Ð→ (Y, η)
(which is the case exactly when the images of ρ1 , ..., ρm are disjoint and contained
in the image of η). We will denote by D⊗ ord ⊗
M ∶= N((DM ) ) the operadic nerve of DM .
ord
∂ (n) ⊆ Mfldn
ord
of vertices. We will denote by BTopord the full subcategory spanned
by {◻ , ◻∂ }, and by BTop (n), BTop+ (n) ⊆ BTopord
n n ord ord
∂ (n) the full subcategories
spanned by ◻n and ◻n∂ respectively. Given an n-manifold M we will then denote
by
ord
M∂ ∶= (Mfldord )/M ×Mfldord
n ∂ (n).
BTopord
ord
N(M ∂ ) Ð→ M ∂ .
is a cartesian fibration. We may identify the fiber of (6.28) over ∆{0} with poset of
opens in M which are homeomorphic to ◻n and the fiber over ∆{1} with the poset
of opens in M which are homeomorphic to ◻n∂ .
For X1 , ..., Xn , Y ∈ {◻n , ◻n∂ } the map Embord (∐i Xi , Y ) Ð→ Sing Emb(∐i Xi , Y )
determines a map of simplicial operads Dord M Ð→ ETop∂ (n) (which, in particular,
∆
(6.30) D⊗ / E⊗
M Top ∂ (n)
∐
M∂ / BTop∂ (n)∐
ord ∐
where the left vertical map is given by the composition D⊗ ∐
M Ð→ (M ∂ ) Ð→ M ∂ .
The square (6.30) then induces a map of ∞-operads
∐
(6.31) D⊗ ⊗ ⊗
M Ð→ M ∂ ×BTop(n)∐ ETop(n) = EM
(6.33) M∂
ord ι / M∂
JJ uu
JJ u
JJ uu
JJ
JJ uuu
$ zuu
BTop∂ (n)
The right vertical map is right fibration and hence its homotopy fibers, which are
also its actual fibers, are ∞-groupoids. It will hence suffice to show that (6.33)
induces an equivalence on homotopy fibers over every object of BTop∂ (n). Let us
hence fix such an object V ∈ {◻n , ◻n∂ }. Then the (homotopy) fiber of the right
diagonal map in (6.33) can be identified with Sing Emb(V, M ), while the homotopy
ord
fiber XV Ð→ M ∂ of the left diagonal map over V can be modeled by the full
ord ord
subcategory of the comma ∞-category (M ∂ )V / ∶= BTop∂ (n)V / ×BTop∂ (n) M ∂
spanned by those (ρ ∶ U Ð→ M, η ∶ V Ð→ U ) such that η is an equivalence in
ord
BTop∂ (n). Let XV ⊆ M ∂ be the full subcategory spanned by those ρ ∶ U Ð→ M
such that U = V ∈ {◻n , ◻n∂ }. We may identify X V with the full subposet of O(M )
spanned by those open subsets which are homeomorphic to either ◻n or ◻n∂ . Then
the forgetful map XV Ð→ XV is a left fibration which corresponds to the functor
χ([ρ ∶ U Ð→ M ]) = Sing Emb(V, U ) (where we note that any embedding V Ð→ U is
an isotopy equivalence when U = V ). In particular, XV is a model for the homotopy
colimit of χ ∶ XV Ð→ S. We may then identify the map on homotopy fibers over V
induced by (6.33) with the natural map
Proposition 6.3.8. The map (6.31) is a weak approximation in the sense of De-
finition 4.2.14.
Proof. Condition (2) of Definition 4.2.14 follows from (6.3.7). We shall now prove
that Condition (1) holds. Arguing as in the proof of Proposition 5.2.4 it will suffice
to show that for every ρ ∶ V Ð→ M in (D⊗ ⊗
M )⟨1⟩ ⊆ DM , the map
(D⊗ ⊗ act
M )/ρ Ð→ (EM )/ρ
act
has weakly contractible homotopy fibers. We first note that the map (E⊗ M)
act
Ð→
⊗
(ETop (n) ) is a right fibration (since it is the pullback of a right fibration), and
act
∂
hence the map
(E⊗ act ⊗
M )/ρ Ð→ (ETop )act
/V
∂ (n)
is a trivial Kan fibration. On the other hand, a direct inspection shows that the
map (D⊗ ⊗ act
V )/ Id Ð→ (DM )/ρ induced by ρ is an isomorphism of simplicial sets. It will
act
(D⊗ act ⊗
V )/ Id Ð→ (ETop )act
/V
∂ (n)
106 YONATAN HARPAZ
Since the right diagonal map is a right fibration it will suffice to show that the
horizontal arrow in (6.35) induces a weak equivalence on homotopy fibers over any
object U ∈ (E⊗ Top∂ (n)
)act , which we can identify with a tuple (U1 , ..., Ul ) of objects
of Top∂ (n) by choosing cocartesian lifts to ρi ∶ ⟨n⟩ Ð→ ⟨1⟩. We now note that the
(homotopy) fiber of the right diagonal map over U = (U1 , ..., Ul ) is naturally equi-
valent to Sing Emb(∐i Ui , V ). To describe the homotopy fiber of the left diagonal
map, let PV ⊆ O(V ) be the subposet consisting of those open subsets W ⊆ V which
⟨l⟩○
are (abstractly) homeomorphic to either ◻n or ◻n∂ . Let XU ⊆ PV be the subpo-
set spanned by those tuples (W1 , ..., Wl ) ∈ (PV )⟨l⟩ such that Wi = Ui ∈ {◻n , ◻n∂ }
○
and the Wi ’s are pairwise disjoint. Unwinding the definitions as in the proof of
Proposition 5.2.4 we see that the map induced by (6.35) on homotopy fibers over
(U1 , ..., Ul ) can be identified with the canonical map
(6.36) hocolim ∏ Sing Emb(Ui , Wi ) Ð→ Sing Emb(∐ Ui , V )
(W1 ,...,Wl )∈XU i∈⟨l⟩○ i
is cofinal. Indeed, taking G to be the trivial group and p to be the identity map
M Ð→ M we get that the first map in (6.38) is cofinal, and taking G = Z/2 and p
to be the split 2-covering M ∐ M Ð→ M we get that the composed map in (6.38)
is cofinal (where we note that E(M ∐ M ) can be naturally identified with E(M ) ×
E(M )). Let us now prove that the composed map in (6.39) is indeed cofinal.
We fix a finite set I and an object η ∶ V ∶= ∐i∈I Vi Ð→ M ̃ of E(M ̃). Consider
the comma ∞-category Disj(M )η/ ∶= E(M ̃)η/ × ̃ Disj(M ). We need to show
E(M )
that Disj(M )η/ is weakly contractible. We remark that the category Disj(M ) is in
fact a poset which can be identified with the poset of open subsets of M which are
homeomorphic to a finite disjoint union of ◻n and ◻n∂ . To simplify notation, let us
hence denote a general object (J, {Uj }j∈J , ρ ∶ ∐j Uj Ð→ M ) of Disj(M ) simply as
U ⊆ M , where U ∶= ∐j Uj is understood as an open subset of M of the above form.
The projection Disj(M )η/ Ð→ Disj(M ) is then a left fibration corresponding to the
functor
̃) ≃ hofibη [Sing Emb(V, U ×M M
U ↦ Map(Mfldn )/N (V, U ×M M ̃) Ð→ Sing Emb(V, M
̃)].
Given two n-manifolds M, N , let us denote by Embloc (M, N ) the space of those
maps M Ð→ N which restrict to an open embedding on each connected compo-
nent of M (endowed with the compact open topology). We define Mfldloc n be the
coherent nerve of the simplicial category whose objects are n-manifolds and whose
mapping spaces are given by Sing Embloc (M, N ). We then see that the active part
(BTop(n)∐ )act of BTop(n)∐ can be identified with the full subcategory of Mfldloc
n
spanned by the finite disjoint unions of n-cubes. In addition, given an n-manifold
M we have a natural isomorphism of simplicial sets
∐
n )/M ×Mfldloc
(M )act ≅ (Mfldloc n
(BTop(n)∐ )act .
∐ act
(M ) / (BTop(n)∐ )act
(E⊗ × ∆{0} / ⊗
A
M) o7 C
act
(6.40)
o o
o oo
o
(E⊗ )act
× ∆ 1 / N(Finact )
M ∗
Remark 6.4.6. In light of Lemma 6.4.4 restriction along Disj(M ) Ð→ E(M ) induces
an equivalence
∫ A≃ colim A(U ).
M U ∈Disj(M )
(6.41) ∫ A ∶ Mfldn Ð→ C
(−)
along (E⊗
Top(n)
)act Ð→ Mfldn which is given on objects by the factorization homo-
logy M ↦ ∫M A (see Remark 6.4.7).
Remark 6.4.9. In the situation of Remark 2.7.5, it follows from Lurie’s theory of
operadic left Kan extensions (using our base hypothesis 6.4.3 and Lemma 6.4.4)
that the left Kan extension (6.41) refines to a map of ∞-operads
(6.42) ∫ A ∶ (Mfldn )⊗ Ð→ C⊗
(−)
110 YONATAN HARPAZ
In this particular case one can even show that (6.42) is a symmetric monoidal func-
tor, that is, it preserves all cocartesian edges over N(Fin∗ ). This eventually follows
from the fact that if M = M0 ∐ M1 then the rule that sends (X0 Ð→ M0 , X1 Ð→ M1 )
to X0 ∐ X1 Ð→ M determines an equivalence
≃
(E⊗ ⊗ ⊗
Top(n) )/M0 × (ETop(n) )/M1 Ð→ (ETop(n) )/M .
act act act
Example 6.4.11 ([1, Corllary 3.12]). Let I be the unit interval. Recall (see Ex-
ample 6.3.3) that an E⊗ I -disk algebra in C is equivalent to the data of a triple
(A, M0 , M1 ) where A is an associative algebra object in C, M0 is a right A-module
in C1C / and M1 is a left A-module in C1C / . In this case we have a natural equivalence
∫ (A, M0 , M1 ) ≃ M0 ⊗A M1
M
Our next goal is to discuss the Fubini property of factorization homology. For
this, we will need to introduce the notion of a bundle map of manifolds with boun-
dary.
Suppose first that N is an open k-manifold and E is an n-manifold, possibly
with boundary. Then we have the notion of a manifold bundle map from E to N ,
which is, by definition a map p ∶ E Ð→ N such that for every open embedding
ρ ∶ ◻k Ð→ N the pullback ρ∗ E ∶= E ×N ◻k Ð→ ◻k admits a trivialization of the form
τ ∶ ρ∗ E ≅ P × ◻k with P an (n − k)-manifold (here by trivialization we simply mean
that τ commutes with the respective projections to ◻k ). In this case the association
[ρ ∶ ◻k Ð→ N ] ↦ E ×N ◻k determines a DN -algebra object in Mfld∂n , which we shall
call [p−1 ]. The local triviality of E now implies that all the 1-ary operations act
on [p−1 ] by equivalences, and by Remark 6.3.11 we have that [p−1 ] descends to an
essentially unique N -disk algebra object in Mfld∂n .
We would like to have a similar story when N is a k-manifold which is not
necessarily open (i.e., can have a boundary).
Definition 6.4.12. Let N be an k-manifold. By a manifold ∂-bundle over N we
shall mean a map E Ð→ N with E an n-manifold and such that the following
conditions hold:
(1) For every open embedding ρ ∶ ◻k Ð→ N the pullback ρ∗ E Ð→ ◻k admits a
trivialization of the form ρ∗ E ≅ P × ◻k with P a (n − k)-manifold.
(2) For every open embedding ρ ∶ ◻k∂ Ð→ N the pullback ρ∗ E Ð→ ◻k∂ admits an
identification of the form
≅ / P × ◻k−1
ρ∗ EB
BB w
BB ww
BB www
B ww (f,Id)
{ww
◻∂k
Proposition 6.4.17 (The Fubini property, [1, Proposition 3.23]). In the above
setting the natural map
(6.43) ∫ p∗ A Ð→ ∫ A
N M
is an equivalence.
Proof. By Remark 6.4.6 the left hand side of (6.43) is given by an iterated colimit
∫ p∗ A ≃ U ∈Disj(N
colim p∗ A(U ) ≃ colim
)
colim
U ∈Disj(M ) V ∈Disj(p−1 (U ))
A(V )
N
which can be assembled to a single colimit indexed by the full subposet Disj(p) ⊆
Disj(M ) × Disj(N ) spanned by those pairs (V, U ) such that p(V ) ⊆ U , since the
forgetful functor Disj(p) Ð→ Disj(N ) is the cocartesian fibration classifying the
functor U ↦ Disj(p−1 (U )) (this follows from a general fubini property of colimits
along cocartesian fibrations). To show that (6.43) is an equivalence it will hence
suffice to show that the composed functor
Disj(p) Ð→ Disj(M ) Ð→ E(M ),
is cofinal. Let us hence fix a finite set I and an object ι ∶ W ∶= ∐i Wi Ð→ M of E(M ).
Then the comma ∞-category Disj(p)ι/ sits in a left fibration Disj(p)ρ/ Ð→ Disj(p)
which classifies the functor
(U, V ) ↦ hofibι [Sing Emb(W, V ) Ð→ Sing Emb(W, M )] .
Since the base change functor hofibι ∶ S/ Sing Emb(W,M ) Ð→ S preserves homotopy
colimits we may instead show that the natural map
(6.44) hocolim Sing Emb(W, V ) Ð→ Sing Emb(W, M )
(U,V )∈Disj(p)
We now observe that the first map in (6.46) is a weak homotopy equivalence by
Lemma 6.1.26 applied to the identity map p−1 (V ) Ð→ p−1 (V ) and the second map
is a weak homotopy equivalence by Lemma 6.1.26 applied to the identity map
N Ð→ N .
Example 6.4.18. The Fubini property can help us to decipher what is the factoriza-
tion homology along the circle. We first note that by Remark 6.2.13 the notion of
an ES 1 -algebra object in C is equivalent to that of a pair (A, τ ) where A is an associ-
ative algebra A and τ ∶ A Ð→ A is an automorphism (associated to the monodromy
along the circle). The projection p ∶ S 1 Ð→ [−1, 1] on the x-axis is a ∂-bundle map
and the E[−1,1] -algebra p∗ (A, τ ) can be identified with the triple (Aop ⊗ A, A−1 , A1 )
where A−1 is a copy of A considered as a pointed right Aop ⊗ A-module and A1
is a copy of A considered as a pointed left Aop ⊗ A-module via the equivalence
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 113
(Id ⊗τ ) ∶ Aop ⊗ A Ð→ Aop ⊗ A. By the Fubini property and Example 6.4.11 we then
have that
∫ (A, τ ) ≃ ∫ p∗ (A, τ ) ≃ A ⊗Aop ⊗A A
S1 [−1,1]
is the τ -twisted Hochschild homology of A.
6.5. Axiomatic characterization of factorization homology. In this section
we will focus attention on open n-manifolds, i.e., those which do not have boundary.
Following [1], our goal is to consider homology theories on suitably framed open n-
manifolds. For this we will need to isolate a particular full subcategory of Mfldn
spanned by manifolds which can be built in finitely many steps by gluing disks of
various dimensions. This gluing is defined via the notion of a collar gluing spelled
out in Example 6.4.16.
Let ∂◻k denote the boundary of the n-cube, which is homeomorphic to the
(k − 1)-sphere (where ∂◻0 = ∅ by convention). We have a canonical right ∂◻k -collar
(0, 1) × ∂◻k ↪ ◻k embedded as the complement of {0} ⊂ ◻k . Let M0 be an open
n-manifold. If ρ ∶ (0, 1) × ∂ ◻k ×◻n−k ↪ M0 is a left [∂ ◻k ×◻n−k ]-collar then we will
say that
M = ◻k × ◻n−k ∐ M0
(0,1)×∂◻k ×◻n−k
is obtained from M0 by adding an open handle of index k.
Definition 6.5.1. Let M be an open n-manifold. We will say that M is of finite
type if it can be obtained from ∅ by adding finitely many open handles. We will
denote by Mfldfinn ⊆ Mfldn the full subcategory spanned by open n-manifolds of
finite type. Similarly, if B Ð→ BTop(n) is a tangent structure then we will denote by
MfldB,fin
n n ×Mfldn Mfldn . We note that the inclusion Mfldn
∶= MfldB fin B,fin
⊆ MfldB
n is
fully faithful and its essential image is spanned by those B-framed open n-manifolds
which are of finite type.
Example 6.5.2. Adding to M0 an open handle of index 0 is simply taking the
coproduct M = M0 ∐ ◻n . In particular, the n-disk ◻n is an n-manifold of finite
type.
Example 6.5.3. ∂ ◻k+1 ×◻n−k is obtained from ◻k × ◻n−k ≅ ◻n by adding a single
open handle of index k. In particular, ∂ ◻k+1 ×◻n−k is an n-manifold of finite type.
Warning 6.5.4. The notion of an open handle is closely related, but not identical,
to the notion of a handle studied in classical geometric topology, which is usually
applied only to compact manifolds. However, if M is a compact manifold with a
finite handle decomposition in the classical sense, then the interior of M is of finite
type in the sense of Definition 6.5.1. In particular, it is known that any closed
manifold of dimension ≠ 4 has a finite handle decomposition, and is hence of finite
type. On the other hand, in dimension 4 a closed manifold admits a finite handle
decomposition if and only if it is smoothable.
Definition 6.5.5. We will say that a manifold ∂-bundle p ∶ M Ð→ N is open if
M is open, and we will say that p has finite type if for every U ⊆ N which is
homeomorphic to either ◻k or ◻k∂ the fiber product M ×N U has finite type.
Now let C be as in Hypothesis 6.4.3 and let B Ð→ BTop(n) be a tangent struc-
ture. In this section we will describe a certain class of symmetric monoidal functors
114 YONATAN HARPAZ
MfldB,fin
n Ð→ C which are called homology theories in [1]. The defining property of
these functors is that they satisfy ⊗-excision, a term we shall now define.
Definition 6.5.6. Let F ∶ MfldB,finn Ð→ C be a symmetric monoidal functor. We
will say that F satisfies ⊗-excision if for every open finite type ∂-bundle p ∶ M Ð→ I
the induced map
∫ p∗ F Ð→ F(M )
I
[p−1 ] F
is an equivalence, where p∗ F is the composed functor E⊗ B,fin ⊗
B Ð→ (Mfldn ) Ð→ C⊗
as above.
Remark 6.5.7. In light of Example 6.4.11 and Example 6.4.16 we may also (somew-
hat informally) phrase the ⊗-excision property as saying that for every collar gluing
M = M0 ∐(0,1)×P M1 of finite type open n-manifolds the induced map
F(M0 ) ⊗F((0,1)×P ) F(M1 ) Ð→ F(M )
is an equivalence.
Definition 6.5.8. Let B Ð→ BTop(n) be a tangent structure and C a presentably
symmetric monoidal ∞-category. A B-framed homology theory is a symmetric
monoidal functor F ∶ MfldB,fin
n Ð→ C which satisfies ⊗-excision. We will denote by
H(MfldB,fin
n , C) ⊆ Fun⊗ (MfldB,fin
n , C) the full subcategory spanned by the B-framed
homology theories.
Example 6.5.9. Let B Ð→ BTop(n) be a boundary tangent structure and let A be
a B-disk algebra. Then the symmetric monoidal functor ∫(−) ∶ (MfldB,fin
n )⊗ Ð→ C⊗
described in Remark 6.4.9 is a B-framed homology theory. This follows immediately
from Proposition 6.4.17.
Remark 6.5.10. If F ∶ MfldB,fin
n Ð→ C is a symmetric monoidal functor then F∣E⊗
B
is by definition an E⊗
B -algebra object. Remark 6.4.8 then furnishes a symmetric
monoidal natural transformation
∫ F∣E⊗ ⇒ F(−)
(−) B
(6.47) ∫ A Ð→ F(M )
M
of Remark 6.5.10 is an equivalence for every B-framed n-manifold M of finite type.
Proof. We prove by double induction on the open handle decomposition of M . For
integers 0 ≤ k, m let us say that an open manifold M has type (k, m) if it can be
obtained from ∅ by adding finitely many handles of index ≤ k out of which at most
m handles are of index exactly k. We first note that an open n-manifold is of type
(0, 1) if and only if it is the n-disk, and the map (6.47) is an equivalence in this
case by definition.
LITTLE CUBE ALGEBRAS AND FACTORIZATION HOMOLOGY COURSE NOTES 115
Now suppose we have proven that (6.47) is an equivalence for every B-framed
manifold of type (k, m) where either k > 0 or k = 0 and m ≥ 1, and let M be a
B-framed manifold of type (k, m+1). Then by definition there exists an n-manifold
M0 of type (k, m) and a left [∂ ◻k ×◻n−k ]-collar ρ ∶ (0, 1) × ∂ ◻k ×◻n−k ↪ M0 such
that
M ∶= ◻k × ◻n−k ∐ M0 .
(0,1)×∂◻k ×◻n−k
≃ ≃
∫M A / F(M )
in which the vertical maps are equivalences since F and ∫(−) A are homology theories.
To show that the bottom horizontal map is an equivalence it will hence suffice to
show that the top vertical map is an equivalence. We now observe that if U ⊆ I is
an open subset homeomorphic to either ◻1 or ◻1∂ then p−1 (U ) is an open manifold
which is homeomorphic to either M0 , ◻k × ◻n−k ≅ ◻n or (0, 1) × ∂ ◻k ×◻n−k , which
are manifolds of types (k, m), (k − 1, 2) and (0, 1), respectively By the induction
hypothesis the map
∫ p∗ A Ð→ F(U )
U
is an equivalence for every such U ⊆ I, and so the top vertical map of (6.48) is
an equivalence, as desired. We may hence conclude that (6.47) is an equivalence
for every manifold of type (k, m + 1). By induction on m we now get that (6.47)
is an equivalence for every manifold of type (k, m′ ) for m′ ≥ 0, and hence for
every manifold of type (k + 1, 0). By induction on k we now get that (6.47) is an
equivalence for any open n-manifold of finite type, as desired.
Corollary 6.5.12 ([1, Theorem 3.24]). Restriction along E⊗ B,fin ⊗
B ↪ (Mfldn ) de-
termines an equivalence
≃
H(MfldB,fin
n , C) Ð→ AlgEB (C)
between B-framed homology theories with values in C and EB -algebra objects in C.
Remark 6.5.13. We focused in this section on open n-manifolds, but there is no
obstacle in extending the notion of homology theories, as well as Theorem 6.5.11
and Corollary 6.5.12, to the case of n-manifolds with boundary. To do so, one just
needs to spell out what it means for an n-manifold with boundary to be of finite
type, a notion which can be formulated using suitable handles with boundary.
7. Poincaré-Koszul duality
7.1. Factorization homology with support.
7.2. En -suspension and En -loops.
116 YONATAN HARPAZ
References
[1] Ayala D., Francis J., Factorization homology of topological manifolds, Journal of Topology,
8.4, 2015, p. 1045–1084.
[2] Dugger D., Isaksen D., Mathematische Zeitschrift, 246(4), 2004, p. 246–667
[3] Hatcher A., Algebraic topology, Cambridge University Press, Cambridge, (2002
[4] Johnstone P., Sketches of an elephant: A topos theory compendium, Vol. 2, Oxford University
Press, 2002.
[5] Lurie J., Higher Topos Theory, Annals of Mathematics Studies, 170, Princeton University
Press, 2009, [Link]
[6] Lurie J. Higher Algebra, 2014, available at [Link]
[7] May J.P., The geometry of iterated loop spaces, Vol. 271, Springer, 2006.