Derived Algebraic Geometry II: Noncommutative Algebra

October 8, 2009

Contents
1 Monoidal ∞-Categories 4
1.1 Monoidal Structures and Algebra Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Cartesian Monoidal Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Subcategories of Monoidal ∞-Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Free Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.5 Limits and Colimits of Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.6 Monoidal Model Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.7 Digression: Segal Monoidal ∞-Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2 Modules 54
2.1 Tensored ∞-Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.2 Nonunital Algebras and Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.3 Limits and Colimits of Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.4 Free Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.5 Modules in a Monoidal Model Category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.6 The ∞-Category of Tensored ∞-Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.7 Endomorphism Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.8 Application: Existence of Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

3 Monads and the Barr-Beck Theorem 95

3.1 ∞-Categories of Endofunctors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.2 Adjunction Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3 Pointed Adjunction Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.4 Split Simplicial Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.5 The Barr-Beck Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.6 Existence of Adjunction Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

4 The Monoidal Structure on Stable Homotopy Theory 133

4.1 Tensor Products of Presentable ∞-Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.2 The Smash Product Monoidal Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.3 Associative Ring Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.4 Modules over A∞ -Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.5 Balanced Pairings and the Bar Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.6 Flat Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.7 Finiteness Properties of Rings and Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

1
Introduction
Let C be an abelian category, and let X be an object of C. Then the set of endomorphisms HomC (X, X) is
endowed with the structure of an associative ring: the addition on HomC (X, X) is determined by the additive
structure on C, and the multiplication is given by composition. In the higher categorical setting, one should
consider not a set of maps HomC (X, X), but instead a space of maps MapC (X, X). In this situation, the
appropriate analogue of an abelian category is a stable ∞-category (see [41]). If C is a stable ∞-category
containing an object X, then the space MapC (X, X) should itself be viewed as a kind of ring. There are
several approaches to making this precise:

(1) Let H denote the homotopy category of topological spaces. If C is an ∞-category, then the homotopy
category hC is naturally enriched over H (see §T.1.2.2). If C is stable, then hC is additive; one can then
use classical reasoning to deduce that for every X ∈ C, the space E = MapC (X, X) has the structure
of a ring object of H: that is, there exist addition and multiplication maps E × E → E which satisfy
the axioms for an associative ring, up to homotopy. Unfortunately, the theory of ring objects in the
homotopy category H is not a very useful one, because the category H is very badly behaved: for
example, H does not admit finite limits or colimits, so it is impossible to carry out even very simple
algebraic constructions in H.

(2) To avoid the difficulties inherent in approach (1), it is tempting to work in a more rigid setting, such as
the theory of topological rings. A topological ring is simply a ring E equipped with a topology, for which
the ring operations are continuous. In this case, one can develop a good theory of (homotopy) limits
and colimits, and carry out a wide variety of constructions which mimic classical algebra. However,
the setting of topological rings is very restrictive. Typically, if C is a stable ∞-category containing an
object X, the space MapC (X, X) is not homotopy equivalent to a topological abelian group, let alone
a topological ring.

To obtain a good theory, it is necessary to find some middle ground between (1) and (2): we cannot
generally arrange that the space E = MapC (X, X) satisfies the ring axioms “on the nose”, but it is also
not sufficient to merely assume that these axioms hold up to homotopy. Fortunately, there is a good theory
which lies between these two extremes: namely, the theory of associative ring spectra or, as we will call them,
A∞ -rings. The theory of A∞ -rings is a generalization of classical (noncommutative) ring theory, in the same
sense that stable homotopy theory is a generalization of the classical theory of abelian groups. The theory of
A∞ -rings and their modules is a fundamental tool in the study of higher categorical mathematics, which can
be applied in a great variety of situations. For example, a theorem of Schwede and Shipley asserts that many
stable ∞-categories (and, by extension, many triangulated categories) can be realized as ∞-categories of
modules over suitably chosen A∞ -rings (Theorem 4.4.9). We also have a more specific reason to study A∞ -
rings: they are a natural stepping stone towards their commutative cousins, E∞ -rings, which are essential
to the foundations of derived algebraic geometry.
Our goal in this paper is to introduce the definition of A∞ -rings from an ∞-categorical point of view. An
ordinary associative ring can be viewed as an algebra object of the category of abelian groups A, where A
is endowed with the structure of a monoidal category via the tensor product. We wish to find an analogue
of this definition, where the ordinary category A is replaced by the ∞-category of spectra. For this, we will
need a higher-categorical version of the theory of monoidal categories. The construction of this theory will
occupy the bulk of this paper.
We will begin in §1 by giving the definition of a monoidal ∞-category. Roughly speaking, a monoidal ∞-
category consists of an ∞-category C, equipped with a distinguished object 1C ∈ C (the unit) and a bifunctor
⊗ : C × C → C, which is unital and associative up to coherent homotopy. We will describe several sources
of examples of monoidal ∞-categories, and various constructions for producing new monoidal ∞-categories
from old. We will also introduce the notion of an algebra object of a monoidal ∞-category C. Roughly
speaking, this is an object A ∈ C equipped with a multiplication A ⊗ A → A and a unit map 1C → A, which
are again unital and associative up to coherent homotopy. The collection of all algebra objects of C can itself

2
be organized into an ∞-category, which we will denote by Alg(C). We will study the ∞-category Alg(C) in
some detail. In particular, we will introduce criteria which imply the existence of a good supply of limits
and colimits in Alg(C), and which guarantee that the forgetful functor Alg(C) → C admits a left adjoint.
Let C be a monoidal ∞-category and A an algebra object of C. In this case, we can speak of A-modules
in C: that is, objects M ∈ C equipped with a structure map A ⊗ M → M which is unital and associative up
to coherent homotopy. In §2, we will study the theory of modules in a slightly more general setting, where
M is taken to be an object of an ∞-category M which is tensored over C (Definition 2.1.1).
Let M be a fixed ∞-category. Then there is a universal example of a monoidal ∞-category C such that M
is tensored over C: namely, one can take C to be the ∞-category Fun(M, M) of endofunctors of M. Algebra
objects of Fun(M, M) are called monads on M. Given a monad T on M, one can define a new ∞-category
ModT (M) of T -modules in M. There is a forgetful functor ModT (M) → M, which admits a left adjoint
M → ModT (M). In general, given a pair of adjoint functors

Mo
F / N,
G

one can canonically associate a monad T on M, such that the functor G : N → M factors through some
functor G0 : N → ModT (M). In classical category theory, the Barr-Beck theorem provides necessary and
sufficient conditions for G0 to be an equivalence. We will prove an ∞-categorical version of the Barr-Beck
theorem in §3 (Theorem 3.5.1).
In §4 we will specialize to the main situation of interest: the case where C is the ∞-category Sp of
spectra. We will see that Sp admits an essentially unique monoidal structure with the property that the
functor ⊗ : Sp × Sp → Sp preserves colimits in each variable (Corollary 4.2.6). The bifunctor can be identified
with the classical smash product operation on spectra. We then define an A∞ -ring to be an algebra object
of Sp. We will then proceed to show that a great deal of classical algebra can be carried out in the setting of
A∞ -rings and their modules: for example, we will introduce a theory of flat modules, and prove an analogue
of Lazard’s theorem: an A-module M is flat if and only if M can be obtained as a filtered colimit of free
A-modules (Theorem 4.6.19).
We should emphasize that the theory of A∞ -rings and their modules is not new. There are various
definitions available in the literature; see, for example, [17]. We have chosen to present the subject using the
language of ∞-categories, which we feel is the natural home for these ideas.

Notation and Terminology

For an introduction to the language of higher category theory (from the point of view taken in this paper),
we refer the reader to [40]. We will use the terminology and results of [40]. We will also use [41] as our
reference for the theory of stable ∞-categories. References to [40] will be indicated by use of the letter T, and
references to [41] will be indicated by use of the letter S. For example, Theorem T.[Link] refers to Theorem
[Link] of [40].
For each integer n ≥ −1, we let [n] denote the linearly ordered set {0, . . . , n} (so that [−1] denotes the
empty set). Throughout this paper, ∆ will denote the category of combinatorial simplices: the objects of
∆ are the linearly ordered sets {[n]}n≥0 , and the morphisms are (nonstrictly) increasing maps of linearly
ordered sets. The category ∆ is equivalent to the larger category of all nonempty linearly ordered finite
sets. We will typically abuse notation by not distinguishing between ∆ and this larger category. In other
words, if J is a nonempty finite linearly ordered set, we will implicitly identify J with an object [n] ∈ ∆ via
an isomorphism of linearly ordered sets α : J ' [n]; we note that there is little risk in doing so, since the
isomorphism α and the integer n are uniquely determined.
If p : X → S is a map of simplicial sets and s is a vertex of S, we will typically write Xs to denote the
fiber X ×S {s}.
We will say that a spectrum X is n-connective if the homotopy groups πi X vanish for i < n. We will say
that X is connective if it is 0-connective: that is, if the negative homotopy groups of X vanish.

3
1 Monoidal ∞-Categories
Our main goal in this section is to introduce the definition of a monoidal ∞-category and to survey some of
the basic examples. We will begin in §1.1 with a brief review of the classical theory of monoidal categories.
We will then show how this theory can be reformulated in such a way that it admits a natural ∞-categorical
generalization (Definition 1.1.2).
Many of the monoidal ∞-categories which we study in this paper will arise via one of the following
constructions:
(1) Let C be an ∞-category which admits finite products. In this case, we can endow C with the Cartesian
monoidal structure, in which the bifunctor ⊗ : C × C → C is given by the Cartesian product. We will
study this example in §1.2.
(2) Let C be a monoidal ∞-category, and let D ⊆ C be a full subcategory which contains the unit object
and is stable under tensor products. Then D inherits the structure of a monoidal ∞-category. In §1.3
we will study this construction together with a dual procedure, which produces monoidal structures
on suitable quotients of C.
(3) Let C be an ordinary monoidal category. Then the nerve N(C) has the structure of a monoidal ∞-
category. In §1.6 we will describe some variations on this observation. For example, we will show that
if A is a (simplicial) monoidal model category, then the underlying ∞-category N(Ao ) of A inherits
the structure of a monoidal ∞-category (Proposition 1.6.5).

If C is a monoidal ∞-category, then we can construct an ∞-category Alg(C) of algebra objects of C; the
definition will be given in §1.1. Our second goal in this section is to analyze the ∞-category Alg(C). In §1.5,
we will establish existence criteria for limits and colimits in Alg(C). The most difficult step is to prove that
Alg(C) admits coproducts (given suitable assumptions on C). To prove this, we will need to know that the
forgetful functor Alg(C) → C admits a left adjoint; in other words, we need to be able to construct the free
algebra generated by an object C ∈ C. We will describe the (rather technical) details of this construction in
§1.4. Our method requires a reformulation of the definition of monoidal ∞-categories, which we will discuss
in §1.7.

1.1 Monoidal Structures and Algebra Objects

Recall that a monoid is a set M equipped with a multiplication M × M → M and a unit object 1 ∈ M
satisfying the identities
1x = x1 = x x(yz) = (xy)z
for all x, y, z ∈ M . Roughly speaking, a monoidal category is a category C equipped with the same kind of
structures: a unit object 1 ∈ C, and a bifunctor ⊗ : C × C. However, in the categorical setting, it is unnatural
to require the identities displayed above to hold as equalities. In general, we do not expect X ⊗ (Y ⊗ Z) to be
equal to (X ⊗ Y ) ⊗ Z. Instead, the associative law should be formulated as the existence of an isomorphism
ηX,Y,Z : (X ⊗Y )⊗Z ' X ⊗(Y ⊗Z). Moreover, the isomorphisms ηX,Y,Z are taken as additional data, and are
required to satisfy further conditions (such as naturality in X, Y , and Z); we refer the reader to §T.A.1.3 for
a detailed definition. If we try to generalize this definition to higher categories, then the equations satisfied
by the isomorphisms ηX,Y,Z should themselves hold only up to isomorphism. It is possible to explicitly
describe all of the relevant data (for example, using the theory of Stasheff associahedra), but the escalation
in complexity is somewhat intimidating; it will be more convenient to proceed in another way.
We begin by considering an example of a monoidal category. Let C be the category of complex vector
spaces, with monoidal structure given by tensor products of vector spaces. Given a pair of vector spaces U and
V , the tensor product U ⊗V is defined by the property that HomC (U ⊗V, W ) can be identified with the set of
bilinear maps U × V → W . In fact, this property really only determines U ⊗ V up to canonical isomorphism:
in order to build an actual tensor product functor, we need to choose some particular construction of U ⊗ V .

4
Because this requires making certain decision in an ad-hoc manner, it is unrealistic to expect an equality of
vector spaces U ⊗ (V ⊗ W ) = (U ⊗ V ) ⊗ W . However, the existence of a canonical isomorphism between
U ⊗ (V ⊗ W ) and (U ⊗ V ) ⊗ W is easily explained: linear maps from either into a fourth vector space X
can be identified with trilinear maps U × V × W → X.
The above example suggests that we might reformulate the definition of a monoidal category as follows.
Rather than give a bifunctor ⊗ : C × C → C, we instead specify, for each n-tuple (C1 , . . . , Cn ) of objects of
C and each D ∈ C, the collection of morphisms C1 ⊗ . . . ⊗ Cn → D. Of course, we also need to specify how
such morphisms are to be composed. The relevant data can be encoded in a new category C⊗ :
Definition 1.1.1. Let (C, ⊗) be a monoidal category. We define a new category C⊗ as follows:
(i) An object of C⊗ is a finite (possibly empty) sequence of objects of C, which we will denote by
[C1 , . . . , Cn ].

(ii) A morphism from [C1 , . . . , Cn ] to [C10 , . . . , Cm

0
] in C⊗ consists of a nonstrictly order-preserving map
f : [m] → [n], and a collection of morphisms Cf (i−1)+1 ⊗ . . . ⊗ Cf (i) → Ci0 for 1 ≤ i ≤ m.

(iii) Composition in C⊗ is determined by composition of order preserving maps, composition in C, and the
associativity and unit constraints of the monoidal structure on C.
Let (C, ⊗) be a monoidal category, and let C⊗ be as in Definition 1.1.1. Then there is forgetful functor
p : C⊗ → ∆op , which carries an object [C1 , . . . , Cn ] to the linearly ordered set [n]. Moreover, p has the
following properties:

(M 1) The functor p is an op-fibration of categories. In other words, for every object [C1 , . . . , Cn ] ∈ C⊗ and
every morphism f : [n] → [m] in ∆op , there exists a morphism f : [C1 , . . . , Cn ] → [C10 , . . . , Cm
0
] which
covers f , and is universal in the sense that composition with f induces a bijection

HomC⊗ ([C10 , . . . , Cm
0
], [C100 , . . . , Ck00 ]) → HomC⊗ ([C1 , . . . , Cn ], [C100 , . . . , Ck00 ]) ×Map∆ ([k],[n]) Map∆ ([k], [m])

for every object [C100 , . . . , Ck00 ] ∈ C⊗ . To achieve this, it suffices to choose f such that the maps
Cf (i−1)+1 ⊗ . . . ⊗ Cf (i) → Ci0 are isomorphisms for 1 ≤ i ≤ m.

(M 2) Let C⊗ ⊗
[n] denote the fiber of p over the object [n] ∈ ∆ . Then C[1] is equivalent to C. More generally,
op

C⊗
[n] is equivalent to an n-fold product of copies of C. The equivalence is induced by functors associated
to the inclusions [1] ' {i − 1, i} ⊆ [n], 1 ≤ i ≤ n.

We now observe that the monoidal structure on C is determined, up to canonical equivalence, by C⊗ .

More precisely, suppose given a functor p : D → ∆op satisfying conditions (M 1) and (M 2). Then:
(a) Condition (M 2) implies that D[0] has a single object, up to equivalence. The projection [1] → [0]
determines a functor D[0] → D[1] ' C, which we can identify with an object 1 ∈ C.
(b) The inclusion [1] ' {0, 2} ⊆ [2] determines a functor C × C ' D[2] → D[1] ' C, which we may denote
by ⊗.
(c) The commutative diagram
{0, 3} / {0, 1, 3}

 
{0, 2, 3} / {0, 1, 2, 3}

5
 in ∆ determines a diagram of categories and functors (which commutes up to canonical isomorphism):

D[1] o D[2]
O O

D[2] o D[3] .

Combining this with the equivalences D[n] ' Cn , we obtain a functorial isomorphism

ηA,B,C : (A ⊗ B) ⊗ C ' A ⊗ (B ⊗ C).

A similar argument can be used to construct canonical isomorphisms

1 ⊗ X ' X ' X ⊗ 1.

It is not difficult to see that (a), (b), and (c) endow C with the structure of a monoidal category. For
example, MacLane’s pentagon axiom asserts that the diagram

((A ⊗ B) ⊗ C) ⊗ D
jj TTTT
j jjjjjj TTηTA⊗B,C,D
TTTT
jjjj TTTT
ujj ηA,B,C ⊗idD
)
(A ⊗ (B ⊗ C)) ⊗ D (A ⊗ B) ⊗ (C ⊗ D)
ηA,B⊗C,D ηA,B,C⊗D
 idA ⊗ηB,C,D 
A ⊗ ((B ⊗ C) ⊗ D) / A ⊗ (B ⊗ (C ⊗ D))

is commutative. This follows from the fact that all five expressions can be canonically identified with the
image of (A, B, C, D) under the composite functor C4 ' D[4] → D{0,4} ' C . In the case where C is equipped
with a monoidal structure and D = C⊗ , then it is easy to see that the data provided by (a), (b) and (c)
recovers the original monoidal structure on C (up to canonical isomorphism). Conversely, an arbitrary functor
D → ∆op satisfying (M 1) and (M 2) determines a monoidal structure on C and an equivalence D ' C⊗ .
In other words, specifying a monoidal structure on C is equivalent to specifying the functor C⊗ → ∆op .
However, the second approach has several advantages over the first:

• As we saw above in the case of vector spaces, it is sometimes easier to specify the category C⊗ than to
specify the bifunctor ⊗, in the sense that it requires fewer arbitrary choices.
• Axioms (M 1) and (M 2) concerning the functor C⊗ → ∆op are a bit simpler than the usual definition
of a monoidal category. Complicated statements, such as the commutativity of MacLane’s pentagon,
are consequences of (M 1) and (M 2).

The significance of the latter point becomes more apparent in the ∞-categorical setting, where we expect
the MacLane pentagon to be only the first step in a hierarchy of coherence conditions of ever-increasing
complexity. Fortunately, the above discussion suggests an approach which does not require us to formulate
these conditions explicitly.
Definition 1.1.2. A monoidal ∞-category is a coCartesian fibration of simplicial sets p : C⊗ → N(∆)op
with the following property:
(∗) For each n ≥ 0, the associated functors C⊗ ⊗
[n] → C{i,i+1} determine an equivalence of ∞-categories

C⊗ ⊗ ⊗ ⊗ n
[n] → C{0,1} × . . . × C{n−1,n} ' (C[1] ) .

6
Remark 1.1.3. Let p : C⊗ → N(∆)op be a monoidal ∞-category. We will refer to the fiber C = C⊗ [1] as the
⊗ ⊗
underlying ∞-category of C ; we will also say that C is a monoidal structure on C. Assertion (∗) implies
that C⊗ ⊗
[0] is a contractible Kan complex. The projection [1] → [0] in ∆ determines a functor C[0] → C[1] ,
⊗

⊗ ⊗
which we can identify with an object 1C ∈ C[1] , well-defined up to equivalence. We will refer to 1C ∈ C[1] as
the unit object of C⊗
[1] .
The three embeddings of [1] into [2] determine a diagram

θ θ0
C × C ' C⊗ ⊗ ⊗ ⊗
{0,1} × C{1,2} ← C[2] → C{0,2} ' C,

where θ is an equivalence. Composing θ0 with a homotopy inverse to θ, we obtain a functor ⊗ : C × C → C,

which is again well-defined up to equivalence.

Remark 1.1.4. Let p : C⊗ → N(∆)op be a monoidal ∞-category in the sense of Definition 1.1.2, and let
f : C⊗[1] → C be an equivalence of ∞-categories. In this situation, we will generally abuse terminology by
saying that C is a monoidal ∞-category, or that p : C⊗ → N(∆)op exhibits C as a monoidal ∞-category.
Informally, we can think of C⊗ as encoding C together with the the tensor operation ⊗ : C × C → C. Of
course, Definition 1.1.2 encodes a good deal more structure; this additional structure expresses the idea that
the operation ⊗ is associative, up to coherent homotopy.
Example 1.1.5. Let C be a monoidal category, and let C⊗ be as in Definition 1.1.1. Then the induced map
N(C⊗ ) → N(∆)op is a monoidal ∞-category. We will generalize this construction in §1.6.
Remark 1.1.6. Let C⊗ → N(∆)op be a monoidal structure on an ∞-category C. Then the induced functor
hC⊗ → ∆op satisfies the axioms (M 1) and (M 2), and therefore determines a monoidal structure on the
homotopy category hC.
In order to make effective use of the theory of monoidal ∞-categories, we will need an associated theory
of monoidal functors.

Definition 1.1.7. A morphism f : [m] → [n] in ∆ is convex if f is injective and the image {f (0), . . . , f (m)} ⊆
[n] is a convex subset of [n].
Definition 1.1.8. Let C⊗ and D⊗ be monoidal ∞-categories. We will say that a functor F : C⊗ → D⊗ is
monoidal if the diagram
C⊗ G
F / D⊗
GG vv
GGp v
GG vv
GG vvv q
# {v
N(∆)op
commutes, and F carries p-coCartesian morphisms to q-coCartesian morphisms. We will say that F is
lax monoidal if the above diagram commutes, and the following weaker condition is satisfied: for every
p-coCartesian morphism α in C⊗ such that p(α) is a convex morphism in ∆, the morphism F (α) is q-
coCartesian. We let FunLax (C⊗ , D⊗ ) denote the full subcategory of MapN(∆)op (C⊗ , D⊗ ) spanned by the lax
monoidal functors, and FunMon (C⊗ , D⊗ ) the full subcategory of FunLax (C⊗ , D⊗ ) spanned by the monoidal
functors.

Remark 1.1.9. Let p : C⊗ → N(∆)op and q : D⊗ → N(∆)op be monoidal ∞-categories, and let F ∈
MapN(∆)op (C⊗ , D⊗ ). Using the equivalence D⊗ ⊗ n
[n] ' (D[1] ) , we see that F is a (lax) monoidal functor if and
only if, for every p-coCartesian morphism f in C⊗ covering a (convex) map [1] → [n] in ∆, the image F (f )
is a q-coCartesian morphism in D.

7
Remark 1.1.10. Roughly speaking, a lax monoidal functor F : C⊗ → D⊗ consists of a functor f between the
underlying ∞-categories C and D equipped with a coherently associative collection of natural transformations

f (C1 ) ⊗ . . . ⊗ f (Cn ) → f (C1 ⊗ . . . ⊗ Cn ).

The functor F is monoidal if and only if each of these transformations is an equivalence.

Remark 1.1.11. The notion of a lax monoidal functor as given by Definition 1.1.8 is the ∞-categorical
analogue of what we called a right lax monoidal functor in §T.A.1.3. There is also a dual notion which
corresponds to left lax monoidal functors, but this notion is not so easily encoded using the formalism
introduced above; see Remark 1.2.16.
It follows immediately from the definition that the class of (lax) monoidal functors is stable under
composition. Consequently, we may define simplicial categories Cat∆,Mon
∞ ⊆ Cat∆,Lax
∞ as follows:

(1) The objects of Cat∆,Mon

∞ and Cat∆,Lax
∞ are monoidal ∞-categories C⊗ → N(∆)op .
(2) Given a pair of monoidal ∞-categories C⊗ and D⊗ , we let

MapCat∆,Mon
∞
(C⊗ , D⊗ ) ⊆ FunMon (C⊗ , D⊗ ) MapCat∆,Lax
∞
(C⊗ , D⊗ ) ⊆ FunLax (C⊗ , D⊗ )

be the largest Kan complexes contained in FunMon (C⊗ , D⊗ ) and FunLax (C⊗ , D⊗ ), respectively.
Definition 1.1.12. We let CatMon
∞ denote the simplicial nerve N(Cat∆,Mon
∞ ), and CatLax
∞ the simplicial nerve
N(Cat∞ ). We will refer to Cat∞ as the ∞-category of monoidal ∞-categories.
∆,Lax Mon

Remark 1.1.13. Let F : C⊗ → D⊗ be a monoidal functor between monoidal ∞-categories. Using Corollary
T.[Link], we deduce that F is an equivalence if and only if F induces an equivalence of underlying ordinary
categories C⊗ ⊗
[1] → D[1] . The analogous assertion for lax monoidal functors is false.

Our next goal is to introduce the notion of an algebra object of a monoidal ∞-category C⊗ → N(∆)op .
We begin by considering the classical case. Let C be a monoidal category. An algebra object of C is an object
A ∈ C equipped with maps
1→A A⊗A→A
which satisfy the usual unit and associativity conditions. In this case, we can define for each n ≥ 0 a map
sn : A⊗n → A, given by iterated multiplication (or by the unit, in the case n = 0); here A⊗n denotes the
n-fold tensor product of A with itself (which is well-defined up to canonical isomorphism, in view of the
associativity constraint on the tensor product ⊗). The associative law can then be reformulated as follows:
given integers n ≥ 0 and k1 , . . . , kn ≥ 0, the composition
sk1 ⊗...⊗skn s
A⊗k1 +...+kn → A⊗n →
n
A

coincides with sk1 +...+kn . When the definition is phrased in this way, an algebra object of C can be regarded
as a section A e of the projection C⊗ → ∆op . More precisely, we associate to each [n] ∈ ∆op the lifting
A([n]) = [A, . . . , A] ∈ C⊗ . To a map f : [m] → [n] in ∆, we associate the “contraction” map A([n])
e e → A([m]),
e
given by (sf (1)−f (0) , . . . , sf (m)−f (m−1) ). The functoriality of this construction encodes the associativity of
the product on A.
Of course, not every section of the projection C⊗ → ∆op corresponds to an algebra object of C. If
A : ∆op → C⊗ is a general section, then we can view A = A([1])
e e ∈ C as our candidate for the “underlying
algebra”. For n ≥ 0, we can view A([n]) e as an n-tuple of objects (C1 , . . . , Cn ) in C. The inclusions
[1] ' {i − 1, i} ⊆ [n] induce maps ηi : Ci → A for 1 ≤ i ≤ n. It is not difficult to see that A e is equivalent
to the section arising from an algebra object (well-defined up to isomorphism) if and only if each ηi is an
isomorphism. This observation, together with Remark 1.1.9, motivates the following definition:

8
Definition 1.1.14. Let C be a monoidal ∞-category. An algebra object of C is a lax monoidal functor
N(∆)op → C⊗ . We let Alg(C) = FunLax (N(∆)op , C⊗ ) denote the ∞-category of algebra objects of C.
More concretely, an algebra object of p : C⊗ → N(∆)op consists of a section of p which carries every
convex morphism in ∆ to a p-coCartesian morphism in C⊗ .
Remark 1.1.15. The notation of Definition 1.1.14 is somewhat abusive: the ∞-category Alg(C) depends
not only on C, but also on its monoidal structure. There is little risk of confusion; the monoidal structure
under consideration should always be clear from context.
Remark 1.1.16. Let C be a monoidal category, and let C⊗ be as in Definition 1.1.1. Then N(C⊗ ) can be
identified with a monoidal structure on the ∞-category N(C). The above discussion shows that Alg(N(C))
is equivalent to the nerve of the ordinary category of algebra objects of C, interpreted in the classical sense.

Remark 1.1.17. Let C be an ∞-category equipped with a monoidal structure. Evaluation at [1] ∈ ∆
determines a functor Alg(C) → C, which we will refer to as the forgetful functor.
Remark 1.1.18. Let C be a monoidal ∞-category and K an arbitrary simplicial set. Then Fun(K, C)
inherits the structure of a monoidal ∞-category, where the tensor product is defined pointwise. More
precisely, suppose that C⊗ → N(∆)op exhibits C as a monoidal ∞-category. Set

Fun(K, C)⊗ = Fun(K, C⊗ ) ×Fun(K,N(∆)op ) N(∆)op .

Then the projection Fun(K, C)⊗ → N(∆)op exhibits Fun(K, C)⊗[1] ' Fun(K, C) as a monoidal ∞-category.
Moreover, we have a canonical isomorphism Alg(Fun(K, C)) → Fun(K, Alg(C)).

1.2 Cartesian Monoidal Structures

Let C be an ordinary category which admits finite products. Then C has the structure of a monoidal category,
with the bifunctor ⊗ : C × C → C given by the Cartesian product. We will refer to this monoidal structure
on C as the Cartesian monoidal structure. Our goal in this section is to give an analogous construction in
the ∞-categorical setting.
Definition 1.2.1. Let C be an ∞-category. We will say that a monoidal structure on C is Cartesian if the
following conditions are satisfied:
(1) The unit object 1C ∈ C is final.
(2) For every pair of objects C, D ∈ C, the canonical maps

C ' C ⊗ 1C ← C ⊗ D → 1C ⊗ D ' D

exhibit C ⊗ D as a product of C and D in the ∞-category C.

If C⊗ → N(∆)op is a Cartesian monoidal structure on an ∞-category C = C⊗ [1] , then we can construct a
⊗ ⊗
functor π : C → C, which is given informally as follows. To an object C ∈ C[n] , corresponding to an n-tuple
(C1 , . . . , Cn ) ∈ Cn , the functor π associates the object π(C) = 1≤i≤n Ci . We will give rigorous construction
Q
of π below (Proposition 1.2.4); first, we axiomatize its properties.

Definition 1.2.2. Let p : C⊗ → N(∆)op be a monoidal ∞-category. A lax Cartesian structure on C⊗ is a

functor π : C⊗ → D satisfying the following condition:
(∗) Let C be an object of C⊗ [n] . For each 1 ≤ i ≤ n, choose a p-coCartesian morphism fi : C → Ci
covering the inclusion [1] ' {i − 1, i} ⊆ [n] in ∆. Then the morphisms π(fi ) exhibit π(C) as a product
D.
Q
1≤i≤n π(C i ) in the ∞-category

9
We will say that π is a weak Cartesian structure if it is a lax Cartesian structure, and the following additional
condition is satisfied:
(∗0 ) Let f : C → C 0 be a p-coCartesian morphism covering the map [1] ' {0, n} → [n]. Then π(f ) is an
equivalence in D.
We will say that a weak Cartesian structure π is a Cartesian structure if π induces an equivalence C⊗
[1] → D.

Example 1.2.3. Let C be an ordinary category which admits finite products. Regard C as endowed with
the Cartesian monoidal structure, and let C⊗ be as in Definition 1.1.1. Define θ : C⊗ → C by the formula
θ([C1 , . . . , Cn ]) = C1 × . . . Cn . The nerve N(θ) : N(C⊗ ) → N(C) is a Cartesian structure on the monoidal
∞-category N(C⊗ ).
It follows immediately from the definition that if C is a monoidal ∞-category and there exists a Cartesian
structure C⊗ → D, then the monoidal structure on C is Cartesian. Our first result is a converse: if C is a
Cartesian monoidal ∞-category, then there exists an essentially unique Cartesian structure on C.
Proposition 1.2.4. Let p : C⊗ → N(∆)op be a Cartesian monoidal structure on an ∞-category C and let D
be an ∞-category which admits finite products. Let Fun× (C⊗ , D) denote the full subcategory of Fun(C⊗ , D)
spanned by the weak Cartesian structures, and let Fun× (C, D) be the full subcategory of Fun(C, D) spanned
by those functors which preserve finite products. The restriction map Fun× (C⊗ , D) → Fun× (C, D) is an
equivalence of ∞-categories.
We will defer the proof until the end of this section.
Our next goal is to show that, if C is an ∞-category which admits finite products, then there exists an
(essentially unique) Cartesian monoidal structure on C. Our strategy is to give an explicit construction of
this monoidal structure.
Notation 1.2.5. The category ∆× is defined as follows:
(1) An object of ∆× consists of an object [n] ∈ ∆ together with a pair of integers i and j which satisfy
0 ≤ i ≤ j ≤ n.
(2) A morphism from ([n], i ≤ j) to ([n0 ], i0 ≤ j 0 ) in ∆× is a map f : [n] → [n0 ] of linearly ordered sets,
with the property that i0 ≤ f (i) ≤ f (j) ≤ j 0 .
The forgetful functor ∆×,op → ∆op is a Grothendieck fibration, so that the induced map of ∞-categories
N(∆× )op → N(∆)op is a Cartesian fibration (Remark T.[Link]).
Remark 1.2.6. The forgetful functor ∆× → ∆ admits a unique section s, defined by s([n]) = ([n], 0 ≤ n).
× ×
Notation 1.2.7. Let C be an ∞-category. We define a simplicial set C e equipped with a map C
e → N(∆)op
by the following universal property: for every map of simplicial sets K → N(∆)op , we have a bijection

e× ) ' HomSet (K ×N(∆)op N(∆× )op , C).

HomN(∆)op (K, C ∆

×
Fix n ≥ 0. We observe that the fiber C e can be identified with the ∞-category of functors f : N(P )op →
[n]
C, where P is the partially ordered set of intervals in [n]: that is, the collection of all subsets of [n] having
e× spanned by those vertices which
the form {i, i + 1, . . . , j} ⊆ [n]. We let C× be the full simplicial subset of C
correspond to those functors f for which the maps f ({i, i + 1, . . . , j}) → f ({k, k + 1}) exhibit f ({i, . . . , j})
as a product f ({i, i + 1}) × . . . × f ({j − 1, j}).
Proposition 1.2.8. Let C be an ∞-category.
e× → N(∆)op is a coCartesian fibration.
(1) The projection p : C

10
 (2) Let α : F → F 0 be a morphism of C e× whose image in N(∆)op corresponds to a map s : [n] → [m]. Then
α is p-coCartesian if and only if the induced map F ({s(i), . . . , s(j)}) → F 0 ({i, . . . , j}) is an equivalence
in C, for every 0 ≤ i ≤ j ≤ m.
(3) The projection p restricts to a coCartesian fibration C× → N(∆)op (with the same class of p-coCartesian
morphisms).
(4) The projection C× → N(∆)op is a monoidal ∞-category if and only if C admits finite products.
(5) Suppose that C admits finite products. Let π : C× → C be the map given by composition with the section
N(∆)op → N(∆× )op (see Remark 1.2.6). Then π is a Cartesian structure on C× .
Proof. Assertions (1) and (2) follow immediately from Corollary T.[Link], and (3) follows from (2) (since
C× is stable under the pushforward functors associated to the coCartesian fibration p).
We now prove (4). If C has no final object, then C× [0] is empty; consequently, we may assume without loss
×
of generality that C has a final object. Then C[1] is equivalent to the ∞-category of diagrams X0 ← X → X1
in C, where X0 and X1 are final. It follows that π induces an equivalence C× ×
[1] ' C. Consequently, C is a
× × ×
monoidal ∞-category if and only if, for each n ≥ 0, the natural map φ : C[n] → C{0,1} × . . . × C{n−1,n} is an
equivalence of ∞-categories. Let P denote the partially ordered set of subintervals of [n], and let P0 ⊆ P
denote the subset consisting of the intervals {k, k + 1}, where 0 ≤ k < n. Then C× [n] can be identified with
the set of functors F : N(P )op → C which are right Kan extensions of F | N(P0 )op , and φ coincides with
the restriction map from P to P0 . According to Proposition T.[Link], φ is fully faithful, and is essentially
surjective if and only if every functor F0 : N(P0 )op → C admits a right Kan extension to N(P )op . Unwinding
the definitions, we see that this is equivalent to the assertion that every finite collection of objects of C admits
a product in C. This completes the proof of (4). Assertion (5) follows immediately from the construction of
C× (and the description of the p-coCartesian morphisms supplied by Corollary T.[Link]).
Proposition 1.2.9. Let p : C⊗ → N(∆)op be a monoidal ∞-category and let D be an ∞-category which
admits finite products. Let Fun×,Lax (C⊗ , D) denote the full subcategory of Fun(C⊗ , D) spanned by the lax
Cartesian structures and Fun× (C⊗ , D) ⊆ Fun×,Lax (C⊗ , D) the full subcategory spanned by the weak Cartesian
structures. Let π : D× → D be the Cartesian structure of Proposition 1.2.8. Then composition with π induces
trivial Kan fibrations

θ : FunLax (C⊗ , D× ) → Fun×,Lax (C⊗ , D) θ0 : FunMon (C⊗ , D× ) → Fun× (C⊗ , D).

Proof. Unwinding the definitions, we can identify FunLax (C⊗ , D× ) with the full subcategory of

Map(C⊗ ×N(∆)op N(∆× )op , D)

spanned by those functors F which satisfy the following conditions:

(1) For every 0 ≤ i ≤ j ≤ n and every C[n] ∈ C⊗
[n] , F induces an equivalence
Y
F (C[n] , i ≤ j) → F (C[n] , i ≤ i + 1)
0≤i<n

in the ∞-category D.
(2) For every p-coCartesian morphism C[n] → C[m] covering a convex morphism α : [m] → [n] in ∆, and
every 0 ≤ i ≤ j ≤ m, the induced map F (C[n] , α(i) ≤ α(j)) → F (C[m] , i ≤ j) is an equivalence in D.
The functor F 0 = π ◦ F can be described by the formula F 0 (C[n] ) = F (C[n] , 0 ≤ n). In other words, F 0 can
be identified with the restriction of F to the full subcategory of C(0) ⊆ C⊗ ×N(∆)op N(∆× )op spanned by
objects of the form (C[n] , 0 ≤ n).

11
 We observe that for every object X = (C[n] , i ≤ j) of the fiber product C⊗ ×N(∆)op N(∆× )op , the ∞-
category C(0)X/ has an initial object. More precisely, if we choose a p-coCartesian morphism α : C[n] →
C{i,...,j} lifting the inclusion α : {i, . . . , j} ⊆ [n], then the induced map α : (C[n] , i ≤ j) → (C{i,...,j} , i ≤ j)
is an initial object of C(0)X/ . It follows that every functor F 0 : C(0) → D admits a right Kan extension to
C⊗ ×N(∆)op N(∆× )op , and that an arbitrary functor F : C⊗ ×N(∆)op N(∆× )op → D is a right Kan extension
of F | C(0) if and only if F (α) is an equivalence, for every α defined as above.
Let E be the full subcategory of Fun(C⊗ ×N(∆)op N(∆× )op , D) spanned by those functors F which satisfy
the following conditions:
(10 ) The restriction F 0 = F | C(0) is a lax Cartesian structure on C⊗ ' C(0).
(20 ) The functor F is a right Kan extension of F 0 .
Using Proposition T.[Link], we conclude that the restriction map E → Fun×,Lax (C⊗ , D) is a trivial fibration
of simplicial sets. To prove that θ is a trivial Kan fibration, it will suffice to show that conditions (1) and
(2) are equivalent to conditions (10 ) and (20 ).
Suppose first that (10 ) and (20 ) are satisfied by a functor F . Condition then (1) follows easily. Choose a
map C[n] → C[m] , covering a convex morphism [m] → [n] in ∆, and let 0 ≤ i ≤ j ≤ m be as in the statement
of (2). We have a homotopy commutative diagram
F (C[n] , α(i) ≤ α(j)) / F (C[m] , i ≤ j)

 
F (C{α(i),...,α(j)} , α(i) ≤ α(j)) F (C{i,...,j} , i ≤ j).

Condition (20 ) implies that the vertical maps are equivalences, and the lower horizontal map is the identity.
It follows that the upper horizontal map is an equivalence, which proves (2).
Now suppose that (1) and (2) are satisfied. The implication (2) ⇒ (20 ) is obvious; it will therefore suffice
to verify (10 ). Let C[n] be an object of C⊗
[n] , and choose p-coCartesian morphisms gi : C[n] → C{i,i+1} . We
wish to show that the induced map
F (C[n] , 0 ≤ n) → F (C{0,1} , 0 ≤ 1) × . . . × F (C{n−1,n} , n − 1 ≤ n)
is an equivalence, which follows immediately from (1) and (20 ). This completes the proof that θ is a trivial
Kan fibration.
To prove that θ0 is a trivial Kan fibration, it will suffice to prove that θ0 is a pullback of θ. In other
words, it will suffice to show that if F : C⊗ ×N(∆)op N(∆× )op → D is a functor satisfying conditions (1) and
(2), then F | C(0) is a weak Cartesian structure on C⊗ if and only if F determines a monoidal functor from
C⊗ into D× . Let q : D× → N(∆)op denote the projection map. Using the description of the q-coCartesian
morphisms provided by Proposition 1.2.8, we see that the latter condition is equivalent to
(2+ ) For every p-coCartesian morphism α : C[n] → C[m] covering a map α : [m] → [n] in ∆, and every
0 ≤ i ≤ j ≤ m, the induced map F (C[n] , f (i) ≤ f (j)) → F (C[m] , i ≤ j) is an equivalence in D.
Moreover, F | C(0) is a weak Cartesian structure if and only if F satisfies the following:
(30 ) For every n ≥ 0 and every p-coCartesian morphism α : C[n] → C[1] in C⊗ lifting the map [1] ' {0, n} ⊆
[n], the induced map F (C[n] , 0 ≤ n) → F (C[ 1], 0 ≤ 1) is an equivalence in D.
It is clear that (2+ ) implies (30 ). Conversely, suppose that (30 ) is satisfied, and let α and 0 ≤ i ≤ j ≤ m
be as in the statement of (2+ ). Consider the diagram
F (C[n] , α(i) ≤ α(j)) / F (C[m] , i ≤ j)

 
F (C{α(i),...,α(j)} , α(i) ≤ α(j)) / F (C{i,...,j} , i ≤ j).

12
Condition (20 ) implies that the vertical maps are equivalences, and two applications of (30 ) implies that
the lower horizontal map is an equivalence as well. It follows that the upper horizontal map is also an
equivalence, as desired.
We are now in a position to establish the uniqueness of Cartesian monoidal structures. Let CatMon,×
∞
denote the full subcategory of CatMon
∞ spanned by the Cartesian monoidal ∞-categories. Let CatCart
∞ denote
the subcategory of Cat∞ whose objects are ∞-categories which admit finite products, and whose morphisms
are functors which preserve finite products.
Corollary 1.2.10. The forgetful functor θ : CatMon,×
∞ → CatCart
∞ is an equivalence of ∞-categories.
Proof. We first observe that if C is an ∞-category which admits finite products, then Proposition 1.2.8
implies that the monoidal ∞-category C× is a preimage of C under the forgetful functor θ. It follows that
θ is essentially surjective. Moreover, if C⊗ is any Cartesian monoidal structure on C, then Proposition 1.2.4
guarantees the existence of a Cartesian structure π : C⊗ → C. Applying Proposition 1.2.9, we can lift π to a
monoidal functor F : C⊗ → C× . Remark 1.1.13 implies that F is an equivalence.
We now show that θ is fully faithful. Let C⊗ and D⊗ be Cartesian monoidal structures on ∞-categories
C and D. We wish to show that the restriction map

MapCatMon,×
∞
(C⊗ , D⊗ ) → MapCatCart
∞
(C, D)

is a homotopy equivalence. We will prove a slightly stronger assertion: namely, that the restrictio map ψ :
FunMon (C⊗ , D⊗ ) → Fun× (C, D) is a categorical equivalence, where Fun× (C, D) denotes the full subcategory
of Fun(C, D) spanned by those functors which preserve finite products. In view of the above remarks, it
suffices to prove this in the case where D⊗ = D× . In this case, the map ψ factors as a composition
ψ0 ψ 00
FunMon (C⊗ , D⊗ ) → Fun× (C⊗ , D) → Fun× (C, D).

Proposition 1.2.9 implies that ψ 0 is a categorical equivalence, and Proposition 1.2.4 implies that ψ 00 is a
categorical equivalence.
Our next goal is to study the algebra objects in an ∞-category equipped with a Cartesian monoidal
structure. We begin by reviewing a bit of classical category theory. Let C be an ordinary category which
admits finite products. A monoid object of C is an object M ∈ C, equipped with maps

∗ → M, M ×M →M

which satisfy the usual associativity and unit conditions; here ∗ denotes a final object of C. Equivalently, we
can define a monoid object of C to be a contravariant functor from C to the category of monoids, such that
the underlying functor Cop → Set is representable by an object M ∈ C.
Example 1.2.11. If C is the category of sets, then a monoid object of C is simply a monoid M . We can
identify the monoid M with a category DM , having only a single object E with HomDM (E, E) = M . The
nerve N(DM ) is a simplicial set, which is typically denoted by BM and called the classifying space of M .
Concretely, the set of n-simplices of BM can be identified with an n-fold product of M with itself, and the
face and degeneracy operations on BM encode the multiplication and unit operations on M . The functor
M 7→ BM is a fully faithful embedding of the category of monoids into the category of simplicial sets.
Moreover, a simplicial set X is isomorphic to the classifying space of a monoid if and only if, for each n ≥ 0,
the natural map X([n]) → X({0, 1}) × . . . × X({n − 1, n}) is a bijection. In this case, the underlying monoid
is given by X([1]), with unit determined by the degeneracy map ∗ ' X([0]) → X([1]) and multiplication by
d
the face map X([1]) × X([1]) ' X([2]) →1 X([1]).
It follows from Example 1.2.11 that we can identify monoids in an arbitrary category C with certain
simplicial objects of C. This observation allows us to generalize the notion of a monoid to higher category
theory.

13
Definition 1.2.12. Let C be an ∞-category. A monoid object of C is a simplicial object X : N(∆)op → C
with the property that, for each n ≥ 0, the collection of maps X([n]) → X({i, i + 1}) exhibits X([n]) as
a product X({0, 1}) × . . . × X({n − 1, n}). We let Mon(C) denote the full subcategory of Fun(N(∆)op , C)
spanned by the monoid objects of C.
Example 1.2.13. Let C be an ∞-category. Every group object of C (see Definition T.[Link]) is a monoid
object of C.

Proposition 1.2.14. Let C⊗ → N(∆)op be a monoidal structure on an ∞-category C = C⊗ [1] , and let
⊗
q : C → D be a Cartesian structure. Then composition with q induces an equivalence Alg(C) → Mon(D).
Proof. In view of Proposition 1.2.9 and Remark 1.1.13, we may assume without loss of generality that
C⊗ = D× . We now apply Proposition 1.2.9 again to deduce that the map

Alg(C) = FunLax (N(∆)op , C⊗ ) → Fun×,Lax (N(∆)op , D) = Mon(D)

is a trivial Kan fibration.

Remark 1.2.15. Definition 1.2.12 allows us to reformulate the notion of a monoidal ∞-category. More
precisely, we claim that a monoidal ∞-category is essentially the same thing as a monoid object in the
∞-category Cat∞ . To see this, we let (Set+ ∆ )/ N(∆)op denote the category of marked simplicial sets equipped
with a map to N(∆)op (see §T.3.1), endowed with the opposite of the model structure defined in §T.3.1.3 (so
that the fibrant objects of (Set+ op
∆ )/ N(∆)op can be identified with coCartesian fibrations X → N(∆) . Then
Cat∞ can be identified with a full subcategory of the underlying ∞-category N((Set∆ )/ N(∆)op ).
Mon + o

Theorem T.[Link] and Proposition T.[Link] furnish equivalences of ∞-categories

Fun(N(∆)op , Cat∞ ) ← N(Fun(∆op , Set+ o + o

∆ ) ) → N((Set∆ )/ N(∆)op ).

Under the composite equivalence, the full subcategory CatMon ∞ ⊆ N((Set+ o

∆ )/ N(∆)op ) can be identified with
Mon(Cat∞ ) ⊆ Fun(N(∆) , Cat∞ ). In other words, we may identify monoidal ∞-categories with monoid
op

objects in the ∞-category Cat∞ .

The above argument suggests yet another possible definition: we can identify monoidal ∞-categories
with certain functors F from ∆op into the ordinary category of ∞-categories, which have the property that
the induced map F ([n]) → F ({0, 1}) × . . . × F ({n − 1, n}) is a categorical equivalence for each n ≥ 0. Such
functors can be identified with bisimplicial sets satisfying appropriate extension conditions; we leave the
details to the reader. We will later obtain an even more concrete model for the ∞-category Mon(Cat∞ ):
namely, monoidal ∞-categories can be identified with ∞-categories C equipped with an object 1 ∈ C and a
multiplication ⊗ : C × C → C which is strictly associative (Example 1.6.20).
Although these alternative approaches are perhaps more concrete, they are more difficult to use in
practice. For example, the theory of algebra objects in a monoidal ∞-category is most easily formulated in
terms of Definition 1.1.2.
Remark 1.2.16. The definition of a monoidal ∞-category is not manifestly self-dual. However, it is nev-
erthless true that any monoidal structure on an ∞-category C determines a monoidal structure on Cop ,
which is unique up to contractible ambiguity. Roughly speaking, we can use Remark 1.2.15 to identify a
monoidal ∞-category C with a monoid object N(∆)op → Cat∞ . We can then obtain a new monoid object
by composing with a self-equivalence R : Cat∞ → Cat∞ , which carries each ∞-category to its opposite (for
the existence and uniqueness of the self-equivalence R, we refer the reader to [48]).
Remark 1.2.15 shows that the theory of monoidal ∞-categories is equivalent to the theory of monoid
objects of Cat∞ . Composing with the self-equivalence R allows us pass between monoidal structures on an
∞-category C and monoidal structures on the opposite ∞-category R C ' Cop .

14
Remark 1.2.17. According to Remark 1.2.16, any monoidal structure on an ∞-category Cop determines a
monoidal structure on C, up to contractible ambiguity. In particular, if C admits finite coproducts, then the
Cartesian monoidal structure on Cop determines a monoidal structure on C, which we will call the coCartesian
monoidal structure. It is characterized up to equivalence by the following properties:
(1) The unit object 1C ∈ C is initial.
(2) For every pair of objects C, D ∈ C, the canonical maps

C ' C ⊗ 1C → C ⊗ D ← 1C ⊗ D ' D

exhibit C ⊗ D as a coproduct of C and D in the ∞-category C.

There is an analogue of Proposition 1.2.14, which applies in the situation where C is equipped with a
coCartesian monoidal structure: in this case, the forgetful functor θ : Alg(C) → C is a trivial Kan fibration.
We will prove an analogue of this statement (for symmetric monoidal ∞-categories) in [43].
We close this section with the proof of Proposition 1.2.4.
Proof of Proposition 1.2.4. We define a subcategory I ⊆ ∆op ×[1] as follows:

(a) An object ([n], i) ⊆ ∆op ×[1] belongs to I if and only if either i = 1 or n > 0.
(b) A morphism ([m], i) → ([n], j) in [1] × ∆op belongs to I if and only if either j = 1 or the induced map
[n] → [m] preserves endpoints.

Let C0 denote the fiber product C⊗ ×N(∆)op N(I), which we regard as a subcategory of C⊗ ×∆1 , and
let p0 : C0 → N(I) denote the projection. Let C00 and C01 denote the intersections of C0 with C⊗ ×{0} and
C⊗ ×{1}, respectively. We note that there is a canonical isomorphism C01 ' C⊗ .
Let E denote the full subcategory of Fun(C0 , D) spanned by those functors F which satisfy the following
conditions:
(i) For every object C ∈ C⊗
[n] , where n > 0, the induced map F (C, 0) → F (C, 1) is an equivalence in D.

(ii) The restriction F | C01 is a weak Cartesian structure on C⊗ .

It is clear that if (i) and (ii) are satisfied, then the restriction F0 = F | C00 satisfies the following additional
conditions:

(iii) The restriction F0 | C⊗

[1] ×{0} is a functor from C to D which preserves finite products.

(iv) For every p0 -coCartesian morphism α in C00 , the induced map F0 (α) is an equivalence in D.

Moreover, (i) is equivalent to the assertion that F is a right Kan extension of F | C01 . Proposition T.[Link]
implies that the restriction map r : E → Fun× (C⊗ , D) induces a trivial Kan fibration onto its essential image.
The map r has a section s, given by composition with the projection map C0 → C⊗ . The restriction map
Fun× (C⊗ , D) → Fun× (C, D) factors as a composition
s e
Fun× (C⊗ , D) → E → Fun× (C, D),

where e is induced by composition with the inclusion C ⊆ C00 ⊆ C0 . Consequently, it will suffice to prove that
e is an equivalence of ∞-categories.
Let E0 ⊆ Fun(C00 , D) be the full subcategory spanned by those functors which satisfy conditions (iii) and
(iv). The map e factors as a composition
e0 e00
E → E0 → Fun× (C, D).

15
Consequently, it will suffice to show that e0 and e00 are trivial Kan fibrations.
Let f : C00 → D be an arbitrary functor, and let C ∈ C⊗ 0
[n] ⊆ C0 , where n > 0. There exists a unique map
α : ([n], 0) → ([1], 0) in I; choose a p0 -coCartesian morphism α : C → C 0 lifting α. We observe that C 0 is an
initial object of C ×(C00 )/C 0 ×C00 C. Consequently, f is a right Kan extension of f | C at C if and only if f (α)
is an equivalence. It follows that f satisfies (iv) if and only if f is a right Kan extension of f | C. The same
argument (and Lemma T.[Link]) shows that every functor f0 : C → D admits a right Kan extension to C00 .
Applying Proposition T.[Link], we deduce that e00 is a trivial Kan fibration.
It remains to show that e0 is a trivial Kan fibration. In view of Proposition T.[Link], it will suffice to
prove the following pair of assertions, for every functor f ∈ E0 :

(1) There exist a functor F : C0 → D which is a left Kan extension of f = F | C00 .

(2) An arbitrary functor F : C0 → D which extends f is a left Kan extension of f if and only if F belongs
to E.

Let (C, 1) ∈ C⊗ 0 0 0
[n] ×{1} ⊆ C . Since there exists a final object 1C ∈ C, the ∞-category C0 ×C C/C also has a
0

⊗
final object, given by the map α : (C , 0) → (C, 1), where C ∈ C[0]?[n]?[0] corresponds, under the equivalence
0 0

C⊗ ⊗
[0]?[n]?[0] ' C × C[n] × C,

to the triple (1C , C, 1C ). We now apply Lemma T.[Link] to deduce (1), together with the following analogue
of (2):

(20 ) An arbitrary functor F : C0 → D which extends f is a left Kan extension of f if and only if, for every
morphism α : (C 0 , 0) → (C, 1) as above, the induced map F (C 0 , 0) → F (C, 1) is an equivalence in D.

To complete the proof, it will suffice to show that F satisfies the conditions stated in (20 ) if and only
if F ∈ E. We first prove the “if” direction. Let α : (C 0 , 0) → (C, 1) be as above; we wish to prove that
F (α) : F (C 0 , 0) → F (C, 1) is an equivalence in D. If n = 0, this is clear, since both sides are final objects of
D. Let us therefore assume that n > 0. The map α factors as a composition
α0 α00
(C 0 , 0) → (C 0 , 1) → (C, 1).

Condition (i) guarantees that F (α0 ) is an equivalence. Condition (ii) guarantees that F (C 0 , 1) is equivalent
to a product F (1C , 1) × F (C, 1) × F (1C , 1), and that F (α00 ) can be identified with the projection onto the
middle factor. Moreover, since 1C is a final object of C, condition (ii) also guarantees that F (1C , 1) is a final
object of D. It follows that F (α00 ) is an equivalence, so that F (α) is an equivalence as desired.
Now let us suppose that F satisfies the condition stated in (20 ). We wish to prove that F ∈ E. Here we
must invoke our assumption that the monoidal structure on C is Cartesian. We begin by verifying condition
(i). Let C ∈ C⊗ 0
[n] , for n > 0, and let α : (C , 0) → (C, 1) be defined as above. There is a unique map
β : ([n], 0) → ([0] ? [n] ? [0], 0) in I, which is the identity on [n] ⊆ [0] ? [n] ? [0]. Choose a p0 -coCartesian
morphism β : (C, 0) → (C 00 , 0) lifting β. Since the final object 1C ∈ C is also the unit object of C, we
β α
can identify C 00 with C 0 . The composition (C, 0) → (C 0 , 1) → (C, 1) is homotopic to the canonical map
γ : (C, 0) → (C, 1) appearing in the statement of (i). Condition (iv) guarantees that F (β) is an equivalence,
and (20 ) guarantees that F (α) is an equivalence. Using the two-out-of-three property, we deduce that F (γ)
is an equivalence, so that F satisfies (i).
To prove that F satisfies (ii), we must verify two conditions:
(ii0 ) If β : (C, 1) → (D, 1) is a p0 -coCartesian morphism in C0 , and the underlying morphism in ∆ is
endpoint-preserving, then F (β) is an equivalence.
(ii1 ) Let C ∈ C⊗ [n] , and choose p-coCartesian morphisms γi : C → Ci covering the inclusions {i − 1, i} ⊆ [n],
for 1 ≤ i ≤ n. Then the maps γi exhibit F (C, 1) as a product F (C1 , 1) × . . . × F (Cn , 1) in D.

16
 Condition (ii0 ) follows immediately from (i) and (iv), provided that C ∈ C⊗ [n] for n > 0. If n = 0, then
the source and target of F (β) are both final objects of D, so F (β) is automatically an equivalence. To prove
(ii1 ), we consider the maps α : (C 0 , 0) → (C, 1) and αi : (Ci0 , 0) → (Ci , 1) which appear in the statement of
(20 ). We have a collection of commutative diagrams

(C 0 , 0)
α / (C, 1)

γi0 γi
 
(Ci0 , 0)
αi
/ (Ci , 1).

Condition (20 ) guarantees that the maps F (α) and F (αi ) are equivalences in D. Consequently, it will suffice
to show that the maps f (γi0 ) exhibit f (C 0 , 0) as a product of f (Ci0 , 0) in D. Let f0 = f | C. Using condition
(iv), we obtain canonical equivalences

f (C 0 , 0) ' f0 (1C ⊗ C1 ⊗ . . . ⊗ Cn ⊗ 1C )

f (Ci0 , 0) ' f0 (1C ⊗ Ci ⊗ 1C )

Since condition (iii) guarantees that f0 preserves products, it will suffice to show that the canonical map
Y
1C ⊗ C1 ⊗ . . . ⊗ Cn ⊗ 1C ' 1C ⊗ Ci ⊗ 1C
1≤i≤n

is an equivalence in the ∞-category C. This follows easily by induction on n (the case n = 2 reduces to our
assumption that the monoidal structure on C is Cartesian.)

1.3 Subcategories of Monoidal ∞-Categories

Let C be an ∞-category. Our goal in this section is to show that a monoidal structure on C determines a
monoidal structure on suitable (full) subcategeroies D ⊆ C. Suppose that D is a full subcategory of C, which
is stable under equivalence in C: that is, if D ∈ D and there is an equivalence C ' D in C, then C ∈ D.
Let C⊗ → ∆op be a monoidal structure on C = C⊗ ⊗ ⊗
[1] . We define a full subcategory D ⊆ C as follows: an
⊗ ⊗ ⊗
object C ∈ C[n] belongs to D if and only if the image of C under the equivalence C[n] ' Cn belongs to
Dn . Our goal in this section is to study circumstances under which D⊗ inherits the structure of a monoidal
∞-category. The most obvious case is that in which D is stable under tensor products in C:
Proposition 1.3.1. Let C⊗ → N(∆)op be a monoidal ∞-category, and let D ⊆ C be a full subcategory which
is stable under equivalence, contains the unit object of C, and is stable under tensor products.
(1) The restricted map D⊗ → N(∆)op is a monoidal ∞-category.
(2) The inclusion D⊗ ⊆ C⊗ is a monoidal functor.
(3) Suppose that the inclusion D ⊆ C admits a right adjoint L (so that D is a colocalization of C). Then
the inclusion D⊗ ⊆ C⊗ admits a right adjoint L⊗ .
(4) Under the hypothesis of (3), the functor L⊗ is lax monoidal.
Proof. Assertions (1) and (2) follow immediately from the definitions. Suppose that D is a colocalization of
C. Let us say that a morphism α : X 0 → X in C⊗ is a colocalization if X ∈ D⊗ , and composition with α
induces a homotopy equivalence MapC⊗ (Y, X 0 ) → MapC⊗ (Y, X) for all Y ∈ C⊗ for all Y ∈ D⊗ . According
to Proposition T.[Link], (3) is equivalent to the following assertion:
(30 ) For every X ∈ C⊗ , there exists a colocalization α : X 0 → X.

17
Assuming this for the moment, (4) is equivalent to the following pair of assertions:
(40 ) For every colocalization α : X 0 → X, the image of α in N(∆)op is an equivalence (and therefore
degenerate).
(400 ) If α : X 0 → X is a colocalization in C⊗
[n] , and [m] → [n] is a convex morphism in ∆, then the image of
⊗
α in C[m] is a colocalization.

Suppose that X ∈ C⊗ ⊗
[n] . Choose a morphism α : X → X in C[n] which corresponds, under the equivalence
0

C⊗[n] ' C , to the canonical map (LC1 , . . . , LCn ) → (C1 , . . . , Cn ). We will show that α is a colocalization
n

map in C⊗ . This will prove (30 ). Assertions (40 ) and (400 ) will likewise follow, since colocalization maps are
unique up to equivalence.
Let Y ∈ D⊗[m] be any object. We wish to show that composition with α induces a homotopy equivalence
0
MapC⊗ (Y, X ) → MapC⊗ (Y, X). Both sides are given a disjoint union, taken over the set of morphisms
β : [n] → [m] in ∆. It will therefore suffice to prove the corresponding result for each summand. Using the
product structure on C⊗ [n] , we can reduce to the case where n = 1 and β is endpoint-preserving. We can
identify Y with an object (D1 , . . . , Dm ) ∈ Dm , and must show that the canonical map

MapC (D1 ⊗ . . . ⊗ Dm , X 0 ) → MapC (D1 ⊗ . . . ⊗ Dm , X)

is an equivalence. This follows, since X 0 → X is a colocalization in C, and D1 ⊗ . . . ⊗ Dm belongs to D.

Example 1.3.2. Let C monoidal ∞-category. We say that an object C ∈ C is invertible if it is an invertible
object of the homotopy category hC: that is, if there exists an object D ∈ C and equivalences

C ⊗ D ' 1C ' D ⊗ C.

Let C0 ⊆ C denote the full subcategory spanned by the invertible objects. It is easy to see that the hypotheses
of Proposition 1.3.1 are satisfied, so that C0 inherits the structure of a monoidal ∞-category.
Remark 1.3.3. Let D ⊆ C be as in the statement of Proposition 1.3.1, and suppose that the inclusion
D ⊆ C admits a right adjoint L. The inclusion i : D⊗ ⊆ C⊗ is a monoidal functor, which induces a
fully faithful embedding Alg(D) ⊆ Alg(C). Let L⊗ be a right adjoint to i, so that L⊗ induces a functor
f : Alg(C) → Alg(D). It is easy to see that f is a right adjoint to the inclusion Alg(D) ⊆ Alg(C). Moreover,
if θ : Alg(C) → C denotes the forgetful functor, then for each A ∈ Alg(C), applying θ to the canonical map
f (A) → A exhibits θ(f (A)) as a colocalization of θ(A).
Proposition 1.3.1 has an evident converse: if D⊗ is a monoidal ∞-category and the inclusion i : D⊗ ⊆ C⊗
is a monoidal functor, then D is stable under tensor products in C. However, it is possible for this converse
to fail if i is only assumed to be weakly monoidal. We now discuss a general class of examples where D is
not stable under tensor products, yet D nonetheless inherits a monoidal structure from C.
Definition 1.3.4. Let C be a monoidal ∞-category, and let L : C → C be a localization functor. We will
say that L is compatible with the monoidal structure on C if the following condition is satisfied:
(∗) Let f : X → Y be a morphism in C, Z ∈ C another object, and f 0 : X ⊗Z → Y ⊗Z, f 00 : Z ⊗X → Z ⊗Y
the induced morphisms. If Lf is an equivalence, then Lf 0 and Lf 00 are equivalences.
Remark 1.3.5. In the situation of Definition 1.3.4, it suffices to verify (∗) in the case where f : X → Y is
equivalent to the localization map X → LX.
Remark 1.3.6. Suppose that C is a presentable ∞-category equipped with a monoidal structure, and that
the tensor product ⊗ : C × C → C preserves small colimits separately in each variable. It is easy to see that
the collection of all maps f which satisfy (∗) is strongly saturated (see Definition T.[Link]). Consequently,
if S is a collection of morphisms of C such that L C = S −1 C, then it suffices to verify (∗) when f ∈ S.

18
Example 1.3.7. Let C be a presentable ∞-category equipped with a monoidal structure, and suppose that
the tensor product ⊗ : C × C → C preserves small colimits separately in each variable. Let n ≥ −2 be an
integer, and let τ≤n C be the collection of n-truncated objects of C (see §T.5.5.6). Then the localization
functor τ≤n is compatible with the monoidal structure on C.
To see this, let F : C → C be a colimit of the constant map ∂ ∆n+2 → {idC } ⊆ Fun(C, C), let α : F → idC
be the canonical transformation, and let S be the collection of all morphisms in C equivalent to those of the
form α(C) : F (C) → C. The proof of Proposition T.[Link] implies that τ≤n C = S −1 C.
Since the tensor product preserves colimits, we have canonical equivalences

F (C) ⊗ D ' F (C ⊗ D) ' C ⊗ F (D).

It follows that if f ∈ S and C ∈ C, then f ⊗ idC and idC ⊗f also belong to S. Now apply Remark 1.3.6.
Our next goal is to prove that if C is a monoidal ∞-category, and L : C → C is a localization functor which
is compatible with the monoidal structure on C, then the full subcategory L C ⊆ C inherits the structure of
a monoidal category (Proposition 1.3.9). First, we need a lemma.
Lemma 1.3.8. Let p : C → D be a coCartesian fibration of ∞-categories. Let L : C → C and L0 : D → D
be localization functors, with essential images L C and L0 D. Suppose that L and L0 are compatible in the
following sense:
(i) The functor p restricts to a functor p0 : L C → L0 D.
(ii) If f is a morphism in C such that Lf is an equivalence, then L0 p(f ) is an equivalence in D.
Then:
(1) The functor L carries p-coCartesian morphisms of C to p0 -coCartesian morphisms of L C.
(2) The functor p0 is a coCartesian fibration.
Proof. Let f : X → Y be a p-coCartesian morphism of C. We wish to prove that Lf is p0 -coCartesian.
According to Proposition T.[Link], it will suffice to show that for every Z 0 ∈ L C, the diagram of Kan
complexes
CLf / ×C {Z 0 } / CLX/ ×C {Z 0 }

 
Dp(Lf )/ ×D {p(Z 0 )} / Dp(LX)/ ×D {p(Z 0 )}

is homotopy Cartesian. Let Z 0 = LZ. Since L2 ' L, we can assume without loss of generality that Z ∈ L C.
Since f is p-coCartesian, Proposition T.[Link] implies that the diagram

Cf / ×C {Z} / CX/ ×C {Z}

 
Dp(f )/ ×D {p(Z)} / Dp(X)/ ×D {p(Z)}

is homotopy Cartesian. Choose a natural transformation α : idC → L which exhibits L as a localization

functor. Then α induces a natural transformation between the above diagrams. It will therefore suffice to
show that each of the induced maps

Cf / ×C {Z} → CLf / ×C {LZ} CX/ ×C {Z} → CLX/ ×C {LZ}

Dp(f )/ ×D {p(Z)} → Dp(Lf )/ ×D {p(LZ)} Dp(LX)/ ×D {p(Z)} → Dp(LX)/ ×D {p(LZ)}

19
is a homotopy equivalence. For the first pair of maps, this follows from the fact that Z ∈ L C. For the second
pair, we observe that (i) and (ii) imply that for every C ∈ C, the map p(α(C)) : p(C) → p(LC) is equivalent
to the L0 -localization p(C) → L0 p(C), and p(Z) ∈ L0 D. This completes the proof of (1).
To prove (2), choose any object C ∈ L C and a morphism f : p(C) → D in L0 D. Choose a p-coCartesian
morphism f : C → D in C. According to (1), the morphism L(f ) : LC → LD is p0 -coCartesian. We now use
0
the fact that p0 is a categorical fibration to lift the equivalence p(α(f )) to an equivalence L(f ) ' f , where
0 0
f : C → D is a p0 -coCartesian morphism lifting f .
Proposition 1.3.9. Let p : C⊗ → N(∆)op be a monoidal structure on the ∞-category C = C⊗
[1] , and let
L : C → C be a compatible localization functor with essential image D ⊆ C. Then:

(1) The inclusion D⊗ ⊆ C⊗ has a left adjoint L⊗ .

(2) The restriction p| D⊗ : D⊗ → N(∆)op is a monoidal structure on the ∞-category D.
(3) The inclusion functor D⊗ ⊆ C⊗ is lax monoidal, and its left adjoint L⊗ : C⊗ → D⊗ is monoidal.

Proof. We first prove (1). According to Proposition T.[Link], it will suffice to show that for every X ∈ C⊗ ,
there exists a localization map X → X 0 where X 0 ∈ D⊗ . Let X lie over [n] ∈ ∆, corresponding to an
n-tuple (C1 , . . . , Cn ) ∈ Cn under the equivalence C⊗ ⊗
[n] ' C . Let α : X → X be a morphism in C[n] , which
n 0

corresponds to a localization map (C1 , . . . , Cn ) → (LC1 , . . . , LCn ). We must show that for every object
Y ∈ D⊗ 0
[m] , composition with α induces a homotopy equivalence MapC⊗ (X , Y ) → MapC⊗ (X, Y ). Both sides
can be expressed as a disjoint union, taken over the set of all maps β : [m] → [n]. Using the product
structure on C⊗ [m] , we can reduce to the case m = 1 and where β is endpoint-preserving. In this case, we can
identify Y with an object D ∈ D. Unwinding the definitions, we are reduced to proving that the natural
map MapC (LC1 ⊗ . . . ⊗ LCn , D) → MapC (C1 ⊗ . . . ⊗ Cn , D) is a homotopy equivalence. Since D ∈ D, it will
suffice to show that the map L(C1 ⊗ . . . ⊗ Cn ) → L(LC1 ⊗ . . . ⊗ LCn ) is an equivalence, which follows easily
from the compatibility of L with the monoidal structure on C. This completes the proof of (1). Moreover,
the proof of (1) shows that L⊗ fits into a commutative diagram
0

C⊗ H
L / C⊗
HH vv
HHp v
HH vv
HH vvv p
# {v
N(∆)op .

Applying the second part of Lemma 1.3.8, we conclude that p| D⊗ is a coCartesian fibration. The equivalence
D⊗ ⊗ ⊗
[n] ' (D) follows by construction. This proves (2). The fact that the inclusion L C ⊆ C is lax monoidal
n

follows by inspection. The assertion that L⊗ is a monoidal functor follows from the first part of Lemma
1.3.8.
We now discuss a slightly more concrete situation, where the localization functors in question are given
by truncation with respect to a t-structure on a stable ∞-category.
Definition 1.3.10. Let C be a stable ∞-category equipped with a t-structure, and let C⊗ → N(∆)op be a
monoidal structure on C. We will say that the monoidal structure on C is compatible with the t-structure on
C if the following conditions are satisfied:
(1) For each C ∈ C, the functors • ⊗ C and C ⊗ • are exact functors from C to itself.
(2) The full subcategory C≥0 contains the unit object of C and is closed under tensor products.
Remark 1.3.11. Let C be as in Definition 1.3.10. Then, for every pair of integers m, n ∈ Z, the tensor
product functor ⊗ carries C≥m × C≥n into C≥n+m . This follows immediately from the exactness of the tensor
product in each variable, and the assumption that the desired result holds in the case m = n = 0.

20
Proposition 1.3.12. Let C be a stable ∞-category equipped with a t-structure, and let C⊗ → N(∆)op be
a compatible monoidal structure on C. Then, for every n ≥ 0, the localization functor τ≤n : C≥0 → C≥0 is
compatible (in the sense of Definition 1.3.4) with the induced monoidal structure on C≥0 (Proposition 1.3.1).
Proof. Let X, Y ∈ C≥0 , and let f : X → τ≤n X be the canonical map. In view of Remark 1.3.5, we need only
show that the maps f ⊗ idY and idY ⊗f become equivalences after applying the functor τ≤n . By symmetry,
it will suffice to treat the case of f ⊗ idY . Since the functor • ⊗ Y is exact, we have a long exact sequence
βk
. . . → πk (τ≥n+1 X ⊗ Y ) → πk (X ⊗ Y ) → πk (τ≤n X ⊗ Y ) → πk−1 (τ≥n+1 X ⊗ Y ) → . . .

in the abelian category C♥ . We wish to show that βk is an isomorphism for k ≤ n. In view of the long exact
sequence, it will suffice to show that πk (τ≥n+1 X ⊗ Y ) vanishes for k ≤ n. In other words, we must show
that τ≥n+1 X ⊗ Y belongs to C≥n+1 , which follows immediately from Remark 1.3.11.
Remark 1.3.13. In the case where C is presentable, the tensor product ⊗ preserves colimits separately in
each variable, and the t-structure on C is accessible, Proposition 1.3.12 is a special case of Example 1.3.7.

1.4 Free Algebras

Let (C, ⊗) be a monoidal category, and let Alg(C) be the category of algebra objects of C. Suppose that
C admits countable coproducts, and that the bifunctor ⊗ : C × C → C preserves countable coproducts
separately in each variable. In this case, the forgetful functor Alg(C) → C admits a left adjoint. This left
adjoint associates to each object C ∈ C the free algebra
a
FreeC (C) = C ⊗n .
n≥0

Our goal in this section is to prove an ∞-categorical analogue of this statement.

Definition 1.4.1. Let C be a monoidal ∞-category, and let θ : Alg(C) → C be the forgetful functor.
Let C ∈ C be an object. A free algebra generated by C is an object A ∈ Alg(C) equipped with a map
φ : C → θ(A), such that composition with φ induces a homotopy equivalence

MapAlg(C) (A, B) → MapC (C, θ(B))

for every B ∈ Alg(C).

The main result of this section is the following:
Theorem 1.4.2. Let C be a monoidal ∞-category. Assume that C admits countable coproducts, and that
the bifunctor ⊗ : C × C → C preserves countable coproducts separately in each variable.

(1) The forgetful functor Alg(C) → C admits a left adjoint.

(2) For every object C ∈ C, there exists a map C → A([1]) which exhibits A ∈ Alg(C) as the free algebra
generated by C.
(3) For every object C ∈ C and every algebra object A ∈ Alg(C), ` a map C → A([1]) exhibits A as a free
algebra generated by C in C if and only if the induced map n≥0 C ⊗n → A([1]) is an equivalence in C
(that is, if and only if A is a free algebra generated by C in the homotopy category hC).

The proof of Theorem 1.4.2 is long and complicated, and will occupy our attention throughout most of
this section. Before embarking on the proof, we will treat a special case which can be established using much
easier arguments (and which requires fewer hypotheses):
Proposition 1.4.3. Let C be a monoidal ∞-category. Then:

21
 (1) The ∞-category Alg(C) has an initial object.

(2) An object A ∈ Alg(C) is initial if and only if the unit map 1C → A([1]) is an equivalence in C.
Proof. Let q : C⊗ → N(∆)op be a monoidal structure on C = C⊗ ⊗
[1] . The fiber C[0] is contractible, so it contains
an object C[0] . Using Lemma T.[Link] we deduce that there exists a map A : N(∆)op → C⊗ which is a
section of q, where A([0]) = C[0] and A is a q-left Kan extension of A|{[0]}. It is easy to see that A carries
each morphism in ∆ to a q-coCartesian edge of C⊗ ; in particular we conclude that A ∈ Alg(C). Using
Proposition T.[Link], we deduce that A is an initial object of Alg(C) (in fact, it is an initial object in the
larger ∞-category of all sections of q). This proves (1).
We observe that the unit map 1C → A([1]) is an equivalence. The “only if” direction of (2) now follows
from the fact that initial objects of Alg(C) are unique up to equivalence. To prove the converse, we consider
an arbitrary algebra object B ∈ Alg(C). Since A is an initial object of Alg(C), there exists a map f : A → B
which induces a commutative triangle

1C E
yy EE
yy EEuB
y
y uA EE
|yy E"
f ([1])
A([1]) / B([1])

in C. Here uA and uB denote the unit maps of A and B, respectively. If uB is an equivalence, then we
conclude that f ([1]) is an equivalence, so that f is itself an equivalence (Corollary 1.5.4). This proves the
“if” direction of (2).
We now return to the proof of Theorem 1.4.2. The argument is long and technical, and can be safely
skipped; the details will not be important elsewhere in this paper. We begin with an outline of our strategy:

(a) Using the results of §1.7, we can reduce to proving an analogous assertion for Segal monoidal ∞-
categories. Let us therefore assume that we are given a map q : C⊗ → N(Lin∗ ) that exhibits C = C⊗ h1i
as a Segal monoidal ∞-category (we will also assume that the reader is familiar with the notation and
terminology from §1.7).
(b) We will define a sequence of categories J0 ⊆ J1 ⊆ J2 ⊆ J3 equipped with (compatible) functors
ψi : Ji → Lin∗ . This will allow us to factor the forgetful functor θ : Algs (C) → C as a composition

Algs (C) → MapN(Lin∗ ) (N(J3 ), C⊗ )

→ MapN(Lin∗ ) (N(J2 ), C⊗ )
→ MapN(Lin∗ ) (N(J1 ), C⊗ )
→ MapN(Lin∗ ) (N(J0 ), C⊗ )
→ C.

(c) We will refine the above picture by selecting appropriate subcategories Ei (C) ⊆ MapN(Lin∗ ) (N(Ji ), C⊗ )
so that the forgetful functor θ factors as a composition
θ θ θ θ θ
4
Alg(C) → E3 (C) →
3
E2 (C) →
2
E1 (C) →
1
E0 (C) →
0
C.

(d) We will show that the functor θ0 is a trivial Kan fibration (Lemma 1.4.7).
(e) We will show that the restriction functor θ1 admits a left adjoint φ1 , given by q-left Kan extension
along the inclusion N(J0 ) ⊆ N(J1 ).

22
 (f ) We will show that the restriction functor θ2 admits a left adjoint φ2 . This is a slightly tricky part
of the argument, since the left adjoint to θ2 is not given by a q-left Kan extension (the relevant Kan
extension typically does not exist).
(g) The functor θ3 does not generally admit a left adjoint. Nevertheless, we will prove that there exists a
partially-defined left adjoint to θ3 , whose domain includes the essential image of φ2 ◦ φ1 . This adjoint
is again given by a q-left Kan extension along the inclusion N(J2 ) ⊆ N(J3 ).

(h) Finally, it will follow immediately from the construction that the functor θ4 has a left adjoint, given
by composition with a section to the functor ψ3 : J3 → Lin∗ .

The definitions of the categories Ji become more complicated as the index i increases. We therefore begin
with the simplest case, where i = 0.
Notation 1.4.4. The category J0 is defined to be the following subcategory of Lin∗ :
(J0) Every object of Lin∗ belongs to J0 .

(J00 ) A morphism α : J∗ → J∗0 of Lin∗ belongs to J0 if and only if every element of J 0 has exactly one
preimage under α.
We let ψ0 : J0 → Lin∗ denote the inclusion.
Notation 1.4.5. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category. We let E0 (C) denote the full
subcategory of MapN(Lin∗ ) (N(J0 ), C⊗ ) spanned by those functors f ∈ MapN(Lin∗ ) (N(J0 ), C⊗ ) which carry
every morphism in J0 to a q-coCartesian morphism in C⊗ .
Example 1.4.6. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category and let A be a section of q. We
observe that A is a Segal algebra object of C if and only if A| N(J0 ) ∈ E0 (C). In particular, restriction
determines a functor Algs (C) → E0 (C).
More informally, a functor f : N(J0 ) → C⊗ belongs to E0 (C) if there exists an object C = f (h1i) ∈ C, such
that for all n ≥ 0, the object f (hni) ∈ C⊗ ⊗
hni corresponds to (C, C, . . . , C) under the equivalence Chni ' C .
n

The functor F is then determined by C, up to canonical equivalence. Our next result makes this statement
more precise:
Lemma 1.4.7. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal structure on an ∞-category C = C⊗
h1i . Then
evaluation at h1i ∈ J0 induces a trivial Kan fibration θ0 : E0 (C) → C.
Proof. Fix n ≥ 0, and regard hni as an object of J0 . Let J−1 denote the full subcategory of J0 spanned by
the object h1i. We observe that a functor f ∈ MapN(Lin∗ ) ((N(J−1 )hni/ )/ , C⊗ ) is a q-limit diagram if and
only if it carries each morphism of ((N(J−1 )op )hni/ )/ to a q-coCartesian morphism in C⊗ . Using Lemma
T.[Link], we deduce that every functor g0 ∈ MapN(Lin∗ ) (N(J−1 ), C⊗ ) admits a q-right Kan extension to
a functor g ∈ MapN(Lin∗ ) (N(J0 ), C⊗ ). Let D ⊆ MapN(Lin∗ ) (N(J0 ), C⊗ ) be the full subcategory spanned by
those functors g which are q-right Kan extensions of g| N(J−1 ). Proposition T.[Link] implies that restriction
determines a trivial Kan fibration D → MapN(Lin∗ ) (N(J−1 ), C⊗ ) ' C . To complete the proof, it will suffice
to show that D = E0 (C). Using the description of the q-limit diagrams given above, we deduce that a functor
g ∈ MapN(Lin∗ ) (N(J0 ), C⊗ ) belongs to D if and only if the following condition is satisfied:
(∗) For every morphism α : hni → h1i in J0 , the morphism g(α) is q-coCartesian.
From this characterization it follows immediately that E0 (C) ⊆ D. To prove the reverse inclusion, let us
suppose that g ∈ MapN(Lin∗ ) (N(J0 ), C⊗ ) satisfies (∗). We wish to show that g is a coCartesian section of q.

23
Let α : hni → hmi be a morphism in J0 , let α = g(α), and let αe : g(hni) → C be a q-coCartesian morphism
of C⊗ which projects to α. There exists an (essentially unique) 2-simplex

< C FF
yyy FF βe
FF
α
y
e
yy FF
y y F"
g(hni)
α / g(hmi)

in C⊗ . We wish to show that β is an equivalence. For this, it will suffice to show that, for every morphism
γ : hmi → h1i in J0 , the image of βe under the associated functor γ! : C⊗ ⊗
hmi → Ch1i is an equivalence. We have
a commutative diagram

γ
e
/ γ! C
< C EE II
e zzz
zz EE βe II γ! βe
α EE II
zz EE II
zz E" I$
g(hni)
α / g(hmi) γ
/ γ! g(hmi).

Consequently, it will suffice to show that the morphisms γ ◦ α and γ e◦α e are q-coCartesian. In the second
case, this follows from the fact that q-coCartesian morphisms are stable under composition. In the first case,
we can use (∗) to assume, without loss of generality, that γ = g(γ). Then γ ◦ α ' g(γ ◦ α) is q-coCartesian
(invoking (∗) once more), as desired.
Notation 1.4.8. The category J1 is defined as follows:

(J1) Objects of J1 are given by morphisms α : hni → hmi in Lin∗ .

(J10 ) Given a pair of objects α, α0 ∈ J1 , a morphism from α to α0 is a commutative diagram

hni
α / hmi

γ
 
α0 / hm0 i
hn0 i

in Lin∗ , such that γ is a morphism in J0 .

We will identify J0 with the full subcategory of J1 spanned by those morphisms α which are isomorphisms
in Lin∗ . The inclusion ψ0 : J0 → Lin∗ extends to a functor ψ1 : J1 → Lin∗ , given by the formula

(α : hni → hmi) 7→ hmi.

Notation 1.4.9. Let C be a Segal monoidal ∞-category. We define E1 (C) ⊆ MapN(Lin∗ ) (N(J1 ), C⊗ ) to be
the full subcategory spanned by those functors f such that f | N(J0 ) belongs to E0 (C).
Remark 1.4.10. Let C be a monoidal ∞-category. Then composition with ψ1 : J1 → Lin∗ induces a functor
Algs (C) → E1 (C).
We now analyze the relationship between E0 (C) and E1 (C).
Lemma 1.4.11. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category. Then:
(1) A functor f ∈ MapN(Lin∗ ) (N(J1 ), C⊗ ) is a q-left Kan extension of f0 = f | N(J0 ) if and only if, for
every object α : hni → hmi in J1 , if eα : id[n] → α denotes the associated morphism of J1 , then f (eα )
is a q-coCartesian morphism in C⊗ .

24
 (2) Every functor f0 ∈ MapN(Lin∗ ) (N(J0 ), C⊗ ) admits an extension f ∈ MapN(Lin∗ ) (N(J1 ), C⊗ ) which is a
q-left Kan extension of f .
(3) Let Map0N(Lin∗ ) (N(J1 ), C⊗ ) denote the full subcategory of MapN(Lin∗ ) (N(J1 ), C⊗ ) spanned by those func-
tors f which are q-left Kan extensions of f0 = f | N(J0 ). Then the restriction map

Map0N(Lin∗ ) (N(J1 ), C⊗ ) → MapN(Lin∗ ) (N(J0 ), C⊗ )

is a trivial Kan fibration.

(4) The restriction map E1 (C) → E0 (C) admits a section φ1 which is simultaneously a left adjoint. The
essential image of φ1 consists of those functors f ∈ E1 (C) which satisfy the condition stated in (1).
Proof. We observe that if α ∈ J1 , then the morphism eα is a final object of the category (J1 )/α ×J1 J0 .
Assertion (1) now follows immediately from Proposition T.[Link], and assertion (2) follows from Lemma
T.[Link] and Proposition T.[Link], since q is a coCartesian fibration. Assertion (3) follows from Proposition
T.[Link], and (4) from Proposition T.[Link].
Notation 1.4.12. The category J2 is defined as follows.
(J2) An object of J2 consists of a quadruple (J, J0 , hni, α), where J is a finite linearly ordered set, J0 is a
subset of J, and α : hni → (J0 )∗ is a morphism in Lin∗ .
(J20 ) Given a pair of objects (J, J0 , hni, α), (J 0 , J00 , hn0 i, α0 ) ∈ J2 , a morphism in J2 from (J, J0 , hni, α) to
(J 0 , J00 , hn0 i, α0 ) consists of a pair of morphisms β : J∗ → J∗0 , γ : hni → hn0 i in Lin∗ , satisfying the
following conditions:

– The morphism γ belongs to J0 .

– Suppose j ∈ J 0 and β −1 {j} ⊆ J0 . Then j ∈ J00 , and γ induces a bijection (α ◦ β)−1 {j} →
(α0 )−1 {j}.

We will identify J1 with the full subcategory of J2 spanned by those objects (J, J0 , hni, α) such that
J0 = J. We note that the functor ψ1 : J1 → Lin∗ admits an extension ψ2 : J2 → Lin∗ , given by the formula

(J, J0 , hni, α) 7→ J∗ .

Notation 1.4.13. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category. We let E2 (C) denote the full
subcategory of MapN(Lin∗ ) (N(J2 ), C⊗ ) spanned by those functors f which possess the following properties:
(i) The restriction f0 = f | N(J0 ) belongs to E0 (C).
(ii) Let (J, J0 , hni, α) be an object of J2 , and let

(β, id) : (J, J0 , hni, α) → (J0 , J0 , hni, α)

be given by the formula (

j if j ∈ J0
β(j) =
∗ if j ∈
/ J0 .

Then f (β, id) is a q-coCartesian morphism in C⊗ .

Remark 1.4.14. Composition with ψ2 : J2 → Lin∗ induces a functor Algs (C) → E2 (C).
Lemma 1.4.15. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category. Assume that the underlying ∞-
category C has an initial object, and a tensor product C1 ⊗ . . . ⊗ Ck in C is initial whenever any of the factors
Ci is initial. Then:

25
 (1) The restriction functor θ2 : E2 (C) → E1 (C) admits a section φ2 , which is simultaneously a left adjoint
to θ2 .
(2) A functor f ∈ E2 (C) belongs to the essential image of φ2 if and only if the following condition is
satisfied:
(∗) For every object (J, J0 , hni, α) ∈ J2 and every element j ∈ J − J0 , if β : J∗ → {j} ∪ {∗} is the
morphism of Lin∗ such that β −1 {j} = {j}, then the induced functor β! : C⊗ ⊗
J∗ → C{j}∗ = C carries
f (J, J0 , hni, α) to an initial object of C.
Proof. We observe that the inclusion J1 ⊆ J2 has a left adjoint U , given by

(J, J0 , hni, α) 7→ (J0 , J0 , hni, α).

Let T : ∆1 × N(J2 ) → N(J2 ) be a natural transformation from id to U which is a unit for the adjunction.
Let T 0 be the composition of T with ψ2 , regarded as a functor from N(J2 ) to Fun(∆1 , N(Lin∗ )).
Let D denote the full subcategory of Fun(∆1 , C⊗ ) spanned by the q-coCartesian morphisms, and let M be
the fiber product N(J2 ) ×Fun(∆1 ,N(Lin∗ )) D . Similarly, we let M0 denote the fiber product N(J2 ) ×N(Lin∗ ) C⊗
where N(J2 ) maps to N(Lin∗ ) via ψ2 , and M1 the fiber product N(J2 ) ×N(Lin∗ ) C⊗ where N(J2 ) maps to
e0 e1
N(Lin∗ ) via ψ1 ◦ U . We have canonical maps M0 ← M→ M1 , where e0 is a trivial Kan fibration.
Let E2 (C) ⊆ MapN(J2 ) (N(J2 ), M) denote the inverse image of E2 (C) ⊆ MapN(J2 ) (N(J2 ), M0 ), so that e0
induces a trivial Kan fibration E2 (C). Since the restriction of T 0 to N(J1 ) is an equivalence, we have a
canonical map E2 (C) × ∆1 → E1 (C) which induces an equivalence from the composition

e ◦ | N(J1 )
E2 (C) →
0
E2 (C) → E1 (C)

to the composition
e ◦
E2 (C) →
1
MapN(J2 ) (N(J2 ), M1 ) → MapN(Lin∗ ) (N(J1 ), C⊗ ).
Unwinding the definitions, we see that there is a pullback diagram
 / MapN(J ) (N(J2 ), M)
E2 (C) 2

g
  
Map0N(J2 ) (N(J2 ), M1 )  / MapN(J ) (N(J2 ), M1 ),
2

where Map0N(J2 ) (N(J2 ), M1 ) denotes the full subcategory of MapN(J2 ) (N(J2 ), M1 ) spanned by those functors f
for which f0 = f | N(J1 ) belongs to E1 (C), and f is a q-right Kan extension of f0 . Using Proposition T.[Link],
we deduce that the projection Map0N(J2 ) (N(J2 ), M1 ) → E0 (C) is a trivial Kan fibration. Consequently, we
may reduce to proving the following analogues of statements (1) and (2) for the map g:

(10 ) The projection g : MapN(J2 ) (N(J2 ), M) → MapN(J2 ) (N(J2 , M1 ) admits a section s, which is simultane-
ously a left adjoint to g.
(20 ) A functor F ∈ MapN(J2 ) (N(J2 ), M) belongs to the essential image of s if and only if the following
condition is satisfied:
(∗0 ) For every object (J, J0 , hni, α) ∈ J2 and every element j ∈ J − J0 , if αj,J : J∗ → {j}∗ is defined
as in Notation 1.7.5, then the induced functor α!j,J : C⊗ ⊗
J∗ → C{j}∗ = C carries F (J, J0 , hni, α) to
an initial object of C.

26
 We observe that the projection e1 : M → M1 is a coCartesian fibration. Moreover, each fiber of e1 is
equivalent to a product of copies of C, and therefore admits an initial object. Let M0 denote the collection
of objects M ∈ M such that M is an initial object of the fiber M ×M1 {e1 (M )}. We observe that (20 ) is
equivalent to the assertion that the essential image of s is the ∞-category MapN(J2 ) (N(J2 ), M0 ). Using our
assumption that a tensor product C ⊗D ∈ C is initial provided that either C or D is initial, and the definition
of the category J2 , we deduce that the collection of objects of M0 is stable under the collection of functors
Mx → My associated to morphisms x → y in M1 . Proposition T.[Link] implies that every object of M0 is
an e1 -initial object of M. It now follows from Proposition T.[Link] that g restricts to a trivial Kan fibration

g 0 : MapN(J2 ) (N(J2 ), M0 ) → MapN(J2 ) (N(J2 , M1 ).

Let s be a section to g 0 . To complete the proof, it will suffice to show that s is a left adjoint to g. In other
words, it suffices to show that if X ∈ MapN(J2 ) (N(J2 ), M1 ) and Y ∈ MapN(J2 ) (N(J2 ), M), then g induces a
homotopy equivalence of Kan complexes

HomR
MapN(J (N(J2 ),M) (s(X), Y ) → HomR
MapN(J (N(J2 ),M1 ) (X, g(Y )).
2) 2)

But this map is a trivial Kan fibration, since s(X) is a g-initial object of MapN(J2 ) (N(J2 ), M) by Lemma
2.3.1.

Notation 1.4.16. The category J3 is defined to be the categorical mapping cylinder of the functor ψ2 :
J2 → Lin∗ . More precisely, this category may be described as follows:
(J3) An object of J3 is either an object of Lin∗ or an object of J2 .
(J30 ) Given a pair of objects hmi, hni ∈ Lin∗ , we have HomJ3 (hmi, hni) = HomLin∗ (hmi, hni).

(J300 ) Given a pair of objects X, Y ∈ J2 , we have HomJ3 (X, Y ) = HomJ2 (X, Y ).

(J3000 ) Given a pair of objects X = (J, J0 , hmi, α) ∈ J2 , hni ∈ Lin∗ , we have

HomJ3 (X, hni) = HomLin∗ (J∗ , hni) HomJ3 (hni, X) = ∅.

By construction, we may regard J2 as a full subcategory of J3 . The functor ψ2 : J2 → Lin∗ extends

canonically to a retraction ψ3 : J3 → Lin∗ .
Notation 1.4.17. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category. We let E02 (C) ⊆ E2 (C) denote
the essential image of the functor φ2 ◦ φ1 : E0 (C) → E2 (C). In other words, E02 (C) is the full subcategory of
MapN(Lin∗ ) (N(J2 ), C⊗ ) spanned by those functors f ∈ N(J2 ) → C⊗ which satisfy the following conditions:

(A0) The composition q ◦ f coincides with the map N(J2 ) → N(Lin∗ ) induced by ψ2 .

(A1) For every morphism α : J∗ → J∗0 in J0 , f carries the associated map

(α, α) : (J, J, J∗ , idJ∗ ) → (J 0 , J 0 , J∗0 , idJ∗0 )

to a q-coCartesian morphism in C⊗ (Notation 1.4.5).

(A2) Let (J, J0 , hni, α) be an object of J2 , and let (β, id) : (J, J0 , hni, α) → (J0 , J0 , hni, α) be given by the
formula (
j if j ∈ J0
β(j) =
∗ if j ∈/ J0 .

Then f (β, id) is a q-coCartesian morphism in C⊗ (Notation 1.4.13).

27
(A3) For every object (J, J0 , hni, α) ∈ J2 and every element j ∈ J − J0 , if αj,J : J∗ → {j}∗ is defined as
in Notation 1.7.5, then the induced functor β! : C⊗ ⊗
J∗ → C{j}∗ = C carries f (J, J0 , hni, α) to an initial
object of C (Lemma 1.4.15).
(A4) For every morphism α : J∗ → J∗0 in Lin∗ , f carries the associated map

(α, idJ∗ ) : (J, J, J∗ , idJ∗ ) → (J 0 , J 0 , J∗ , α)

to a q-coCartesian morphism in C⊗ (Lemma 1.4.11).

Lemma 1.4.18. Let A and B be marked simpicial sets (see §T.3.1), and let j : A → B be a weak equivalence.
Let C be an ∞-category, and p : B → C\ a map of marked simplicial sets. Then the induced map q : C/p →
C/pj is a categorical equivalence.

Proof. Choose a ` marked anodyne map A → A0 , where A0 is a fibrant object of Set+

∆ . Now choose a marked
anodyne map B A A0 → B 0 , where B 0 is fibrant. Let j 0 : A0 → B 0 denote the induced map; since (Set+
∆ ) is
left proper, j 0 is a marked equivalence. The inclusion B → B 0 is marked anodyne, so the map p extends to
a map p0 : B 0 → C\ . We have a commutative diagram

q0
C/p0 / C/p0 j 0

 
C/p
q
/ C/pj .

Proposition T.[Link] implies that the vertical maps are trivial Kan fibrations. It will therefore suffice to
show that q 0 is an equivalence of ∞-categories. We now observe that Theorem T.[Link] implies that j 0 is a
categorical equivalence of the underlying simplicial sets, so that q 0 is a categorical equivalence by Proposition
T.[Link].
Lemma 1.4.19. Let p : D → C be a functor between ∞-categories. Let D0 ⊆ D be the essential image of a
localization functor L : D → D, and suppose that for every D ∈ D, p carries the localization map D → LD to
an equivalence in C. Let p0 = p| D0 . Then the induced map q : C/p → C/p0 is a trivial fibration of simplicial
sets.
Proof. Since the map q is a right fibration, it will suffice to show that q is a categorical equivalence. Let
α : idD → L be a unit for the adjunction between L and the inclusion D0 ⊆ D. Let B be the marked
simplicial set (D, S), where S is the collection of morphisms f of D such that L(f ) is an equivalence, and let
A = D\0 ⊆ B. Then α determines a homotopy (∆1 )] × B → B from idB to the map L : B → A, and restricts
to a homotopy from idA to the induced map L|A : A → A. It follows that L is a homotopy inverse to the
inclusion j : A ⊆ B, so that j is a weak equivalence. The desired result now follows from Lemma 1.4.18.
Lemma 1.4.20. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category. Assume that the underlying ∞-
category C = C⊗h1i admits countable coproducts, and that the bifunctor ⊗ : C × C → C preserves countable
coproducts separately in each variable. Let f0 ∈ E02 (C). Then:
(1) There exists a commutative diagram

N(J 2 ) / C⊗
f0
_ 9
ttt
tt f
t
tt
q
 tt ψ3 
N(J3 ) / N(Lin∗ ),

where f is a q-left Kan extension of f0 .

28
 (2) The restriction A = f | N(Lin∗ ) is a Segal algebra object of C.
◦ ◦
(3) Let C = f0 (h1i , h1i , h1i, idh1i ) ∈ C. Then the canonical map C → A(h1i) and the algebra structure on
A determine an equivalence a
C ⊗n → A(h1i).
n≥0

Remark 1.4.21. The hypotheses of Lemma 1.4.20 have the following consequence, which we will use
repeatedly: given a collection of objects C1 , . . . , Ck ∈ C, if any Ci is an initial object of C, then the tensor
product C1 ⊗ . . . ⊗ Ck is initial.
Proof. We begin by formulating a more precise version of (1), which will be used to establish (2) and (3).
First, we need to introduce a bit of notation. Let J be a finite linearly ordered set, and let j be an element
of J. Let αj,J : J∗ → {j}∗ be defined as in Notation 1.7.5, and let α!j,J : C⊗ ⊗
J∗ → C{j}∗ = C denote the
associated functor. For each n ≥ 0, we let ej,J n denote the object of J2 given by (J, {j}, hni, β), where the
map β : hni → {j}∗ satisfies β −1 (∗) = ∗. We note that ψ2 (ej,J j,J
n ) = J∗ , so there is a canonical map en → J∗
in J3 . In addition to (1), we will need the following characterization of the left Kan extensions of f0 :
(10 ) Suppose that f ∈ MapN(Lin∗ ) (N(J3 ), C⊗ ) is such that f0 = f | N(J2 ). Then f is a q-left Kan extension
of f0 if and only if, for every finite linearly ordered set J and every element j, f exhibits α!j,J f (J∗ ) as
a coproduct of the objects α!j,J f (ej,J
n ) in C. Here n ranges over all nonnegative integers.

Fix an object J∗ ∈ Lin∗ , let I denote the category J2 ×Lin∗ (Lin∗ )/J∗ , and let f1 denote the composition
f0
N(I) → N(J2 ) → C⊗ .

According to Lemma T.[Link], assertion (1) is equivalent to the assertion that, for any choice of J∗ , the
lifting problem
f1
/ ⊗
t: C
N(I)
_
f1 t
t q
t
 t 
N(I). / N(Lin∗ )

admits a solution f 1 which is a q-colimit diagram. Choose a q-coCartesian transformation f1 → f2 covering

the canonical natural transformation from q ◦ f1 to the constant functor N(I) → N(Lin∗ ) taking the value
J∗ . Proposition T.[Link] shows that we can identify q-colimits of f1 with q-colimits of f2 . Combining this
observation with Proposition T.[Link], we are reduced to proving the following pair of assertions:

(a) The diagram f2 : N(I) → C⊗ ⊗

J∗ admits a colimit f 2 : N(I) → CJ∗ .
.

(b) Let δ : J∗ → J∗0 be a morphism in Lin∗ and δ! : C⊗ ⊗

J∗ → CJ∗0 the associated functor. Then δ! ◦ f 2 :
⊗
N(I). → CJ∗0 is a colimit diagram.

Moreover, (10 ) translates into the following condition:

(c) Let f 2 : N(I). → C⊗

J∗ be an arbitrary extension of f2 . Then f 2 is a colimit diagram if and only if, for
each j ∈ J, α! ◦ f 2 exhibits the image of the cone point as a coproduct of the objects (α!j,J ◦ f2 )(ej,J
j,J
n )
in C, where n ranges over all nonnegative integers.

We will begin with the proofs of (a) and (c), and return later to the proof of (b). Using the fact that
q : C⊗ → N(Lin∗ ) is a Segal monoidal ∞-category, we see that (a) and (c) can be reformulated as follows:

(a0 ) For each j ∈ J, the map f3 = α!j,J ◦ f2 admits a colimit in C.

29
 (c0 ) An arbitrary map f 3 : N(I). → C extending f3 is a colimit diagram if and only if it exhibits the image
n ) ∈ C; here n ranges over all nonnegative integers.
of the cone point as a coproduct of the objects f3 (ej,J

Let us now fix j ∈ J and prove the assertions (a0 ) and (c0 ). An object of I can be identified with a quintuple
(I, I0 , hni, α, α0 ), where I0 ⊆ I are linearly ordered sets, hni is an object of Lin∗ , and α : hni → (I0 )∗ and
α0 : I∗ → J∗ are morphisms in Lin∗ . Let I0 denote the full subcategory of I spanned by those objects for
which α0 is an isomorphism. We observe that the inclusion I0 ⊆ I has a left adjoint, given by

(I, I0 , hni, α, α0 ) 7→ (J, J0 , hni, β ◦ α, id),

where J0 = {j ∈ J : (α0 )−1 {j} ⊆ I0 } and β : (I0 )∗ → (J0 )∗ has the property that β −1 {j} = (α0 )−1 {j} for all
j ∈ J0 . It follows that the inclusion N(I0 ) ⊆ N(I) is cofinal. We therefore obtain the following reformulations
of (a0 ) and (c0 ):

(a00 ) The map f4 = f3 | N(I0 ) admits a colimit in C.

(c00 ) An arbitrary extension f 4 : N(I0 ). → C of f4 is a colimit diagram if and only if it exhibits the image of
n ) ∈ C; here n ranges over all nonnegative integers.
the cone point as a coproduct of the objects f4 (ej,J

We may identify objects of I0 with triples (J0 , hni, α), where J0 is a subset of J, hni is an object of Lin∗ ,
and α : hni → (J0 )∗ is a morphism in Lin∗ . Let I1 be the full subcategory of I0 spanned by those objects for
which j ∈ J0 . Using condition (A3) of Notation 1.4.17, we conclude that if X ∈ I0 and X ∈ / I1 , then f4 (X)
is an initial object of C. It follows that f4 is a left Kan extension of f5 = f4 | N(I1 ). Lemma T.[Link] shows
that (a00 ) and (c00 ) are equivalent to the following conditions:

(a000 ) The map f5 admits a colimit in C.

(c000 ) An arbitrary extension f 5 : N(I1 ). → C of f5 is a colimit diagram if and only if it exhibits the image of
n ) ∈ C; here n ranges over all nonnegative integers.
the cone point as a coproduct of the objects f5 (ej,J

Let I2 denote the full subcategory of I1 spanned by those objects (J0 , hni, α) where J0 = {j}. We observe
that the inclusion I2 ⊆ I1 has a right adjoint, given by

(J0 , hni, α) 7→ ({j}, hni, αj,J0 ◦ α).

Using condition (A2) of Notation 1.4.17, we observe that f5 carries each counit map ({j}, hni, αj,J0 ◦ α) →
(J0 , hni, α) to an equivalence in C. Applying (the dual of) Lemma 1.4.19, we obtain the following reformu-
lations of (a000 ) and (c000 ):

(a0000 ) The map f6 = f5 | N(I2 ) admits a colimit in C.

(c0000 ) An arbitrary extension f 6 : N(I2 ). → C of f6 is a colimit diagram if and only if it exhibits the image of
n ) ∈ C; here n ranges over all nonnegative integers.
the cone point as a coproduct of the objects f6 (ej,J

Let I3 denote the full subcategory of I2 spanned by triples of the form ({j}, hni, α), where α : hni → {j}
has the property that α−1 {∗} = {∗}. The inclusion I3 ⊆ I2 has a left adjoint, so the induced inclusion
N(I3 ) ⊆ N(I2 ) is cofinal. We therefore obtain the following reformulations of (a0000 ) and (c0000 ):

(a00000 ) The map f7 = f6 | N(I3 ) admits a colimit in C.

(c00000 ) An arbitrary extension f 7 : N(I3 ). → C of f7 is a colimit diagram if and only if it exhibits the image of
n ) ∈ C; here n ranges over all nonnegative integers.
the cone point as a coproduct of the objects f7 (ej,J

30
 We now observe that the category I3 is discrete, and its objects can be identified with nonnegative integers
00000
n via the bijection n 7→ (ej,J
n → J∗ ). Consequently, assertion (c ) is a tautology, and (a00000 ) follows from
our assumption that the ∞-category C admits countable coproducts. This completes the proofs of (a) and
(c).
We now return to the proof of (b). Using the fact that q : C⊗ → N(Lin∗ ) is a Segal monoidal ∞-category,
we see that (b) is equivalent to the following slightly weaker statement:

(b0 ) Let β : J∗ → h1i be a morphism in Lin∗ and β! : C⊗ ⊗

J∗ → Ch1i = C the associated functor. Then
g 3 = δ! ◦ f 2 : N(I). → C is a colimit diagram.

Noting once again that the inclusion N(I0 ) ⊆ N(I) is cofinal, we can reformulate (b0 ) as follows:

(b00 ) The map g 4 = g 3 | N(I0 ). → C is a colimit of g4 = g 4 | N(I0 ).

The morphism β : J∗ → h1i is uniquely determined by the the subset J1 = J − β −1 {∗} ⊆ J. Let I01
denote the full subcategory of I0 spanned by those objects (J0 , hni, α) such that J1 ⊆ J0 . Using Condition
(A3) of Notation 1.4.17 and Remark 1.4.21, we deduce that if (J0 , hni, α) ∈ I0 does not belong to I01 , then
g4 (J0 , hni, α) is an initial object of C. It follows that g4 is a left Kan extension of g5 = g4 | N(I01 ). Invoking
Lemma T.[Link], we see that (b00 ) is equivalent to:

(b000 ) The map g 5 = g 4 | N(I01 ). → C is a colimit diagram.

Let I02 denote the full subcategory of I01 spanned by those objects (J0 , hni, α) for which J0 = J1 . The
inclusion I02 ⊆ I01 admits a right adjoint V . Moreover, condition (A3) of Notation 1.4.17 guarantees that
g5 carries each of the unit maps V (J1 , hni, α) → (J0 , hni, α) to an equivalence in C. Applying (the dual of)
Lemma 1.4.19, we deduce that (b000 ) is equivalent to:

(b0000 ) The map g 6 = g 5 | N(I02 ). → C is a colimit diagram.

Finally, let I03 denote the full subcategory of I02 spanned by those objects (J1 , hni, α) for which α−1 {∗} =
{∗}. The inclusion I03 ⊆ I02 admits a left adjoint, so the induced inclusion N(I03 ) ⊆ N(I02 ) is cofinal. We
therefore obtain the following final reformulation of (b0000 ):

(b00000 ) The map g 7 = g 6 | N(I03 ). → C is a colimit diagram.

We now observe that the category I03 is discrete. If we let J1 = {j0 , . . . , jk }, then the objects of I03 are in
bijection with finite sequences of nonnegative integers (n0 , . . . , nk ). The bijection is given by the formula

(n0 , . . . , nk ) 7→ (J1 , {n0 + . . . + nk }, α),

where α−1 {ji } = {m|n0 + . . . + ni−1 ≤ m < n0 + . . . + ni }. Using conditions (A1) and (A4) of Notation
1.4.17, we conclude that g 7 (n0 , . . . , nk ) is canonically isomorphic to the tensor product

f1 (ejn00,J ) ⊗ . . . ⊗ f1 (ejnkk,J )

in the homotopy category hC. On the other hand, the same argument and condition (c) imply that the
image of the cone point under g 7 is equivalent to the tensor product
a a
( f1 (ejn00,J )) ⊗ . . . ⊗ ( f1 (ejnkk,J )).
n0 ≥0 nk ≥0

Condition (b00000 ) now follows from our assumption that the tensor product on C preserves countable coprod-
ucts in each variable. This completes our proof of (b), and therefore also our proofs of (1) and (10 ).

31
 For the remainder of the proof, we will assume that f ∈ MapN(Lin∗ ) (N(J3 ), C⊗ ) is a q-left Kan extension
of f0 . Let us prove (2). We wish to show that, if α : J∗ → J∗0 is a morphism in Lin∗ which induces a bijection
from α−1 J 0 onto J 0 , then f (α) is a q-coCartesian morphism in C⊗ . Let j 0 be an element of J 0 , and j ∈ J its
preimage under α. In view of our assumption that q : C⊗ → N(Lin∗ ) is a Segal monoidal ∞-category, it will
0 0
suffice to show that α induces (for every choice of j 0 ∈ J 0 ) an equivalence α!j,J f (J∗ ) → α!j ,J f (J∗0 ). Invoking
(1), we see that f induces equivalences
j 0 ,J 0
a j,J a j 0 ,J 0 0 0
α!j,J f (J∗ ) ' α! (f0 (ej,J
n )) α! f (J∗0 ) ' α! (f0 (ejn ,J )).
n≥0 n≥0

0 0 0 0
It will therefore suffice to show that each of the maps α!j,J f0 (ej,J j ,J
n ) → α! f0 (ejn ,J ) is an equivalence; this
follows from (A2) of Notation 1.4.17.
We now prove (3). Invoking (10 ), we deduce that f exhibits A(h1i) as a coproduct of the objects
◦ ◦
f0 (h1i , h1i , hni, α(n)), where α(n) denotes the unique map hni → h1i such that α−1 {∗} = {∗}. The
functor f induces a commutative diagram
◦ ◦
f0 (hni , hni , hni, id) / f0 (h1i◦ , h1i◦ , hni, α(n))

 
A(hni) / A(h1i)

in C∗ . Condition (A1) of Notation 1.4.17 guarantees that the left vertical map corresponds, under the
equivalence C⊗ ⊗ ⊗
hni ' C , to the nth power of the map C → A(h1i). Let α(n)! : Chni → Ch1i = C be the functor
n

induced by α(n). Then the above diagram induces a commutative square

◦ ◦
α(n)! f0 (hni , hni , hni, id)
u / f0 (h1i◦ , h1i◦ , h1i, α(n))

 
α(n)! A(hni) / A(h1i)

in C. Condition (A4) of Notation 1.4.17 guarantees that the map u is an equivalence. Consequently, the
◦ ◦
map f0 (h1i , h1i , hni, α(n)) → A(h1i) can be identified with the composition

C ⊗n → A(h1i)⊗n ' α(n)! A(hni) → A(h1i),

so that (3) follows from (10 ).

Notation 1.4.22. Let C be a monoidal ∞-category. We define E3 (C) to be the full subcategory of

MapN(Lin∗ ) (N(J3 ), C⊗ )

spanned by those functors f ∈ MapN(Lin∗ ) (N(J3 ), C⊗ ) with the following properties:

(i) The restriction f | N(J2 ) belongs to E2 (C).
(ii) The restriction f | N(Lin∗ ) belongs to Algs (C).
Lemma 1.4.23. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category. Then:
(1) Composition with ψ3 : J3 → Lin∗ and the inclusion Lin∗ → J3 defines a pair of adjoint functors

φ04
MapN(Lin∗ ) (N(J3 ), C⊗ ) o / Map ⊗
N(Lin∗ ) (N(Lin∗ ), C ) .
θ40

32
 (2) The functors φ04 and θ40 restrict to a pair of adjoint functors E3 (C) o
φ4
/ Algs (C) .
θ4

Proof. In view of Proposition T.[Link], to prove (1) it will suffice to show that θ40 is a q-right Kan extension
functor. In other words, it will suffice to show that if A ∈ MapN(Lin∗ ) (N(Lin∗ ), C⊗ ), then A0 = A ◦ ψ3 is
a q-right Kan extension of A. Fix an object (J, J0 , hni, α) ∈ J2 . We observe that Lin∗ ×J3 (J3 )/J has a
final object, given by J∗ ∈ Lin∗ . Consequently, A0 is a q-right Kan extension of A at J if and only if the
induced map A0 (J, J0 , hni, α) → A0 (J∗ ) is q-Cartesian. We now observe that this map is an equivalence by
construction.
To prove (2), it will suffice to show that φ04 (E3 (C)) ⊆ Algs (C) and θ40 (Algs (C)) ⊆ E3 (C). The first inclusion
is obvious, and the second follows from Remark 1.4.14.
Proof of Theorem 1.4.2. The equivalence of (1) and (2) follows from Proposition T.[Link]. We will prove
(2). Using Remark 1.7.11, we may assume without loss of generality that the monoidal structure on C is the
restriction of a Segal monoidal category q : C⊗ → N(Lin∗ ). Let θ : Algs (C) → C denote the forgetful functor.
Using Proposition 1.7.15, we are reduced to proving the following analogue of (2):

(20 ) For every C ∈ C, there exists a Segal algebra object A0 ∈ Algs (C) and a map η : C → θ(A0 ) such that,
for every B 0 ∈ Algs (C), composition with η induces a homotopy equivalence MapAlgs (C) (A0 , B 0 ) →
MapC (C, θ(B 0 )).

Fix an object C ∈ C. Using Lemmas 1.4.7, 1.4.11, and 1.4.15, we can choose C ∈ E02 (C) such that
◦ ◦ e ∈ E3 (C) which is a q-left Kan extension
C(h1i , h1i , h1i, idh1i ) = C. Using Lemma 1.4.20, we can choose C
0
of C. Let A = C| e N(Lin∗ ), so that C e determines a morphism C → θ(A0 ) in C. We claim that this
morphism has the desired property. Let B 0 be an arbitrary object of Algs (C). The canonical map g :
MapAlgs (C) (A0 , B 0 ) → MapC (C, θ(B 0 )) factors as a composition
g4
MapAlgs (C) (A0 , B 0 ) → MapE3 (C) (C,
e B 0 ◦ ψ3 )
g3
→ MapE2 (C) (C, B 0 ◦ ψ2 )
g2
→ MapE1 (C) (C| N(J1 ), B 0 ◦ ψ1 )
g1
→ MapE0 (C) (C| N(J0 ), B 0 ◦ ψ0 )
g0
→ MapC (C, θ(B 0 )).

The maps g0 , g1 , g2 , g3 , and g4 are all homotopy equivalences (Lemmas 1.4.7, 1.4.11, 1.4.15, T.[Link], and
1.4.23, respectively). It follows that g is a homotopy equivalence, as desired. This completes the proof of
(20 ). Taking A ∈ Alg(C) to be the composition of A0 with the map N(∆)op`→ N(Lin∗ ), we deduce (2).
The third assertion of Lemma 1.4.20 implies that the canonical map n C ⊗n → A([1]) ' θ(A0 ) is an
equivalence. This (together with observation that free algebras generated by C are uniquely determined
up to equivalence) proves the “only if” direction of (3). For the converse, suppose given an algebra object
B ∈ Alg(C) and a map C → B([1]) which induces an equivalence n C ⊗n → B([1]). Since A is freely
`
generated by C, there exists a map f : A → B in Alg(C) and a commutative triangle

A([1])
zz= HH 0
HH f
zz HH
zz HH
zz H$
C / B([1])

33
in C. Consider the induced diagram

A([1])
sss9 HH
HH f 0
sss HH
sss HH
s H$
⊗n /
`
n≥0 C B([1]).

Using the two-out-of-three property, we deduce that f 0 is an equivalence. Corollary 1.5.4 now implies that
f is an equivalence, so that B is freely generated by C as desired.

Remark 1.4.24. The proof of Theorem 1.4.2 can be adapted to prove the following analogue of Theorem
1.4.2:
(∗) Suppose that C is a monoidal ∞-category which admits countable coproducts, and that the tensor
product ⊗ : C × C → C preserves countable coproducts separately in each variable. Then the forgetful
functor θ : Algnu (C) → C admits a left adjoint (here Algnu (C) denotes the ∞-category of nonunital
algebra objects of C; see Definition 2.2.1). Moreover, the composition of θ with this left adjoint is
(canonically) identified with the functor C 7→ n>0 C ⊗n .
`

In fact, assertion (∗) is slightly easier to prove that Theorem 1.4.2, since it is possible to avoid the formalism
of Segal monoidal categories. We leave the details to reader.

1.5 Limits and Colimits of Algebras

Let C be a monoidal ∞-category. Our goal in this section is to prove the existence of limits and colimits in
Alg(C), given suitable assumptions on C. The case of limits is straightforward: our main result is Corollary
1.5.3, which asserts that limits in Alg(C) can be computed at the level of the underlying objects of C.
Colimits in Alg(C) are more complicated. One might naively guess that one can form a colimit of a
diagram K → Alg(C) by first forming a colimit of the underlying diagram K → C, and then endowing it
with an algebra structure. However, this obviously fails in some simple cases. For example, if C is the (nerve
of the) category of sets, equipped with its Cartesian monoidal structure then Alg(C) is (the nerve of) the
category of associative monoids. In this case, both C and Alg(C) admit coproducts, but the forgetful` functor
θ : Alg(C) → C does not preserve coproducts: if M and N are monoids, then the disjoint union M N does
not inherit the structure of a monoid.
Nevertheless, there are some colimits in Alg(C) which can (often) be computed at the level of the un-
derlying objects of C. Namely, we will see that this is true when K is a sifted simplicial set (Definition
T.[Link]); see Proposition 1.5.10 for a precise statement.
The construction of general colimits in Alg(C) is more difficult. Using general arguments, we can reduce
to problem of constructing sifted colimits (which is addressed by Proposition 1.5.10) and the problem of
constructing coproducts. To handle the latter, we will not proceed directly. Instead, we will use the results
of §3.5 and §1.4 to resolve arbitrary objects of Alg(C) by free algebras. We can then reduce to the problem
of constructing coproducts of free algebras, which are easily computed in terms of coproducts in C.
We now begin our analysis by considering limits in Alg(C).
Lemma 1.5.1. Let C⊗ and D⊗ be monoidal structures on ∞-categories C and D, respectively, and let
α : F → F 0 be a morphism in FunLax (C⊗ , D⊗ ). The following are equivalent:

(1) The transformation α is an equivalence in FunLax (C⊗ , D⊗ ).

(2) For every object C ∈ C, the morphism α(C) is an equivalence in D.
Proof. The implication (1) ⇒ (2) is obvious. Conversely, suppose that (2) is satisfied. Let X ∈ C⊗ [n] ; we wish
to prove that α(X) is an equivalence in D⊗
[n] . Since D ⊗
is a monoidal ∞-category, it suffices to show that for

34
0 ≤ i < n, the image of α(X) under the map D⊗ ⊗
[n] → D[1] ' D induced by the inclusion [1] ' {i, i + 1} ,→ [n]
is an equivalence in D. Since F and F 0 are lax monoidal functors, this morphism can be identified with
α(Xi ), where Xi is the image of X under the corresponding map C⊗ ⊗
[n] → C[1] ' C. The desired result now
follows immediately from (2).
Proposition 1.5.2. Let C⊗ and D⊗ be monoidal structures on ∞-categories C and D, respectively, and let
q : K → FunLax (C⊗ , D⊗ ). Suppose that, for every object C ∈ C, the induced diagram K → D admits a limit
in D. Then:

(1) The diagram q has a limit in FunLax (C⊗ , D⊗ ).

(2) An arbitrary extension q : K / → FunLax (C⊗ , D⊗ ) of q is a limit diagram if and only if, for every object
C ∈ C, the induced map K / → D is a limit diagram.

Proof. We will prove that there exists a limit diagram q : K / → FunLax (C⊗ , D⊗ ) satisfying the condition
of (2). This will prove (1). Moreover, the “only if” direction of (2) will follow from the uniqueness of limit
diagrams up to equivalence. To prove the “if” direction of (2), suppose that q 0 : K / → FunLax (C⊗ , D⊗ ) is an
arbitrary extension of q which satisfies the condition of (2). Since q is a limit diagram, we obtain a natural
transformation α : q 0 → q in FunLax (C⊗ , D⊗ )/q . It follows from Lemma 1.5.1 that α is an equivalence.
For every object X ∈ C⊗ ⊗
[n] , let qX : K → D[n] be the diagram induced by q. We observe that if X
corresponds to a sequence of objects (C1 , . . . , Cn ) under the equivalence C⊗[n] ' C , then qX has a limit,
n

⊗
which projects under the equivalence D[n] ' D to a collection of limits for the diagrams {qCi }1≤i≤n .
n

Applying Proposition T.[Link], we conclude that q admits a limit in the ∞-category MapN(∆)op (C⊗ , D⊗ ),
and that the limit of q belongs to FunLax (C⊗ , D⊗ ). It is readily verified that this limit has the desired
properties.
Corollary 1.5.3. Let D be an ∞-category equipped with a monoidal structure, let θ : Alg(D) → D be the
forgetful functor, and let q : K → Alg(D) be a diagram. Suppose that θ ◦ q has a limit in D. Then:
(1) The diagram θ has a limit in Alg(D).
(2) An arbitrary extension q : K / → Alg(D) is a limit diagram if and only if θ ◦ q is a limit diagram.
Proof. Apply Proposition 1.5.2 in the case where C⊗ = N(∆)op .
Corollary 1.5.4. Let C be an ∞-category equipped with a monoidal structure. Then the forgetful functor
θ : Alg(C) → C is conservative. In other words, a morphism f : A → A0 of algebra objects of C is an
equivalence in Alg(C) if and only if θ(f ) is an equivalence in C.
Corollary 1.5.5. Let C be a monoidal ∞-category, and suppose that C admits a final object. Then:
(1) The ∞-category Alg(C) admits a final object.
(2) An object A ∈ Alg(C) is final if and only if its image in C is final.
In other words, if C is a monoidal ∞-category which admits a final object U , then U can be promoted
(in an essentially unique way) to an algebra object of C.
Remark 1.5.6. Let C be a monoidal ∞-category which admits a final object. Corollary 1.5.5 has the
following analogue for nonunital algebras, which can be proven using a similar argument (see §2.2 for an
explanation of the notation):
(1) The ∞-category Algnu (C) admits a final object.
(2) An object A ∈ Algnu (C) is final if and only if its image in C is final.

35
Definition 1.5.7. Let K be a simplicial set, and let C be an ∞-category which admits K-indexed colimits.
We will say that a monoidal structure on C is compatible with K-indexed colimits if, for every object C ∈ C,
the functors
•⊗C :C→C C ⊗•:C→C
preserve K-indexed colimits.
In the case where K is sifted (see Definition T.[Link]), Definition 1.5.7 admits a convenient reformulation.
Lemma 1.5.8. Let K be a sifted simplicial set, and let C⊗ → N(∆)op be a monoidal ∞-category which is
compatible with K-indexed colimits. For each morphism f : [m] → [n] in N(∆)op , the associated functor
θ : C⊗ ⊗
[n] → C[m] preserves K-indexed colimits.

Proof. Let k be the largest element of the set {f (i + 1) − f (i)}0≤i<m ; we will work by induction on k. Let
C = C⊗ ⊗
[1] . Using the equivalence C[m] ' C , we can reduce to the case m = 1. Similarly, we may reduce to
m

the case where the map f : [1] → [n] preserves initial and final objects. If n > 2, then we can factor f as
f0 f 00
a composition [1] → [2] → [n], and the desired result holds for f 0 and f 00 by the inductive hypothesis. We
may therefore suppose that n ≤ 2. If n = 1, then θ is an equivalence and there is nothing to prove. If n = 0,
then θ is equivalent to a constant map; the desired result then follows from Corollary T.[Link], since K is
weakly contractible. When n = 2, we apply Proposition T.[Link].
Lemma 1.5.9. Let p : X → S be a coCartesian fibration of simplicial sets, and let K be an arbitrary
simplicial set. Suppose that:
(i) For each vertex s of S, the fiber Xs admits K-indexed colimits.
(ii) For each edge s → s0 of S, the associated functor Xs → Xs0 preserves K-indexed colimits.
Then:

(1) Every diagram K → MapS (S, X) admits a colimit.

(2) An arbitrary map K . → MapS (S, X) is a colimit diagram if and only, for each s ∈ S, the associated
map K . → Xs is a colimit diagram.
Proof. Combine Proposition T.[Link] with (the dual of) Lemma 2.3.1.

Proposition 1.5.10. Let C⊗ and D⊗ be monoidal structures on ∞-categories C and D, respectively. Let K
be a sifted simplicial set. Suppose that D admits K-indexed colimits, and that the monoidal structure on D
is compatible with K-indexed colimits. Then:

(1) The ∞-category MapN(∆)op (C⊗ , D⊗ ) admits K-indexed colimits.

(2) The full subcategories

FunMon (C⊗ , D⊗ ) ⊆ FunLax (C⊗ , D⊗ ) ⊆ MapN(∆)op (C⊗ , D⊗ )

are stable under K-indexed colimits.

(3) The forgetful functor θ : FunLax (C⊗ , D⊗ ) → Fun(C, D) detects K-indexed colimits. More precisely, a
map q : K . → FunLax (C⊗ , D⊗ ) is a colimit diagram if and only if θ ◦ q is a colimit diagram.

36
Proof. Assertions (1), (2), and the “only if” part of (3) follow immediately from Lemmas 1.5.8 and 1.5.9. To
prove the “if” direction of (3), let us suppose that q : K . → FunLax (C⊗ , D⊗ ) is is such that θ ◦ q is a colimit
diagram. We wish to prove that q is a colimit diagram. In view of (2) and Lemma 1.5.9, it will suffice to
show that for each object X ∈ C⊗ ⊗
[n] , the induced map q X : K → D[n] is a colimit diagram. In view of the
.

equivalence D⊗[n] → D , it will suffice to show that each of the composite maps
n

q
K. →
X
D⊗
[n] ' D → D
n

is a colimit diagram. Since the values assumed by q are lax monoidal functors, these compositions can be
qX
identified with K . →i D, where X corresponds to (X1 , . . . , Xn ) ∈ Cn under the equivalence C⊗
[n] ' C .
n

Corollary 1.5.11. Let K be a sifted simplicial set, let D be an ∞-category which admits K-indexed colimits,
and let D⊗ → N(∆)op be a monoidal structure on D which is compatible with K-indexed colimits. Then:

(1) The ∞-category Alg(D) admits K-indexed colimits.

(2) The forgetful functor θ : Alg(D) → D detects K-indexed colimits. More precisely, a map q : K . →
Alg(D) is a colimit diagram if and only if θ ◦ q is a colimit diagram.
Proof. Apply Proposition 1.5.10 in the case C⊗ = N(∆)op .

Let us now turn to the problem of constructing general colimits in Alg(C). In view of Corollary T.[Link]
and Lemma S.14.8, arbitrary colimits in Alg(C) can be built out of filtered colimits, geometric realizations,
and finite coproducts. The cases of filtered colimits and geometric realizations are addressed by Corollary
1.5.11 (in view of Example T.[Link] and Lemma T.[Link]). It will therefore suffice construct finite coproducts
in Alg(C).

Proposition 1.5.12. Let κ be an uncountable regular cardinal. Let C be an ∞-category which admits κ-small
colimits, endowed with a monoidal structure which is compatible with κ-small colimits. Then:
(1) The forgetful functor θ : Alg(C) → C has a left adjoint ψ.
(2) For every object A ∈ Alg(C), there exists a simplicial object A• of Alg(C) having A as a colimit, such
that each An belongs to the essential image of ψ.
(3) The ∞-category Alg(C) admits κ-small colimits.
Proof. Part (1) was established as Proposition 1.4.2, and (2) follows from Corollary 1.5.11 and Proposition
3.5.5. We now prove (3). Since Alg(C) admits κ-small sifted colimits (Corollary 1.5.11), it will suffice to
show that Alg(C) admits κ-small coproducts (in fact, it suffices to treat the case of finite coproducts, but we
will not need this). Choose a κ-small collection {Aβ }β∈B of objects of Alg(C). We wish to show that there
exists a coproduct for this collection in Alg(C). Let ψ : C → Alg(C) be a left adjoint to the inclusion functor.
Let us say that an object of Alg(C) is free if it belongs to the essential image of ψ. Since ψ preserves all
colimits which exist in C (Proposition T.[Link]), the coproduct of the collection {Aβ }β∈B exists whenever
each Aβ is free.
We now treat the general case. According to (2), each Aβ can be obtained as the geometric realization
of a simplicial object Aβ• of Alg(C), where each Aβn is free. Let p : β∈B N(∆)op → Alg(C) be the result of
`
amalgamating all of these simplicial objects. We can now apply the methods of §T.4.2.3 to decompose the
diagram p in two different ways:

(i) Since each summand of p has the object Aβ as a colimit, we can identify colimits of p with coproducts
of the family {Aβ }β∈B .

37
 op
)/[n] , and let p0[n] be the
`
(ii) For each object [n] ∈ ∆, let p[n] denote the restriction of p to β∈B (N(∆)
`
restriction of p to β∈B {[n]}. Since the inclusion
a a
{[n]} ⊆ (N(∆)op )/[n]
β∈B β∈B

is cofinal, we can identify colimits of p[n] with colimits of p0[n] , which are coproducts of the family
{Aβn }β∈B . The coproducts An = β∈B Aβn exist since each Aβn is free. Using the methods of §T.4.2.3,
`
we can organize the coproducts An into a simplicial object A• , such that colimits of p can be identified
with colimits of A• .
We now observe that the simplicial object A• of Alg(C) has a colimit in Alg(C), by virtue of Corollary
1.5.11.

Corollary 1.5.13. Let C be an ∞-category which admits small colimits, and suppose that C is endowed with
a monoidal structure which is compatible with small colimits. Then Alg(C) admits small colimits.
We now treat the question of whether or not Alg(C) is a presentable ∞-category.
Lemma 1.5.14. Let C⊗ → N(∆)op be a monoidal structure on an ∞-category C = C⊗
[1] . The following
conditions are equivalent:

(1) The ∞-category C is accessible, and the monoidal structure on C is compatible with κ-filtered colimits
for all sufficiently large κ.
(2) Each fiber of p is accessible, and for every morphism [m] → [n] in ∆, the associated functor C⊗ ⊗
[n] → C[m]
is accessible.

Proof. Suppose first that (2) is satisfied. We deduce immediately that C is accessible, and that the tensor
product ⊗ : C × C ' C⊗ [2] → C is an accessible functor. It now suffices to observe that for each C ∈ C, the
inclusion {C} × C ⊆ C × C is also an accessible functor.
Now suppose that (1) holds. Each fiber C⊗ [n] is equivalent to an n-fold product of copies of C, and is
therefore accessible. Now choose a morphism [m] → [n] in ∆; we wish to show that the associated functor
C⊗ ⊗ ⊗
[n] → C[m] is accessible. Using the equivalences C[m] ' C
m
and C⊗[n] ' C , we reduce easily to the case
n

where m = 1 and the map [m] → [n] preserves initial and final elements. The functor C⊗ ⊗
[n] → C[1] can be
identified with an iterated tensor product

⊗ : C×... × C → C.

If n = 0 then this functor is constant, if n = 1 it is equivalent to the identity, and if n = 2 it is accessible in

view of our assumption. For n > 2, we can use the associativity of ⊗ to reduce to the case n = 2.

Proposition 1.5.15. Let C be a monoidal ∞-category. Then:

(1) If C is accessible and the monoidal structure on C is compatible with κ-filtered colimits, for κ sufficiently
large, then Alg(C) is an accessible ∞-category.
(2) If C is a presentable ∞-category and the monoidal structure on C is compatible with small colimits,
then Alg(C) is a presentable ∞-category.

Proof. Assertion (1) follows from Lemma 1.5.14 and Proposition T.[Link]. To prove (2), we combine (1)
with Corollary 1.5.13.

38
1.6 Monoidal Model Categories
Let C be an ∞-category. In Example 1.1.5, we saw that if C is the nerve of a monoidal category, then C
inherits the structure of a monoidal ∞-category. Our goal in this section is to obtain some generalizations
of this result. We will be principally interested in the case where C = N(Ao ) is the ∞-category underlying
a simplicial model category A. We will show that if A admits a monoidal structure which is suitably
compatible with its simplicial and model structures (Definition 1.6.4), then C again inherits the structure of
a monoidal ∞-category. Moreover, there is a close relationship between the ∞-category of algebra objects
Alg(C) and the ordinary category of (strictly associative) algebras in A: see Theorem 1.6.17.
Definition 1.6.1. Let C be simplicial category. We will say that a monoidal structure on C is weakly
compatible with the simplicial structure on C provided that the operation ⊗ : C × C → C is endowed with the
structure of a simplicial functor, which is compatible with associativity and unit transformations.
We will say that a closed monoidal structure on C is compatible with the simplicial structure on C if it is
weakly compatible and for every triple of objects A, B, C ∈ C, the natural maps

Map(B, AC) → Map(A ⊗ B, C) ← Map(A, C B )

are isomorphisms of simplicial sets (see §T.A.1.3).

Remark 1.6.2. The compatibility of Definition 1.6.1 is not merely a condition, but additional data which
must be supplied. However, in practice the simplicial structure on the functor ⊗ tends to be clear in practice.
Let C be a simplicial category equipped with a weakly compatible monoidal structure, and let C⊗ the
category described in Definition 1.1.1. Then C⊗ is naturally endowed with the structure of a simplicial
category, where we define the simplicial mapping space MapC⊗ ([C1 , . . . , Cn ], [C10 , . . . , Cm
0
]) to be the disjoint
union a Y
MapC (Cf (i−1)+1 ⊗ . . . ⊗ Cf (i) , Ci0 )
f :[m]→[n] 1≤i≤m

Proposition 1.6.3. Let C be a fibrant simplicial category with a weakly compatible monoidal structure, and
let C⊗ be the simplicial category constructed above. Then the induced map p : N(C⊗ ) → N(∆)op is a monoidal
structure on the ∞-category N(C).

Proof. We first show that p is a coCartesian fibration. It is obviously an inner fibration, since N(C⊗ ) is an
∞-category and N(∆)op is the nerve of an ordinary category. Choose an object [C1 , . . . , Cn ] ∈ C⊗ and a
morphism f : [m] → [n] in ∆. Then there exists a morphism f : [C1 , . . . , Cn ] → [C10 , . . . , Cm
0
] in N(C⊗ )
0
which covers f , where Ci ' Cf (i−1)+1 ⊗ . . . ⊗ Cf (i) . It follows immediately from Proposition T.[Link]
that f is p-coCartesian. We conclude the proof by observing that the fiber N(C⊗ )[n] is isomorphic to
N(C)n , and the projection onto the ith factor can be identified with the functor induced by the inclusion
[1] ' {i − 1, i} ⊆ [n].
In practice, the hypotheses of Proposition 1.6.3 are often too strong. Suppose, for example, that A is
a simplicial model category, and let Ao be the full (simplicial) subcategory of A spanned by the fibrant-
cofibrant objects. Then Ao is a fibrant simplicial category, and we refer to the simplicial nerve N(Ao ) as the
underlying ∞-category of A. A monoidal structure on A need not restrict to a monoidal structure on Ao ,
even if it is compatible with the model structure of A (see Definition 1.6.4 below). Nevertheless, we can use
a variant of Proposition 1.6.3 to endow N(Ao ) with the structure of a monoidal ∞-category.
Definition 1.6.4. Let A be a simplicial model category. We will say that a monoidal structure ⊗ on
A is compatible with the simplicial model structure if it is compatible with the simplicial structure on A
(Definition 1.6.1) and satisfies the following conditions:

39
 (1) For every pair of cofibrations i : A → A0 , j : B → B 0 in A, the induced map
a
k : (A ⊗ B 0 ) (A0 ⊗ B) → A0 ⊗ B 0
A⊗B

is a cofibration. Moreover, if either i or j is a weak equivalence, then k is a weak equivalence.

(2) The unit object 1 of A is cofibrant.
(3) The monoidal structure on A is closed.
Proposition 1.6.5. Let A be a simplicial model category with a compatible monoidal structure, let A⊗ be
defined as in Definition 1.1.1, endowed with the simplicial structure considered in Proposition 1.6.3. Let
Ao be the full subcategory of A spanned by the fibrant-cofibrant objects, and let A⊗,o be the full subcategory
of A⊗ spanned by those objects [C1 , . . . , Cn ] such that each Ci belongs to Ao . Then the natural map p :
N(A⊗,o ) → N(∆)op determines a monoidal structure on the ∞-category N(Ao ).
Proof. Our first step is to prove that N(A⊗,o ) is an ∞-category. To prove this, it will suffice to show that
A⊗,o is a fibrant simplicial category. Let [C1 , . . . , Cn ] and [C10 , . . . , Cm
0
] be objects of A⊗,o . Then
MapA⊗,o ([C1 , . . . , Cn ], [C10 , . . . , Cm
0
])
is a disjoint union of products of simplicial sets of the form MapA (Ci ⊗ . . . ⊗ Cj , Ck0 ). Each of these simplicial
sets is a Kan complex, since Ci ⊗ . . . ⊗ Cj is cofibrant and Ck0 is fibrant.
Since p is a map from an ∞-category to the nerve of an ordinary category, it is automatically an inner
fibration. We next claim that p is a coCartesian fibration. Let [C1 , . . . , Cn ] be an object of A⊗,o , and let
f : [m] → [n] be a map in ∆. For each 1 ≤ i ≤ m, choose a trivial cofibration
ηi : Cf (i−1)+1 ⊗ . . . ⊗ Cf (i) → Ci0 ,

where Ci0 is a fibrant object of A. Together these determine a map f : [C1 , . . . , Cn ] → [C10 , . . . , Cm
0
] in
⊗,o
N(A ). We claim that f is p-coCartesian. In view of Proposition T.[Link], it will suffice to show that
for every morphism g : [k] → [m] in ∆ and every [C100 , . . . , Ck00 ] ∈ A⊗,o , the induced map
MapA⊗ ([C10 , . . . , Cm
0
], [C100 , . . . , Ck00 ]) ×Hom∆ ([k],[m]) {g} → MapA⊗ ([C1 , . . . , Cn ], [C100 , . . . , Ck00 ])
determines a homotopy equivalence onto the summand of MapA⊗ ([C1 , . . . , Cn ], [C100 , . . . , Ck00 ]) spanned by
those morphisms which cover the map h = f ◦ g : [k] → [n]. Unwinding the definitions, it will suffice to prove
that for 1 ≤ i ≤ k, the induced map
0 0
MapA (Cg(i−1)+1 ⊗ . . . ⊗ Cg(i) , Ci00 ) → MapA (Ch(i−1)+1 ⊗ . . . ⊗ Ch(i) , Ci00 )
is a homotopy equivalence of Kan complexes. For this, we need only show that the map
0 0
η : Ch(i−1)+1 ⊗ . . . ⊗ Ch(i) → Cg(i−1)+1 ⊗ . . . ⊗ Cg(i)
is a weak equivalence. This follows from the observation that η can be identified with the tensor product of
the maps {ηj }g(i−1)<j≤g(i) , each of which is a weak equivalence between cofibrant objects.
We now conclude the proof by observing that the fiber N(A⊗,o )[n] is isomorphic to an n-fold product of
N(Ao ) with itself, and that the projection onto the jth factor can be identified with the functor associated
to the inclusion [1] ' {j − 1, j} ⊆ [n].
Example 1.6.6. Let A be a simplicial model category. Suppose that the Cartesian monoidal structure on
A is compatible with the model structure (in other words, that the final object of A ` is cofibrant, and that for
any pair of cofibrations i : A → A0 , j : B → B 0 , the induced map i ∧ j : (A × B 0 ) A×B (A0 × B) → A0 × B 0
is a cofibration, trivial if either i or j is trivial). Then there is a canonical map of simplicial categories
θ : A⊗ → A, given on objects by the formula θ([A1 , . . . , An ]) = A1 × . . . × An . Since the collection of
fibrant-cofibrant objects of A is stable under finite products, θ induces a map A⊗,o → Ao . It follows that
N(A⊗,o ) → N(∆)op induces the Cartesian monoidal structure on the ∞-category N(Ao ).

40
Definition 1.6.7. Let C be a monoidal ∞-category. We will say that C is left closed if, for each C ∈ C, the
functor D 7→ C ⊗ D admits a right adjoint. Similarly, we will say that C is right closed if, for each C ∈ C,
the functor D 7→ D ⊗ C admits a right adjoint. We will say that C is closed if it is both left closed and right
closed.
Remark 1.6.8. In view of Proposition T.[Link], the condition that a monoidal ∞-category C be closed can
be checked at the level of the (H-enriched) homotopy category of C, with its induced monoidal structure.
More precisely, C is right closed if and only if, for every pair of objects C, D ∈ C, there exists another
object DC and a map DC ⊗ C → D with the following universal property: for every E ∈ C, the induced
map MapC (E, DC ) → Map(E ⊗ C, D) is a homotopy equivalence. In this case, the construction D 7→ DC
determines a right adjoint to the functor E 7→ E ⊗ C.
Remark 1.6.9. Let A be a monoidal model category equipped with a compatible simplicial structure, and
let C be a cofibrant object of A. Then the construction D 7→ C ⊗ D determines a left Quillen functor
fC : A → A. Suppose that C is also fibrant, so that C can be identified with an object of the underlying
∞-category N(Ao ). Then tensor product with C also induces a functor LfC from N(Ao ) to itself. The proof
of Proposition 1.6.5 shows that LfC can be identified with a left derived functor of fC . In particular, LfC
admits a right adjoint. It follows that the monoidal ∞-category N(Ao ) is left closed in the sense of Definition
1.6.7 (the same argument shows that N(Ao ) is also right closed). In particular, the functor LfC preserves
all colimits which exist in N(Ao ).
Let C be a monoidal category. An algebra object of C is an object A ∈ C equipped with maps

1 → A, A⊗A→A

which satisfy the usual unit and associativity axioms. The collection of algebra objects of C can be organized
into a category which we will denote by Alg(C).
Suppose now that C is a monoidal category equipped with a compatible simplicial structure. Then Alg(C)
inherits the structure of a simplicial category, where we let MapAlg(C) (A, B) ⊆ MapC (A, B) be the simplicial
subset described by the following property: a map K → MapC (A, B) factors through MapAlg(C) (A, B) if and
only if the diagrams
∆0 × K / MapC (1, A) × MapC (A, B)

 
∆0 / MapC (1, B)

KF / K ×K / MapC (A, B) × MapC (A, B) / MapC (A ⊗ A, B ⊗ B)

FF
FF
FF
FF
# 
∆0 × K / MapC (A ⊗ A, A) × MapC (A, B) / MapC (A ⊗ A, B)

commute. The simplicial nerve of Alg(C) is typically not equivalent to the category of algebra objects of
N(C), where N(C) is endowed with the monoidal structure of Proposition 1.6.3. In fact, Alg(C) need not be
a fibrant simplicial category, even when C is itself fibrant. However, in the case where C is a monoidal model
category we can often remedy the situation by passing to a suitable subcategory of Alg(C).
In the arguments which follow, we will need to invoke the following hypothesis (formulated originally by
Schwede and Shipley; see [60]):
Definition 1.6.10 (Monoid Axiom). Let A be a combinatorial symmetric monoidal model category. Let U
be the collection of all morphisms of A having the form
idX ⊗f
X ⊗Y → X ⊗ Y 0,

41
where f is a trivial cofibration, and let U denote the weakly saturated class of morphisms generated by U
(Definition T.A.1.2.2). We will say that A satsifies the monoid axiom if every morphism of U is a weak
equivalence in A.
Remark 1.6.11. Let A be a combinatorial symmetric monoidal model category in which every object is
cofibrant, and let U and U be as in Definition 1.6.10. Then every morphism belonging to U is a trivial
cofibration. Since the collection of trivial cofibrations in A is weakly saturated, we conclude that A satisfies
the monoid axiom.
Notation 1.6.12. Let f : X → X 0 and g : Y → Y 0 be morphisms in a monoidal category A which admits
pushouts. We define the pushout product of i and j to be the induced map
a
f g : (X ⊗ Y 0 ) (X 0 ⊗ Y ) → X 0 ⊗ Y 0 .
X⊗Y

The operation ∧ endows the category Fun([1], A) with a monoidal structure, which is symmetric if the
monoidal structure on A is symmetric.
Lemma 1.6.13. Let A be a combinatorial symmetric monoidal model category which satisfies the monoid
axiom, and let U be as in Definition 1.6.10. Then:

(1) If f : X → X 0 belongs to U and Y is an object of A, then f ⊗ idY : X ⊗ Y → X 0 ⊗ Y belongs to U .

(2) If f, g ∈ U , then f ∧ g ∈ U .
Proof. To prove (1), let S denote the collection of all morphisms f in A such that f ⊗ idY belongs to U . It
is easy to see that S is weakly saturated. It will therefore suffice to show that U ⊆ S, which is obvious.
To prove (2), we use the same argument. Fix g, and let S 0 be the set of all morphisms f ∈ A such that
f ∧ g belongs to U . We wish to prove that U ⊆ S 0 . Since S 0 is weakly saturated, it will suffice to show that
U ⊆ S 0 . In other words, we may assume that f is of the form f0 ⊗ idA , where f0 is a trivial cofibration in
A. Similarly, we may assume that g = g0 ⊗ idB . Then f ∧ g = (f0 ∧ g0 ) ⊗ (idA⊗B ), which belongs to U since
f0 ∧ g0 is a trivial cofibration in A.
Lemma 1.6.14. Let A be a monoidal model category. Suppose given commutative diagrams

A /B X /Y

   
A0 / B0 X0 / Y0

where every object is cofibrant, the horizontal arrows are cofibrations, and the vertical arrows are weak
equivalences. Then in the induced diagram

/ B⊗Y
`
(A ⊗ Y ) A⊗X (B ⊗ X)

`  
(A0 ⊗ Y 0 ) A0 ⊗X 0 (B
0
⊗ X 0) / B0 ⊗ Y 0

has the same properties.

Proof. The assertion that the horizontal arrows are cofibrations follows immediately from the definition of
a monoidal model category. Since every object appearing in either of the original diagrams, we deduce that
each of the morphisms
A ⊗ X → A0 ⊗ X 0 A ⊗ Y → A0 ⊗ Y 0

42
 B ⊗ X → B0 ⊗ X 0 B ⊗ Y → B0 ⊗ Y 0
is a weak equivalence between cofibrant objects. We obtain a weak equivalence of cofibrant diagrams

A⊗Y o A⊗X / B⊗X

 
A0 ⊗ Y 0 o A0 ⊗ X 0 / B0 ⊗ X 0,
`
from which we obtain (by passing to the colimit) a weak equivalence of cofibrant objects (A ⊗ Y ) A⊗X (B ⊗
X) → (A0 ⊗ Y 0 ) A0 ⊗X 0 (B 0 ⊗ X 0 ) as desired.
`

Proposition 1.6.15. [Schwede-Shipley] Let A be a combinatorial monoidal model category. Assume that
either every object of A is cofibrant, or that A is a symmetric monoidal model category which satisfies the
monoid axiom. Then:

(1) The category Alg(A) admits a combinatorial model structure, where:

(W ) A morphism f : A → B of algebra objects of A is a weak equivalence if it is a weak equivalence

when regarded as a morphism in A.
(F ) A morphism f : A → B of algebra objects of A is a fibration if it is a fibration when regarded as
a morphism in A.

(2) The forgetful functor θ : Alg(A) → A is a right Quillen functor.

(3) If A is equipped with a compatible simplicial structure, then Alg(A) inherits the structure of a simplicial
model category.
(4) Suppose that every object of A is cofibrant and that the collection of weak equivalences in A is stable
under filtered colimits. Then Alg(A) is left proper.

Proof. We first observe that the category Alg(A) is presentable (this follows, for example, from Proposition
1.5.15). Recall that a collection S of morphisms in a presentable category C is weakly saturated if it is
stable under pushouts, retracts, and transfinite composition (see Definition T.A.1.2.2); we will say that S is
generated by a subset S0 ⊂ S if S is the smallest weakly saturated collection of morphisms containing S0 .
Since C is combinatorial, there exists a (small) collection of morphisms I = {iα : C → C 0 } which generates
the class of cofibrations in A, and a (small) collection of morphisms J = {jα : D → D0 } which generates the
class of trivial cofibrations in A.
Let F : A → Alg(A) be a left adjoint to the forgetful functor. Let F (I) be the weakly saturated class
of morphisms in Alg(A) generated by {F (i) : i ∈ I}, and let F (J) be defined similarly. Unwinding the
definitions, we see that a morphism in Alg(A) is a trivial fibration if and only if it has the right lifting
property with respect to F (i), for every i ∈ I. Invoking the small object argument, we deduce that every
f0 f 00
morphism f : A → C in Alg(A) admits a factorization A → B → C where f 0 ∈ F (I) and f 00 is a trivial
fibration. Similarly, we can find an analogous factorization where f 0 ∈ F (J) and f 00 is a fibration.
Using a standard argument, we may reduce the proof of (1) to the problem of showing that every
morphism belonging to F (J) is a weak equivalence in Alg(A). If A is symmetric and satisfies the monoid
axiom, then let U be as in Definition 1.6.10; otherwise let U be the collection of all trivial cofibrations in A.
Let S be the collection of all morphisms in Alg(A) such that the induced map in A belongs to U . Then S
is weakly saturated, and every element of S is a weak equivalence. To complete the proof, it will suffice to
show that F (J) ⊆ S. In other words, we must prove:

43
 (∗) Let
F (i)
F (C) / F (C 0 )

 
A / A0 f

be a pushout diagram in Alg(A). If i is a trivial cofibration in A, then f ∈ S.

Let ∅ be an initial object of A, and let j : ∅ → A be the unique morphism. We now observe that A0 can
be obtained as the direct limit of a sequence
f1 f2
A = A(0) → A(1) → . . .
of objects of A, where each fn is a pushout of j ∧ i ∧ j ∧ . . . ∧ i ∧ j; here the factor i appears n times.
If every object of A is cofibrant, then we conclude that fn is a trivial cofibration using the definition of a
monoidal model category. If the monoidal structure on A is symmetric and satisfies the monoid axiom, then
repeated application of Lemma 1.6.13 shows that fn ∈ U . Since U is weakly saturated, it follows that f ∈ S
as desired. This completes the proof of (1).
Assertion (2) is obvious. To prove (3), we observe both A and Alg(A) are cotensored over simplicial
sets, and that we have canonical isomorphisms θ(AK ) ' θ(A)K for A ∈ Alg(A), K ∈ Set∆ . To prove that
Alg(A) is a simplicial model category, it will suffice to show that Alg(A) is tensored over simplicial sets,
and that given a fibration i : A → A0 in Alg(A) and a cofibration j : K → K 0 in Set∆ , the induced map
0 K0
AK → AK ×A0 K A0 is a fibration, trivial if either i or j is a fibration. The second claim follows from the
fact that θ detects fibrations and trivial fibrations. For the first, it suffices to prove that for K ∈ Set∆ , the
functor A 7→ AK has a left adjoint; this follows from the adjoint functor theorem.
We now prove (4). Let T be the collection of all morphisms α : X → Y in Alg(A) with the following
property: for every pushout diagram
X
α /Y

 
A / A0

β β0
 
B / B0,
if β is a weak equivalence, then β 0 is a weak equivalence. To prove that A is left proper, it will suffice to
show that every cofibration belongs to T . Using the assumption that the collection of weak equivalences
in A is stable under filtered colimits, we deduce that T is a weakly saturated class of morphisms. It will
therefore suffice to show that T contains every generating cofibration of the form F (i) : F (x) → F (y), where
i : x → y is a cofibration in A. The algebra objects A0 and B 0 can be defined as the direct limit of sequences
f1 f2
A = A(0) → A(1) → . . .
g1 g2
B = B (0) → B (1) → . . .
as in the proof of (∗). Since the collection of weak equivalences in A is stable under filtered colimits, it will
suffice to show that each of the maps βk : A(k) → B (k) is a weak equivalence. For k = 0 there is nothing
to prove; in the general case we work by induction on k. We have a map between the homotopy pushout
diagrams
fk0 0
gk
K / A ⊗ y ⊗ ... ⊗ y ⊗ A L / B ⊗ y ⊗ ... ⊗ y ⊗ B

  
A(k−1)
fk
/ A(k) B (k−1)
gk
/ B (k) .

44
Consequently, to prove that βk is a weak equivalence, it will suffice to show that βk−1 is a weak equivalence
(which follows from the inductive hypothesis) and that the maps

K→L A ⊗ y ⊗ ... ⊗ y ⊗ A → B ⊗ y ⊗ ... ⊗ y ⊗ B

are weak equivalences, which follows from Lemma 1.6.14.

We now come to the main result of this section, which asserts that if A is a simplicial model category
with a compatible monoidal structure, then every algebra object of the ∞-category N(Ao ) is equivalent to
a strictly associative algebra object in A. This is a kind of “straightening theorem” which we will use in §4
to compare our approach to the theory of A∞ -rings with more classical definitions.
Lemma 1.6.16. Let A be a combinatorial monoidal model category with a compatible simplicial structure,
and let C be a small category such that N(C) is sifted (Definition T.[Link]). Assume either that every object
of A is cofibrant, or that A satisfies the following pair of conditions:
(A) The monoidal structure on A is symmetric, and A satisfies the monoid axiom.
(B) The model category A is left proper, and the class of cofibrations in A is generated by cofibrations
between cofibrant objects (this is automatic if every object of A is cofibrant).
Then the forgetful functor N(Alg(A)o ) → N(Ao ) preserves N(C)-indexed colimits.
Proof. In view of Theorem T.[Link] and Proposition T.[Link], it will suffice to prove that the forgetful
functor θ : Alg(A) → A preserves homotopy colimits indexed by C. Let us regard Alg(A)C and AC as
endowed with the projective model structure (see §T.A.3.3). Let F : AC → A and FAlg : Alg(A)C → Alg(A)
be colimit functors, and let θC : Alg(A)C → AC be given by composition with θ. Since N(C) is sifted, there
is a canonical isomorphism of functors α : F ◦ θC ' θ ◦ FAlg . We wish to prove that this isomorphism persists
after deriving all of the relevant functors. Since θ and θC preserve weak equivalences, they can be identified
with their right derived functors. Let LF and LFAlg be the left derived functors of F and FAlg , respectively.
Then α induces a natural transformation α : LF ◦ θC → θ ◦ LFAlg ; we wish to show that α is an isomorphism.
Let A : C → Alg(A) be a strongly cofibrant object of Alg(A)C ; we must show that the natural map

LF (θC (A)) → θ(LFAlg (A)) ' θ(FAlg (A)) ' F (θC (A))

is a weak equivalence in A.
Let us say that an object X ∈ AC is good if each of the objects X(C) ∈ A is cofibrant, the object
F (X) ∈ A is cofibrant, and the natural map LF (X) → F (X) is a weak equivalence in A: in other words, if
the colimit of X is also a homotopy colimit of X. To complete the proof, it will suffice to show that θC (A) is
good, whenever A is a strongly cofibrant object of Alg(A)C . This is not obvious, since θC is a right Quillen
functor and does not preserve strongly cofibrant objects in general (note that we have not yet used the full
strength of our assumption that N(C) is sifted). To continue the proof, we will need a relative version of the
preceding condition. We will say that a morphism f : X → Y in AC is good if the following conditions are
satisfied:
(i) The objects X, Y ∈ AC are good.
(ii) For each C ∈ C, the induced map X(C) → Y (C) is a cofibration in A.
(iii) The map F (X) → F (Y ) is a cofibration in A.
We now make the following observations:
(1) The collection of good morphisms is stable under transfinite composition. More precisely, suppose
given an ordinal α and a direct system of objects {X β }β<α of AC . Suppose further that for every
0 < β < α, the map lim{X γ }γ<β → X β is good. Then the induced map X 0 → lim{X β }β<α is good.
−→ −→

45
 The only nontrivial point is to verify that the object X = lim{X β }β<α is good. For this, we observe
−→
X is a homotopy colimit of the system {X β } (in virtue of (ii)), while F (X) is a homotopy colimit of
the system {F (X β )} (in virtue of (iii)), and that the collection of homotopy colimit diagrams is stable
under homotopy colimits.

(2) Suppose given a pushout diagram

X
f
/Y

 f0 
X0 / Y0

in AC . If f is good and X 0 is good, then f 0 is good. Once again, the only nontrivial point is to show
that Y 0 is good. To see this, we observe that our hypotheses imply that Y 0 is homotopy pushout of
Y with X 0 over X. Similarly, F (Y 0 ) is a homotopy pushout of F (Y ) with F (X 0 ) over F (X). We
now invoke once again the fact that the class of homotopy colimit diagrams is stable under homotopy
colimits.
(3) Let F : C → A be a constant functor whose value is a cofibrant object of A. Then F is good. This
follows from the fact that N(C) is weakly contractible (Proposition T.[Link]).
(4) Every strongly cofibrant object of AC is good. Every strong cofibration between strongly cofibrant
objects of AC is good.

(5) If X and Y are good objects of AC , then X ⊗ Y is good. To prove this, we first observe that the
collection of cofibrant objects of A is stable under tensor products. Because N(C) is sifted, we have a
chain of isomorphisms in hA:

LF (X ⊗ Y ) ' LF (X) ⊗ LF (Y ) ' F (X) ⊗ F (Y ) ' F (X ⊗ Y ).

(6) Let f : X → X 0 be a good morphism in AC , and let Y be a good object of AC . Then the morphism
f ⊗ idY is good. Condition (i) follows from (5), condition (ii) follows from the fact that tensoring with
each Y (C) preserves cofibrations (since Y (C) is cofibrant), and condition (iii) follows by applying the
same argument to F (Y ) (and invoking the fact that F commutes with tensor products).

(7) Let f : X → X 0 and g : Y → Y 0 be good morphisms in AC . Then

a
f ∧ g : (X ⊗ Y 0 ) (X 0 ⊗ Y ) → X 0 ⊗ Y 0
X⊗Y

is good. Condition (ii) follows immediately from the fact that A is a monoidal model category.
Condition (iii) follows from the same argument, together with the observation that F commutes with
pushouts and tensor products. Condition (i) follows by combining (5), (6), and (2).
We observe that our assumption (B) (which is a consequence of the assumption that every object of A
is cofibrant) implies an analogous result for AC :
(B 0 ) The collection of all strong cofibrations in AC is generated by strong cofibrations between strongly
cofibrant objects.
Let T : AC → Alg(A)C be a left adjoint to θC . Using the small object argument and (B 0 ), we conclude
that for every strongly cofibrant object A ∈ Alg(A)C there exists a transfinite sequence {Aβ }β≤α in Alg(A)C
with the following properties:
(a) The object A0 is initial in Alg(A)C .

46
 (b) The object A is a retract of Aα .
(c) If λ ≤ α is a limit ordinal, then Aλ ' colim{Aβ }β<λ .
(d) For each β < α, there is a pushout diagram

T (f )
T (X 0 ) / T (X)

 
Aβ / Aβ+1

where f is a strong cofibration between strongly cofibrant objects of AC .

We wish to prove that θC (A) is good. In view of (b), it will suffice to show that θC (Aα ) is good. We
will prove a more general assertion: for every γ ≤ β ≤ α, the induced morphism uγ,β : θC (Aγ ) → θC (Aβ ) is
good. The proof is by induction on β. If β = 0, then we are reduced to proving that θC (A0 ) is good. This
follows from (a) and (3). If β is a nonzero limit ordinal, then the desired result follows from (c) and (1). It
therefore suffices to treat the case where β = β 0 + 1 is a successor ordinal. Moreover, we may suppose that
γ = β 0 : if γ < β 0 , then we observe that uγ,β = uβ 0 ,β ◦ uγ,β 0 and invoke (1), while if γ > β 0 , then γ = β and
we are reduced to proving that θC (Aβ ) is good, which follows from the assertion that uβ 0 ,β is good. We are
now reduced to proving the following:

(∗) Let
T (f )
T (X 0 ) / T (X)

 
B0
v /B

be a pushout diagram in Alg(A)C , where f : X 0 → X is a strong cofibration between strongly cofibrant

objects of AC . If θC (B 0 ) is good, then θC (v) is good.

To prove (∗), we set Y = θC (B) ∈ AC , Y 0 = θC (B 0 ) ∈ AC . Let g : ∅ → Y 0 the unique morphism, where

∅ denotes an initial object of AC . As in the proof of Proposition 1.6.15, Y can be identified with the colimit
of a sequence
w w
Y (0) →1 Y (1) →2 . . .
where Y (0) = Y 0 , and wk is a pushout of the morphism f (k) = g ∧ f ∧ g ∧ . . . ∧ f ∧ g, where the factor f
appears k times. In view of (1) and (2), it will suffice to prove that each f (k) is a good morphism. Since Y 0
is good, we conclude immediately that g is good. It follows from (4) that f is good. Repeated application
of (7) allows us to deduce that f (k) is good, and to conclude the proof.
We now come to our main result:
Theorem 1.6.17. Let A be a combinatorial monoidal model category equipped with a compatible simplicial
structure. Assume either:
(A) Every object of A is cofibrant.
(B) The model category A is left proper, the class of cofibrations in A is generated by cofibrations between
cofibrant objects, the monoidal structure on A is symmetric, and A satisfies the monoid axiom.
Then the canonical map
N(Alg(A)o ) → Alg(N(Ao ))
is an equivalence of ∞-categories.

47
Example 1.6.18. Let k be a field, and let A denote the category of complexes of k-vector spaces

. . . → Mn+1 → Mn → Mn−1 → . . . ,

with monoidal structure given by the formation of tensor products of complexes. The category A admits a
model structure, where:
(C) A map of complexes f : M• → N• is a cofibration if it induces an injection Mn → Nn of k-vector
spaces, for each n ∈ Z.
(F ) A map of complexes f : M• → N• is a fibration if it induces an surjection Mn → Nn of k-vector spaces,
for each n ∈ Z.
(W ) A map of complexes f : M• → N• is a weak equivalence if it is a quasi-isomorphism; that is, if f induces
an isomorphism on homology groups Hn (M• ) → Hn (N• ) for each n ∈ Z.
The category A also admits a simplicial structure, compatible with its monoidal model structure; see the
discussion in §S.13. Moreover, every object of A is cofibrant.
The ∞-category N(Ao ) can be identified with the (unbounded) derived ∞-category of k-vector spaces
(see Definition S.13.6); let us denote this ∞-category by D(k). According to Theorem 1.6.17, the ∞-category
of algebra object Alg(D(k)) is equivalent to N(Alg(A)o ): that is, to a suitable ∞-category of differential
graded algebras over k.
An analogous result holds if we replace the field k by an arbitrary commutative ring, and endow A with
the projective model structure. In this case, not every object of A is cofibrant, but A still satisfies hypothesis
(B) of Theorem 1.6.17 (see [60]).

Example 1.6.19. Let A be the category of symmetric spectra, as defined in [28]. Then A admits several
model structures which satisfy assumption (B) of Theorem 1.6.17. Using Corollary 4.2.6, we deduce that
N(Ao ) is equivalent, as a monoidal ∞-category, to the ∞-category Sp of spectra (endowed with the smash
product monoidal structure; see §4.2). Using Theorem 1.6.17, we deduce that our ∞-category A∞ = Alg(Sp)
of A∞ -rings is equivalent to the ∞-category underlying the category Alg(A) of symmetric ring spectra.

Example 1.6.20. [∞-Categorical MacLane Coherence Theorem] Let A be the category of marked simplicial
sets (see §T.3.1). Then A is a simplicial model category, which satisfies the hypotheses of Example 1.6.6.
The underlying ∞-category N(Ao ) can be identified with Cat∞ , the ∞-category of ∞-categories. Proposition
1.2.14 implies that composition with the Cartesian structure N(A⊗,o ) → N(Ao ) induces an equivalence of
∞-categories Alg(N(Ao )) → Mon(Cat∞ ). Combining this observation with Theorem 1.6.17, we conclude
that the ∞-category of monoid objects of Cat∞ is equivalent to the ∞-category underlying the category
of strictly associative monoids in A. In other words, every monoidal ∞-category C is equivalent (as a
monoidal ∞-category) to an ∞-category C0 equipped with a strictly associative monoid structure C0 × C0 → C0
(which determines a monoidal structure on C0 via Proposition T.[Link]). We regard this assertion as an
∞-categorical analogue of MacLane’s coherence theorem, which asserts that every monoidal category is
equivalent to a strict monoidal category (that is, a monoidal category in which the tensor product operation
⊗ is associative up to equality, and the associativity isomorphisms are simply the identity maps).
Proof of Theorem 1.6.17. Consider the diagram

N(Alg(A)o ) / Alg(N(Ao ))
MMM qq
MMMG qqq
MMM qq
M& xqqq G
0

o
N(A ).

It will suffice to show that this diagram satisfies the hypotheses of Corollary 3.5.7:

48
 (a) The ∞-categories N(Alg(A)o ) and Alg(N(Ao )) admit geometric realizations of simplicial objects.
In fact, both of these ∞-categories are presentable. For N(Alg(A)o ), this follows from Proposi-
tions T.A.3.7.6 and 1.6.15. For Alg(N(A)o ), we first observe that N(A)o is presentable (Proposi-
tion T.A.3.7.6) and that the tensor product preserves colimits separately in each variable, and apply
Proposition 1.5.15.
(b) The functors G and G0 admit left adjoints F and F 0 . The existence of a left adjoint to G follows from
the fact that G is determined by a right Quillen functor. The existence of a left adjoint to G0 follows
from Proposition 1.4.2.
(c) The functor G0 is conservative and preserves geometric realizations of simplicial objects. This follows
from Corollaries 1.5.11 and 1.5.4.
(d) The functor G is conservative and preserves geometric realizations of simplicial objects. The first
assertion is immediate from the definition of the weak equivalences in Alg(A), and the second follows
from Lemma 1.6.16.
(e) The canonical map G0 ◦ F 0 → G ◦ F is an equivalence of functors. This follows from the observation
that both sides induce, on the level of homotopy categories, the free algebra functor C 7→ n≥0 C ⊗n
`
(Proposition 1.4.2).

Remark 1.6.21. Proposition 1.6.5 admits a converse. Suppose that C is a presentable ∞-category endowed
with a monoidal structure, and that the associated bifunctor ⊗ : C × C → C preserves small colimits separately
in each variable. Then C is equivalent (as a monoidal ∞-category) to N(Ao ), where A is a combinatorial
simplicial model category, endowed with a compatible monoidal structure. Since we will not need this fact,
we will only give a sketch of proof.
First, we apply Example 1.6.20 to reduce to the case where C is a strict monoidal ∞-category; that is, C
is a simplicial monoid. Now choose a regular cardinal κ such that C is κ-accessible. Enlarging κ if necessary,
we may suppose that the full subcateogry Cκ ⊆ C spanned by the κ-compact objects contains the unit object
of C and is stable under tensor products.
The ∞-category Cκ is essentially small. We define a sequence of simplicial subsets

D(0) ⊆ D(1) ⊆ . . . ⊆ Cκ

as follows. Let D(0) = ∅, and for i ≥ 0 let D(i + 1) be a small simplicial subset S
of Cκ which is categorically
equivalent to C and contains the submonoid of C generated by D(i). Let D = D(i), so that D is a small
κ κ

simplicial submonoid of Cκ such that the inclusion D ⊆ Cκ is a categorical equivalence.

The proof of Theorem T.[Link] shows that C can be identified with an accessible localization of P(D). Ac-
cording to Proposition T.[Link], we can identify P(D) with N(Ao ), where A denotes the category (Set∆ )/ D
endowed with the contravariant model structure (see §T.2.1.4). Let L : P(D) → C be a localization functor,
and let B be the category (Set∆ )/ D endowed with the following localized model structure:
(C) A morphism α : X → Y in (Set∆ )/ D is a cofibration in B if and only if α is a monomorphism of
simplicial sets.

(W ) A morphism α : X → Y in (Set∆ )/ D is a weak equivalence in B if and only if the L(β) is an isomorphism

in the homotopy category hC, where β denotes the corresponding morphism in hP(D) ' hA.
(F ) A morphism α : X → Y in (Set∆ )/ D is a fibration in B if and only if it has the right lifting property
with respect to every morphism which is simultaneously a cofibration and a weak equivalence in B.

49
Proposition T.A.3.7.3 implies that B is a (combinatorial) simplicial model category, and that the underlying
∞-category N(Bo ) is equivalent to C.
The category (Set∆ )/ D is endowed with a monoidal structure, which may be described as follows: given
a finite collection of objects X1 , . . . , Xn ∈ S/ D , we let X1 ⊗ . . . ⊗ Xn denote the product X1 × . . . × Xn
of the underlying simplicial sets, mapping to D via the composition X1 × . . . × Xn → Dn → D, where the
second map is given by the monoid structure on D. It is not difficult to verify that this monoidal structure is
compatible with the model structure on B. Applying Proposition 1.6.5, we deduce that N(Bo,⊗ ) determines
a monoidal structure on N(Bo ) ' C. One can show that this monoidal structure coincides (up to equivalence)
with the structure determined by the associative multiplication on C.
Corollary 1.6.22. Let A be combinatorial simplicial model category. Assume that:
(i) The Cartesian product functor endows A with the structure of a monoidal model category, and that
this structure is compatible with the simplicial structure on A.
(ii) The model category A is left proper and the class of cofibrations is generated by cofibrations between
cofibrant objects.
(iii) The Cartesian product functor on A satisfies the monoid axiom.
op
Let G : Alg(A) → A∆ be the functor which assigns to each algebra object of A the corresponding monoid
object (see Proposition 1.2.14). Then:
(1) The functor G admits a left adjoint F .
op
(2) The functors F and G determine a Quillen adjunction between Alg(A) and A∆ , where the latter
category is endowed with the projective model structure.
op
(3) The right derived functor RG is fully faithful, and its essential image in A∆ consists of those sim-
plicial objects A• of A which determine monoid objects of the homotopy category hA.
Remark 1.6.23. Conditions (ii) and (iii) in the statement of Corollary 1.6.22 are automatic if we assume
that every object of A is cofibrant.
Proof. Assertion (1) follows from the adjoint functor theorem, since G is an accessible functor which com-
mutes with small limits. Assertion (2) is clear, since G preserves weak equivalences and fibrations. To prove
op
(3), it will suffice to show that the induced map of ∞-categories G0 : N Alg(A)o → N(A∆ )o (which is
0
well-defined up to homotopy) is fully faithful, and that the essential image of G consists of those fibrant-
cofibrant simplicial objects A• of A which determine monoid objects in the homotopy category hA. We have
a homotopy commutative diagram

N Alg(A)o / N(A∆op )o

p q
 00 
G/
Alg(Ao ) Fun(N(∆op ), N(Ao ))
where p is the categorical equivalence of Theorem 1.6.17, q is the categorical equivalence of Proposition
T.[Link]. It will therefore suffice to show that G00 is fully faithful, and that its essential image is the class
of monoid objects of N(Ao ). This follows immediately from Proposition 1.2.14.

1.7 Digression: Segal Monoidal ∞-Categories

In this section, we will introduce the definition of a Segal monoidal ∞-category. The theory of a Segal
monoidal ∞-categories is equivalent to the theory of monoidal ∞-categories which we use throughout this
paper (see Remark 1.7.11). However, the formalism of Segal monoidal ∞-categories is better suited to the
technical arguments we use in §1.4 to construct free algebras. The results of this section will be used only
in §1.4, and may be skipped without loss of continuity.

50
Notation 1.7.1. If J is a linearly ordered set, we let J∗ denote the set J ∪ {∗}, where ∗ is a new distinct
element. We do not regard this new element as in any way related to the ordering on J. For every nonnegative
◦
integer n, we let hni denote the linearly ordered set {1 < 2 < . . . < n}.
Definition 1.7.2. The category Lin∗ is defined as follows.
(1) The objects of Lin∗ are the sets hni, where n is a nonnegative integer.

(2) Given a pair of objects hmi, hni ∈ Lin∗ , we let HomLin∗ (hmi, hni) denote the collection of all maps
α : hmi → hni with the following properties:
(i) The map α satisfies α(∗) = ∗.
(ii) The restriction α|(hmi − α−1 {∗}) is a (nonstrictly) order-preserving map.
◦
Remark 1.7.3. We note that every finite linearly ordered set J is uniquely isomorphic to hni , where n is
the cardinality of J. We will typically abuse notation by referring to J∗ as an object of Lin∗ ; in this case,
we are implicitly identifying J∗ with hni.
Notation 1.7.4. We define a functor ψ : ∆op → Lin∗ as follows:

(1) For each n ≥ 0, we have ψ([n]) = hni.

(2) Given a morphism α : [n] → [m] in ∆, the associated morphism ψ(α) : hmi → hni is given by the
formula (
j if (∃j)[α(j − 1) < i ≤ α(j)]
ψ(α)(i) =
∗ otherwise.

More informally, ψ assigns to a nonempty linearly ordered set [n] the set of all “gaps” between adjacent
elements of [n].
Notation 1.7.5. If J is a finite linearly ordered set containing an element j, we let αj,J : J∗ → {j}∗ denote
the map defined by the formula (
j,J j if i = j
α (i) =
∗ otherwise.

Definition 1.7.6. Let C be an ∞-category. A Segal monoid object of C is a functor f : N(Lin∗ ) → C with
the following property:
(∗) For every finite linearly ordered set J, the collection of maps {f (αj,J )} exhibits f (J∗ ) as a product of
the objects {f ({j}∗ )}j∈J .
We let Mons (C) denote the full subcategory of Fun(N(Lin∗ ), C) spanned by the Segal monoid objects.
Proposition 1.7.7. Let C be an ∞-category. Composition with the functor ψ : ∆op → Lin∗ of Notation
1.7.4 induces an equivalence of ∞-categories q : Mons (C) → Mon(C).

We will give the proof of Proposition 1.7.7 at the end of this section.
Definition 1.7.8. A Segal monoidal category is a coCartesian fibration C⊗ → N(Lin∗ ) with the following
property: for every finite linearly ordered set J, the associated functors α!j,J : C⊗ ⊗
J∗ → C{j}∗ induce an
⊗ ⊗
equivalence CJ∗ ' j∈J C{j}∗ .
Q

Remark 1.7.9. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category. We will write C to indicate the fiber
C⊗ ×N(Lin∗ ) {h1i}. We will refer to C as the underlying ∞-category of C⊗ , and say that q : C⊗ → N(Lin∗ ) is
a Segal monoidal structure on C.

51
Remark 1.7.10. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category, and let ψ : ∆op → Lin∗ be
defined as in Notation 1.7.4. The fiber product C⊗ ×N(Lin∗ ) N(∆)op is a monoidal ∞-category in the sense
of Definition 1.1.2.
Remark 1.7.11. Let q : C⊗ → N(Lin∗ ) be a coCartesian fibration, classified by a functor f : N(Lin∗ ) →
Cat∞ . Then C⊗ is a Segal monoidal ∞-category if and only if f is a Segal monoid object of Cat∞ . Propo-
sition 1.7.7 implies that the pullback functor Mons (Cat∞ ) → Mon(Cat∞ ) is an equivalence of ∞-categories.
Consequently, up to equivalence, every monoidal ∞-category arises from a Segal monoidal ∞-category via
the construction of Remark 1.7.10.
Our final objective in this section is to study the analogue of the theory of algebras in the setting of Segal
monoidal ∞-categories.
Definition 1.7.12. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal structure on an ∞-category C = C⊗ h1i . A
⊗
Segal algebra object of C is a section A : N(Lin∗ ) → C of q with the following property: for every morphism
α : J∗ → J∗0 in Lin∗ which induces a bijection α−1 (J 0 ) → J 0 , the morphism A(α) is q-coCartesian. We let
Algs (C) denote the full subcategory of MapN(Lin∗ ) (N(Lin∗ ), C⊗ ) spanned by the Segal algebra objects.

Remark 1.7.13. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal ∞-category and let A : N(Lin∗ ) → C⊗ be a
section of q. Then A is a Segal algebra object if and only if, for every finite linearly ordered set J and every
element j ∈ J, the morphism A(αj,J ) is q-coCartesian (see Notation 1.7.5).
Remark 1.7.14. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal structure on an ∞-category C = C⊗ h1i and let
⊗
A : N(Lin∗ ) → C be a section of q. Using Remark 1.7.13, we deduce that A is a Segal algebra object of C if
and only if A ◦ ψ : N(∆)op → N(∆)op ×N(Lin∗ ) C⊗ is an algebra object of C, in the sense of Definition 1.1.14.
The following comparison result will be needed in §1.4:
Proposition 1.7.15. Let q : C⊗ → N(Lin∗ ) be a Segal monoidal structure on an ∞-category C. Composition
with the functor ψ : ∆op → Lin∗ of Notation 1.7.4 determines an equivalence of ∞-categories θ : Algs (C) →
Alg(C).
Proposition 1.7.7 should be regarded as a special case of Proposition 1.7.15, where we take C to be the
∞-category Cat∞ , equipped with the Cartesian Segal monoidal structure. Since we do not wish to press
any further into the theory, we will not make this precise; instead, we will give two separate (but nearly
identical) proofs for Propositions 1.7.15 and 1.7.7.
Proof of Proposition 1.7.7. It follows immediately from the definition that composition with the functor ψ
carries Mons (C) into Mon(C). We define a category I as follows:
(1) An object of I is either an object of ∆op or an object of Lin∗ .
(2) Morphisms in I are give by the formulas

HomI ([m], [n]) = Hom∆op ([m], [n]) HomI (hmi, hni) = HomLin∗ (hmi, hni)

MapI (hmi, [n]) = MapLin∗ (hmi, ψ([n])) MapI ([n], hmi) = ∅.

The map ψ extends to a retraction ψ : I → Lin∗ . Let Mon(C) denote the full subcategory of Fun(N(I), C)
consisting of those functors f : N(I) → C which possess the following properties:
(i) For each n ≥ 0, f carries the canonical map hni → [n] in I to an equivalence in C.
(ii) The restriction f | N(∆)op is a monoid object of C.
(ii0 ) The restriction f | N(Lin∗ ) is a Segal monoid object of C.

52
 We observe that if (i) is satisfied, then (ii) and (ii0 ) are equivalent to one another. Moreover, (i) is
equivalent to the assertion that f is a left Kan extension of f | N(Lin∗ ). Since every functor f0 : N(Lin∗ ) → C
admits a left Kan extension (given, for example, by f0 ◦ψ), Proposition T.[Link] implies that the restriction
map p : Mon(C) → Mons (C) is a trivial Kan fibration. The map q is the composition of a section to p (given
by composition with ψ) and the restriction map p0 : Mon(C) → Mon(C). It will therefore suffice to show
that p0 is a trivial fibration. In view of Proposition T.[Link], this will follow from the following pair of
assertions:

(a) Every f0 ∈ Mon(C) admits a right Kan extension f : N(I) → C.

(b) Given f : N(I) → C such that f0 = f | N(∆)op is a monoid object of C, f is a right Kan extension of f0
if and only if f satisfies condition (i) above.

To prove (a), we fix an object J∗ ∈ Lin∗ . Let J be the category (∆)op ×Lin∗ (Lin∗ )J∗ / , and let g denote
f0
the composition N(J) → N(∆)op → C . According to Lemma T.[Link], it will suffice to show that g admits
a limit in C (for every choice of J). The objects of J can be identified with morphisms α : J∗ → I∗ in
Lin∗ . Let J0 ⊆ J denote the full subcategory spanned by those objects for which every element of I has a
unique preimage under α. The inclusion J0 ⊆ J has a right adjoint, so that N(J0 )op → N(J)op is cofinal.
Consequently, it will suffice to show that g0 = g| N(J0 ) admits a limit in C.
Let J1 denote the full subcategory of J0 spanned by the morphisms αj,J : J∗ → {j}∗ . Using our
assumption that f0 is a monoid object of C, we deduce that g0 is a right Kan extension of g1 = g0 | N(J1 ).
In view of Lemma T.[Link], it will suffice to show that the map g1 has a limit in C. But this is clear; our
assumption that f0 is a monoid object of C guarantees that f exhibits f ([n]) as a limit of g1 . This proves (a).
Moreover, the proof shows that f is a right Kan extension of f0 at hni if and only if f induces an equivalence
f (hni) → f ([n]); this immediately implies (b) as well.
Proof of Proposition 1.7.15. We will essentially repeat the proof of Proposition 1.7.7. Let I be the category
introduced in the proof of Proposition 1.7.7, and observe again that the functor ψ : ∆op → Lin∗ extends
canonically to a retraction ψ : I → Lin∗ . Let Alg(C) denote the full subcategory of MapN(Lin∗ ) (N(I), C⊗ )
consisting of those functors f : N(I) → C⊗ such that q ◦ f = ψ and the following additional conditions are
satisfied:
(i) For each n ≥ 0, f carries the canonical map hni → [n] in I to an equivalence in C⊗ .
(ii) The restriction f | N(∆)op is an algebra object of C.

(ii0 ) The restriction f | N(Lin∗ ) is a Segal algebra object of C.

We observe that if (i) is satisfied, then (ii) and (ii0 ) are equivalent to one another. Moreover, (i) is
equivalent to the assertion that f is a q-Kan extension of f | N(Lin∗ ). Since every functor f0 : N(Lin∗ ) →
C admits a q-left Kan extension (given, for example, by f0 ◦ ψ), Proposition T.[Link] implies that the
restriction map p : Alg(C) → Algs (C) is a trivial Kan fibration. The map θ is the composition of a section
to p (given by composition with ψ) and the restriction map p0 : Alg(C) → Alg(C). It will therefore suffice to
show that p0 is a trivial fibration. In view of Proposition T.[Link], this will follow from the following pair
of assertions:

(a) Every f0 ∈ Alg(C) admits a q-right Kan extension f ∈ MapN(Lin∗ ) (N(I), C⊗ ).

(b) Given f ∈ MapN(Lin∗ ) (N(I), C⊗ ) such that f0 = f | N(∆)op is an algebra object of C, f is a q-right Kan
extension of f0 if and only if f satisfies condition (i) above.

To prove (a), we fix an object J∗ ∈ Lin∗ . Let J denote the category (∆)op ×Lin∗ (Lin∗ )J∗ / , and let g
f0
denote the composition N(J) → N(∆)op → C . According to Lemma T.[Link], it will suffice to show that g

53
admits a q-limit in C (for every choice of J). The objects of J can be identified with morphisms α : J∗ → I∗
in Lin∗ . Let J0 ⊆ J denote the full subcategory spanned by those objects for which every element of I has
a unique preimage under α. The inclusion J0 ⊆ J has a right adjoint, so that N(J0 )op → N(J)op is cofinal.
Consequently, it will suffice to show that g0 = g| N(J0 ) admits a q-limit in C.
Let J1 denote the full subcategory of J0 spanned by the morphisms αj,J : J∗ → {j}∗ . Using our
assumption that f0 is a monoid object of C, we deduce that g0 is a q-right Kan extension of g1 = g0 | N(J1 ).
In view of Lemma T.[Link], it will suffice to show that the map g1 has a q-limit in C. But this is clear; our
assumption that f0 is an algebra object of C guarantees that f exhibits f ([n]) as a limit of g1 . This proves
(a). Moreover, the proof shows that f is a q-right Kan extension of f0 at hni if and only if f induces an
equivalence f (hni) → f ([n]); this immediately implies (b) as well.

2 Modules
Let C be a monoidal category, and let A be an algebra object of C. Then we can consider left A-modules
in C: that is, objects M ∈ C equipped with a multiplication A ⊗ M → M satisfying suitable unit and
associativity conditions. The purpose of this section is to introduce an analogue of these ideas in the case
where C is replaced by a monoidal ∞-category. With an eye towards later applications, we will work in a
somewhat more general setting, where M is allowed to be an object of an ∞-category which is left-tensored
over C. We will introduce the formalism of tensored ∞-categories in §2.1. In particular, we will see that for
every ∞-category M which is left-tensored over C, and every algebra object A ∈ Alg(C), we can introduce
an ∞-category of (left) A-modules ModA (M). Moreover, every monoidal ∞-category C can be regarded as
(left) tensored over itself in a canonical way (Example 2.1.3).
In §2.3 we will study the ∞-categories ModA (M) in more detail. In particular, we will explain how to
compute limits and colimits in ModA (M). Our main results, Corollaries 2.3.5 and 2.3.7, assert that (in
favorable cases) limits and colimits in ModA (M) are detected by the forgetful functor θ : ModA (M) → M.
Our technique for computing colimits is based on a general result, Proposition 2.3.6, which can be used to
show that ModA (M) inherits many pleasant features of M.
The forgetful functor θ is generally not an equivalence, but it always admits a left adjoint. In §2.4 we
will construct this adjoint, and show that it can be described by the formula M 7→ A ⊗ M .
In §2.5, we will study the theory of module objects in the underlying ∞-category of a monoidal model
category A. In this case, we will show that our ∞-categorical theory of module objects of N(Ao ) is closely
related to the classical theory of (strictly associative) modules in A.
The theory of ∞-categories left-tensored over a given monoidal ∞-category C is really a special case of
the general theory of module objects. More precisely, we have already seen that C can be identified with
an algebra object of Cat∞ (Remark 1.2.15). In §2.6 we will extend this result by showing ∞-categories
left-tensored over C can be identified with C-module objects of Cat∞ . These identifications allow us to apply
our study of module objects to deduce results about tensored ∞-categories. For example, we can use this
method to show that C is freely generated (as an ∞-category tensored over C) by the unit object 1C ∈ C
(Corollary 2.6.7).
In §2.7 we will shift our focus somewhat; rather than studying the ∞-category ModA (M) over a fixed
algebra object A ∈ Alg(C), we will instead fix an object M ∈ M and study algebras which act on M . We will
show that algebra objects of C which act on M can be identified with algebra objects in a larger ∞-category
C[M ] (Proposition 2.7.6), and that in some cases one can construct a universal example of such an algebra,
which we will denote by End(M ).
For certain applications, it is convenient to work with nonunital algebras in a monoidal ∞-category C:
that is, objects A ∈ C equipped with a multiplication A ⊗ A → A which is coherently associative, but no
unit map 1C → A. In §2.2 we will give a precise definition for the category Algnu (C) of nonunital algebra
objects of C, and introduce an accompanying theory of nonunital modules. In §2.8, we will show that the
∞-categories Algnu (C) and Alg(C) are closely related: namely, Alg(C) is equivalent to a subcategory of
Algnu (C) (Theorem 2.8.1). In other words, if a nonunital algebra object of C admits a unit, then that unit

54
is essentially unique.

2.1 Tensored ∞-Categories

Let C be a monoidal ∞-category. Our first goal is to introduce the notion of an ∞-category M left-tensored
over C. Roughly speaking, this means we have an action ⊗ : C × M → M of C on M, which is unital and
associative up to coherent homotopy.
Definition 2.1.1. Let p : C⊗ → N(∆)op be a monoidal ∞-category. An ∞-category left-tensored over C⊗
is a categorical fibration q : M⊗ → C⊗ with the following properties:
(1) The composition (p ◦ q) : M⊗ → N(∆)op is a coCartesian fibration.
(2) The map q carries (p ◦ q)-coCartesian edges of M⊗ to p-coCartesian edges of C⊗ .
(3) For each n ≥ 0, the inclusion {n} ⊆ [n] induces an equivalence of ∞-categories M⊗ ⊗ ⊗
[n] → C[n] × M{n} .

Remark 2.1.2. Let q : M⊗ → C⊗ be as in Definition 2.1.1. We will refer to the fiber M = M⊗ [0] as the
⊗ ⊗ ⊗ ⊗
underlying ∞-category of M . Let C = C[1] be the underlying ∞-category of C . Then M[n] is equivalent to
the product Cn × M. The coCartesian fibration q induces functors

M⊗ ⊗ ⊗ ⊗
{0} ← M[1] → C[1] × M{1} ,

where the right map is an equivalence. We therefore obtain a bifunctor ⊗ : C × M → M, well-defined up to

homotopy. Moreover, the structure associated to M⊗ [n] for n > 1 ensures that the bifunctor ⊗ : C × M → M is
coherently associative; in particular, the homotopy category hC is tensored over hM in the sense of classical
category theory. We will generally abuse terminology by saying simply that M is left-tensored over C.

Example 2.1.3. Let C be a monoidal ∞-category. Then the tensor product ⊗ : C × C → C exhibits C as
left-tensored over itself. More precisely, suppose that the monoidal structure on C is given by a projection
p : C⊗ → N(∆)op . Let e : ∆ → ∆ denote the “shift” functor [n] 7→ [n] ? [0], where ? denotes the operation
of concatenating linearly ordered sets. Then e induces a functor N(∆)op → N(∆)op , which we will also
denote by e. The inclusion [n] ⊆ [n] ? [0] determines a natural transformation α : e → id, which we can
view as a functor from ∆1 × N(∆)op to N(∆)op . We now define a simplicial set C⊗,L equipped with a map
C⊗,L → N(∆)op via the formula

HomN(∆)op (K, C⊗,L ) = Hom0N(∆)op (∆1 × K, C⊗ )

α
where ∆1 ×K maps to N(∆)op via the composition ∆1 ×K → ∆1 ×N(∆)op → N(∆)op , and Hom0N(∆)op (∆1 ×
K, C⊗ ) ⊆ HomN(∆)op (∆1 × K, C⊗ ) is the subset consisting of those maps ∆1 × K → C⊗ which carry each
edge ∆1 × {k} to a p-coCartesian edge of C⊗ . Using Proposition T.[Link], we deduce that the projection
ψ q
C⊗,L → N(∆)op is a coCartesian fibration, and that the projections C⊗ L ←C
⊗,L
→ C⊗ preserve coCartesian
⊗
edges. Here CL denotes the pullback of p along the map e : N(∆) → N(∆) . It is easy to see that q is a
op op

categorical fibration (since the inclusion {1} × K ⊆ ∆1 × K is a cofibration of simplicial sets). Since ψ is an
equivalence, we deduce that C⊗,L[n] is canonically equivalent to the ∞-category C
n+1
. Using this equivalence,
we can easily verify that q satisfies condition (4) of Definition 2.1.1. Consequently, q : C⊗,L → C⊗ can be
viewed as an ∞-category left-tensored over C. The map ψ induces an equivalence C⊗,L ⊗
[0] → C[1] ' C. We will
summarize the situation by saying that C⊗,L exhibits C as left-tensored over itself.

Definition 2.1.4. Let p : C⊗ → N(∆)op be a monoidal ∞-category, and let q : M⊗ → C⊗ be an ∞-category

equipped with a left action of C⊗ . A module object of M⊗ is a functor F : N(∆)op → M⊗ with the following
properties:

55
 (1) The composition q ◦ F is an algebra object of C⊗ . In particular, p ◦ q ◦ F is the identity on N(∆)op .

(2) Let α : [m] → [n] be a convex map in ∆ such that α(m) = n. Then F (α) is a p ◦ q-coCartesian
morphism of M⊗ .
In this case, we will also abuse terminology by saying that F is a module object of the underlying ∞-category
M = M⊗ ⊗
[0] . We let Mod(M) denote the full subcategory of MapN(∆)op (N(∆) , M ) spanned by the module
op

objects of M.

Remark 2.1.5. There is an analogous notion of an ∞-category right-tensored over C⊗ , obtained by replacing
the inclusion {n} ⊆ [n] with the inclusion {0} ⊆ [n] in condition (∗) of Definition 2.1.1. Similarly, we have
a dual notion of module object in such a category. When necessary, we will distinguish these two notions by
referring to left modules and right modules, respectively.

Remark 2.1.6. Let C be a monoidal ∞-category, and let C⊗,L be defined as in Example 2.1.3. We will
abuse terminology by referring to module objects of C⊗,L as module objects of C. The ∞-category of module
objects of C⊗,L will be denoted by Mod(C), or ModL (C) when it is necessary to emphasize that we are
working with left modules. Note that this notation is not quite compatible with the notation introduced in
Definition 2.1.4, since C⊗,L
[0] is equivalent but not isomorphic to C.

Example 2.1.7. Let p : C⊗ → N(∆)op be a monoidal structure on an ∞-category C = C⊗ [1] . Let α :

∆1 × N(∆)op → N(∆)op be defined as in Example 2.1.3. Composition with α determines a section s of the
forgetful functor θ : Mod(C) → Alg(C). This section carries each A ∈ Alg(C) to a left A-module s(A). We
can view s(A) as A, equipped with the canonical left action on itself. For this reason, we will generally not
distinguish in our notation between A and s(A).
Remark 2.1.8. Let q : M⊗ → C⊗ be as in Definition 2.1.4. Then composition with q induces a functor
θ : Mod(M) → Alg(C), which we will refer to as the forgetful functor. If A is an algebra object of C, we
let ModA (M) denote the fiber θ−1 {A}. It is easy to see that θ is a categorical fibration of simplicial sets,
so that each ModA (M) is an ∞-category, which we will call the ∞-category of (left) A-modules in M. We
will soon see (Corollary 2.3.3) that θ is a Cartesian fibration, so that ModA (M) is an ∞-category which is
contravariantly functorial in A. In the case where C is acting on itself (as in Example 2.1.3), we will also
denote this ∞-category by ModA (C) or simply ModA , and refer to it as the ∞-category of (left) A-modules.
In classical category theory, one can define a similar notion of a category (left) tensored over a fixed
monoidal category C. However, this notion plays a secondary role to the theory of enriched categories (see,
for example, §T.A.1.4). For example, every category C can be regarded as enriched over the category Set of
sets. However, C is enriched and tensored over Set if and only if C admits arbitrary coproducts.
It is possible to modify Definition 2.1.1 to obtain the definition of an ∞-category enriched over a given
monoidal ∞-category C. This notion is useful in a variety of situations. For example, suppose we wanted to
develop the theory of (∞, n)-categories: that is, higher categories in which we allow noninvertible morphisms
of order ≤ n. One reasonable definition is inductive: one defines an (∞, n)-category to be an ∞-category
which is enriched over Cat(∞,n−1) , the ∞-category of (∞, n − 1)-categories (endowed with the Cartesian
monoidal structure). We will not pursue the theory of enriched ∞-categories in this paper, since the requisite
modifications to Definition 2.1.1 are somewhat cumbersome to work with in practice. However, it requires
very little additional effort to formulate the following strengthening of Definition 2.1.1:
Definition 2.1.9. Let C be a monoidal ∞-category, and let M be an ∞-category which is left-tensored over
C.

(1) Let M and N be objects of M. A morphism object for M and N is an object MorM (M, N ) ∈ C together
with a map α : MorM (M, N ) ⊗ M → N with the following universal property: for every object C ∈ C,
composition with α induces a homotopy equivalence MapC (C, MorM (M, N )) → MapM (C ⊗ M, N ).

56
 (2) Let M be an object of M and C an object of C. An exponential object is an object CM ∈ M together
with a map β : C ⊗C M → M with the following universal property: for every object N ∈ M,
composition with β induces a homotopy equivalence MapM (N,C M ) → MapM (C ⊗ N, M ).

We will say that M is tensored and enriched over C if, for every pair of objects M, N ∈ M, there exists a
morphism object Mor(M, N ) ∈ C. We will say that M is tensored and cotensored over C if, for every M ∈ M
and C ∈ C, there exists an exponential object CM ∈ M.
Remark 2.1.10. Let C and M be as in Definition 2.1.9. Suppose that M is enriched over C. Then the
morphism object MorM (M, N ) ∈ C can be constructed as functor of M and N . To see this, we observe
that we have a trifunctor (C × M)op × M → S, given by composing the tensor product ⊗ : C × M → M,
the Yoneda embedding j : Mop → Fun(M, S), and the evaluation map Fun(M, S) × M → S. We may
identify this map with a bifunctor e : Mop × M → Fun(Cop , S). If M is enriched over C, then the image of
e is contained in full subcategory Fun0 (Cop , S) ⊆ Fun(Cop , S) spanned by the essential image of the Yoneda
embedding j 0 : C → Fun0 (Cop , S). Composing e with a homotopy inverse to j 0 , we obtain the desired functor
MorM : Mop × M → C . Similarly, if M is cotensored over C, then the exponential CM can be regarded as a
functor Cop × M → M .
Example 2.1.11. Let C be a monoidal ∞-category, and regard C as left-tensored over itself (Example 2.1.3).
Then C is enriched over itself if and only if it is right closed, and cotensored over itself if and only if it is left
closed (Definition 1.6.7).
The following criterion provides a large supply of examples of enriched and cotensored ∞-categories:
Proposition 2.1.12. Let C be a monoidal ∞-category, and let M be an ∞-category which is left-tensored
over C. Suppose further that C and M are presentable.
(1) If, for every C ∈ C, the functor C ⊗ • : M → M preserves small colimits, then M is cotensored over C.
(2) If, for every M ∈ M, the functor • ⊗ M : C → M preserves small colimits, then M is enriched over C.
Proof. This follows immediately from the representability criterion of Proposition T.[Link].
We conclude this section with a few technical observations which will be needed later, when we need
q p
study tensored ∞-categories in detail. Let M⊗ → C⊗ → N(∆)op be as in Definition 2.1.1. Then p and (p ◦ q)
are coCartesian fibrations, but q is usually not a coCartesian fibration. Nevertheless, q shares some of the
pleasant features of coCartesian fibrations.
Lemma 2.1.13. Let C be a monoidal ∞-category, and let M be an ∞-category which is left tensored over
C. Then the associated functor p : M⊗ → C⊗ is a locally coCartesian fibration.
Proof. For each n ≥ 0, the map p[n] : M⊗ ⊗ ⊗ ⊗
[n] → C[n] is equivalent to the projection C[n] × M → C[n] , and
therefore a coCartesian fibration. The desired result now follows from Proposition T.[Link].
Remark 2.1.14. In the situation of Lemma 2.1.13, suppose we are given a morphism α : C → D in C⊗ ,
covering a map α : [m] → [n] in ∆. We can identify C with an n-tuple of objects (C1 , . . . , Cn ) ∈ Cn , D with
an m-tuple (D1 , . . . , Dm ) ∈ Cm , and α with a collection of morphisms Cα(i−1)+1 ⊗ . . . ⊗ Cα(i) → Di . The
fibers of the map p : M⊗ → C⊗ over the objects C and D are both canonically equivalent to M, and the
induced functor α! : M → M is given (up to equivalence) by the formula M 7→ Cα(m)+1 ⊗ . . . ⊗ Cn ⊗ M.

Remark 2.1.15. Let p : M⊗ → C⊗ be as in Lemma 2.1.13, and suppose given a commutative triangle
0
{= M DD
f { { DD g
{{ DD
{ DD
{{ "
M
h / M 00

57
in M⊗ , covering a triangle
[m]
} `AA
}} AA
}}α AA
} β AA
}~ }
[n] o γ
[k]

in the category ∆. Suppose furthermore that f and g are locally p-coCartesian, and that α induces a
bijection
{i ∈ [m] : β(k) < i} ' {j ∈ [n] : γ(k) < j ≤ α(m)}
(so that, in particular, β(k) < i ≤ m implies γ(k) < α(i)). Then the description given in Remark 2.1.14
implies that h is also locally p-coCartesian.
Lemma 2.1.16. Let p : M⊗ → C⊗ be as in Lemma 2.1.13, and let α be a morphism in M⊗ which covers a
map α : [m] → [n] in ∆. Then:
(1) Suppose that α(m) = n and α is locally p-Cartesian. Then α is p-Cartesian.
(2) Suppose that m ≤ n, that α : [m] → [n] is the canonical inclusion, and that α is locally p-coCartesian.
Then α is p-coCartesian.
Proof. Assertion (1) follows from Remark 2.1.15 and Lemma T.[Link], while (2) follows from Remark 2.1.15
and Lemma T.[Link].

2.2 Nonunital Algebras and Modules

Let A be an associative ring. A nonunital A-module is an abelian group M equipped with a bilinear
multiplication map A × M → M which satisfies the associativity formula a(bm) = (ab)m. This is slightly
weaker than the condition that M be an A-module, since we do not require the unit object 1 ∈ A to act
by the identity on M . For example, there is a trivial nonunital A-module structure on any abelian group
0
M , given by the zero map A × M → M. Of course, the category of A-modules can be identified with a full
subcategory of the category of nonunital A-modules. Our goal in this section is to formulate and prove an
∞-categorical analogue of this statement (Proposition 2.2.16). This result will play an essential role in our
study of the Barr-Beck theorem in §3.
We first observe that the notion of a nonunital A-module makes no reference to the identity element of
A; it therefore makes sense in the case where A is allowed to be a nonunital algebra. Our first goal in this
section is to set up a theory of nonunital algebras in the ∞-categorical context.
Definition 2.2.1. We define a subcategory ∆nu ⊆ ∆ as follows:
(1) The objects of ∆nu are the objects of ∆: that is, they are linearly ordered sets [n] where n ≥ 0.
(2) A morphism α : [m] → [n] of ∆ belongs to ∆nu if and only if α is an injective map of linearly ordered
sets.
Let q : C⊗ → N(∆)op be a monoidal structure on an ∞-category C = C⊗ [1] . A nonunital algebra object
⊗
of C is a functor A : N(∆ ) → C such that q ◦ A coincides with the inclusion N(∆nu )op ⊆ N(∆)op ,
nu op

and A carries convex morphisms in ∆nu to q-coCartesian morphisms in C⊗ . We let Algnu (C) denote the full
subcategory of MapN(∆)op (N(∆nu )op , C⊗ ) spanned by the nonunital algebra objects of C.
Warning 2.2.2. In [40], we denoted the category ∆nu by ∆s .
Remark 2.2.3. The definition of a nonunital algebra object of a monoidal ∞-category C makes sense in
the more general setting of nonunital monoidal ∞-categories. We will have no need for this additional level
of generality.

58
Remark 2.2.4. Let C be a monoidal ∞-category. Evaluation on the object [1] ∈ ∆nu defines a forgetful
functor Algnu (C) → C. We will generally abuse notation by not distinguishing between a nonunital algebra
A ∈ Algnu (C) and its image in C.
Remark 2.2.5. Let q : C⊗ → N(∆)op be a monoidal structure on an ∞-category C = C⊗ [1] , and let
⊗
A : N(∆) → C be a section of q. Then A is an algebra object of C if and only if A| N(∆nu )op is a
op

nonunital algebra object of C. Consequently, restriction to the subcategory ∆nu ⊆ ∆ defines a forgetful
functor θ : Alg(C) → Algnu (C). We will later show that the functor θ induces an equivalence of Alg(C) with
a subcategory of Algnu (C) (Theorem 2.8.1).
q p
Definition 2.2.6. Let M⊗ → C⊗ → N(∆)op exhibit M = M⊗ [0] as left-tensored over the monoidal ∞-
category C = C[1] . A nonunital module object of M is a functor M : N(∆nu )op → M⊗ with the following
⊗

properties:
(1) The composition q ◦ M is a nonunital algebra object of C.
(2) Let α : [m] → [n] be a convex map in ∆nu such that α(m) = n. Then M (α) is a (p ◦ q)-coCartesian
morphism of M⊗ .

We let Modnu (M) denote the full subcategory of MapN(∆)op (N(∆nu )op , M⊗ ) spanned by the nonunital mod-
ule objects. If A ∈ Algnu (C), then we let Modnu nu
A (M) denote the fiber Mod (M)×Algnu (M) {A}. If A ∈ Alg(C),
nu nu nu
then we let ModA (M) = Modθ(A) (M), where θ : Alg(C) → Alg (C) denotes the forgetful functor (see Re-
mark 2.2.5).
Remark 2.2.7. The terminology of Definition 2.2.6 is perhaps slightly confusing: remember that a module
object of M always refers to a pair (A, M ), where A is an algebra object of C and M is a module over A.
To obtain the notion of a nonunital module object, we allow A to be a nonunital algebra (and drop the
requirement that the unit of A, if it exists, acts by the identity on M ).
Remark 2.2.8. In the situation of Definition 2.2.6, evaluation at [0] ∈ ∆nu induces a forgetful functor
Modnu (M) → M. We will generally abuse notation by not distinguishing between a nonunital module object
M ∈ Modnu (M) and its image in M.
q p
Remark 2.2.9. Let M⊗ → C⊗ → N(∆)op exhibit M = M⊗ [0] as left-tensored over the monoidal ∞-category
⊗ ⊗
C = C[1] , and let M : N(∆) → M be a section of p ◦ q. Then M is a module object of M if and only
op

if M | N(∆nu )op is a nonunital module object of M. In particular, restriction induces a functor Mod(M) →
Modnu (M).
Definition 2.2.10. Let C be a monoidal ∞-category, let A be a nonunital algebra object of C, and let
u : 1C → A be an arbitrary map. We say that u is a left unit if the composition
u
A ' A ⊗ 1C → A ⊗ A → A

is homotopic to the identity in C. Similarly, we will say that u is a right unit if the composition
u
A ' 1C ⊗ A → A ⊗ A → A

is homotopic to the identity in C. We will say that u is a quasi-unit if it is both a left unit and a right unit. We
will say that A is quasi-unital if there exists a quasi-unit u : 1C → A. Let f : A → A0 be a morphism between
nonunital algebra objects of C. We will say that f is quasi-unital if there exists a quasi-unit u : 1C → A, and
the composition f ◦ u : 1C → A0 is a quasi-unit for A0 . We let Algqu (C) denote the subcategory of Algnu (C)
whose objects are quasi-unital algebra objects of C, and whose morphisms are quasi-unital maps of algebras.

59
Remark 2.2.11. Let A be a nonunital algebra object of a monoidal ∞-category C. Suppose that u : 1C → A
is a left unit and v : 1C → A is a right unit. Then the composition
u⊗v
1C ' 1C ⊗ 1C → A ⊗ A → A
is homotopic to both u and v, so that u and v are homotopic to one another. It follows that A is quasi-unital
if and only if A admits both a left and right unit. In this case, a quasi-unit u : 1C → A is uniquely determined
(up to homotopy).
Remark 2.2.12. We will see later that there is essentially no difference between algebras and quasi-unital
algebras. More precisely, we will show that for every monoidal ∞-category C, the forgetful functor Alg(C) →
Algqu (C) is a trivial Kan fibration (Theorem 2.8.1).
Definition 2.2.13. Let C be a monoidal ∞-category, let M be an ∞-category which is left-tensored over C,
and let A be a quasi-unital algebra object of C. We will say that an object M ∈ Modnu
A (M) is quasi-unital
if the composition
u
φ : M ' 1C ⊗ M → A ⊗ M → M
is homotopic to the identity in M, where u is a quasi-unit for A. We let Modqu
A (M) denote the full subcategory
of Modnu
A (M) spanned by the quasi-unital A-modules.

Remark 2.2.14. In view of Remark 2.2.11, the condition of Definition 2.2.13 does not depend on the choice
of a quasi-unit u : 1C → A.
Remark 2.2.15. In the situation of Definition 2.2.13, the condition that ψ be homotopic to the identity
is equivalent to the (apparently weaker) condition that ψ be an equivalence. For suppose that ψ is an
equivalence. Since the composition
u⊗u
1C ' 1C ⊗ 1C → A ⊗ A → A
is equivalent to u, we conclude that ψ 2 is homotopic to ψ (that is, ψ 2 and ψ belong to the same connected
component of MapM (M, M ) ). If ψ is invertible in the homotopy category hM, this forces ψ to be homotopic
to the identity.
Our goal in this section is to prove the following result:
Proposition 2.2.16. Let M be an ∞-category which is left-tensored over a monoidal ∞-category C, and let
A be an algebra object of C. The restriction map ModA (M) → Modnu A (M) induces a trivial Kan fibration
θ : ModA (M) → ModquA (M).

In other words, the theory of A-modules is equivalent to the theory of nonunital A-modules in which the
unit of A happens to act by the identity.
Proof. It is clear that θ is a categorical fibration. It will therefore suffice to show that θ is a categorical
equivalence.
We define a subcategory I ⊆ [1] × ∆op as follows:
(1) Every object of [1] × ∆op belongs to I.
(2) A morphism α : (i, [m]) → (j, [n]) in [1] × ∆op belongs to I if and only if either i = 0 or the map
[n] → [m] is injective.
For i = 0, 1, we let Ii denote the full subcategory of I spanned by the objects {(i, [n])}n≥0 .
Form a pullback diagram
N / M⊗

p
 
N(∆)op
A / C⊗ .

60
Let D denote the full subcategory of MapN(∆)op (N(I), N) spanned by those functors f : N(I) → N with the
following properties:
(i) The restriction f | N(I0 ) belongs to ModA (M).
(ii) The restriction f | N(I1 ) belongs to Modqu
A (M).

(iii) For each n ≥ 0, the induced map f (0, [n]) → f (1, [n]) is an equivalence in N[n] ' M.
We observe that, if (iii) is satisfied, then (i) and (ii) are equivalent. For each n ≥ 0, the cat-
egory I0 ×I I/(1,[n]) has a final object, given by the map (0, [n]) → (1, [n]). It follows that a functor
f ∈ MapN(∆)op (N(I), N) satisfies (iii) if and only if f is a p-left Kan extension of f | N(I0 ). Invoking
Proposition T.[Link], we deduce that the restriction map r0 : D → ModA (M) is a trivial Kan fibration
onto its essential image. Moreover, r0 has a section s, given by composition with the projection I → ∆op .
Consequently, r0 is a trivial Kan fibration, so s is a categorical equivalence. The restriction map θ factors
s r1
as a composition ModA (M) → D → Modqu
A (M). Consequently, it will suffice to show that r1 is a categorical
equivalence. In view of Proposition T.[Link], it will suffice to prove the following:

(a) Every f1 ∈ Modqu

A (M) admits a p-right Kan extension f ∈ MapN(∆)op (N(I), N).

(b) Let f ∈ MapN(∆)op (N(I), N) be an arbitrary extension of f1 = f | N(I1 ) ∈ Modqu

A (M). Then f is a
p-right Kan extension of f1 if and only if f satisfies condition (iii).

In what follows, let us fix a quasi-unital module f1 ∈ Modqu A (M) and an integer n ≥ 0. Let J =
I1 ×I I(0,[n])/ , and let g : N(J) → N be the composition of f1 with the projection J → N(I1 ). Let v ∈ J
denote the object corresponding to the morphism (0, [n]) → (1, [n]). In view of Lemma T.[Link], it will
suffice to prove the following assertions (for every choice of f1 , n ≥ 0):

(a0 ) There exists a p-limit diagram g : N(J). → N rendering the following diagram commutative:
g
/
N(J)
_ t: N
t
g
t p
t
 t 
N(J). / N(∆)op .

(b0 ) Given an arbitrary map g which renders the above diagram commutative, g is a p-limit diagram if and
only if g carries {v}. ⊆ N(J). to an equivalence in N[n] ' M.

Let J0 denote the full subcategory of J spanned by those maps α : (0, [n]) → (1, [m]) for which the image
of the induced map [m] → [n] contains n. We claim that the inclusion N(J0 )op ⊆ N(J)op is cofinal. In view
of Theorem T.[Link], it will suffice to show that, for every object α ∈ J, the category J0 ×J J/α has weakly
contractible nerve. If α ∈ J0 , this is obvious. Suppose that α classifies a map γ : [m] → [n]. Then J0 ×J J/α
can be identified with a product of categories {Eop i }0≤i≤n , where
(
(∆nu
+ )γ −1 (i)/ if i < n
Ei ' nu
∆ if i = n.

The categories Ei have initial objects for i < n, and En has weakly contractible nerve (because the in-
clusion N(∆nu )op ⊆ N(∆)op is cofinal (Lemma T.[Link]), cofinal maps are weak homotopy equivalences
(Proposition T.[Link]), and N(∆)op is weakly contractible (Lemma T.[Link] and Proposition T.[Link])),
so 0≤i≤n N(Ei )op is likewise weakly contractible.
Q
Let g0 = g| N(J0 ). In view of the cofinality statement above, (a0 ) and (b0 ) are equivalent to the following
assertions:

61
(a00 ) There exists a p-limit diagram g 0 : N(J0 ). → N rendering the following diagram commutative:
g0
/
N(J 0 )
_ s s9 N
g0
s s p
 ss 
N(J0 ). / N(∆)op .

(b00 ) Given an arbitrary map g 0 which renders the above diagram commutative, g 0 is a p-limit diagram if
and only if g 0 carries {v}. ⊆ N(J0 ). to an equivalence in N[n] ' M.
We now observe that, for every morphism α : [m] → [n] in ∆ for which α(m) = n classifying a map
∆1 → N(∆)op the pullback N ×N(∆)op ∆1 is equivalent to a product M ×∆1 . It follows that for every object
N ∈ N[m] , there exists a locally p-Cartesian morphism α : N 0 → N in N covering α. Lemma 2.1.16 implies
that α is p-Cartesian.
Let h1 : N(J0 ) → N(∆)op denote the composition N(J0 ) → N(I) → N(∆)op , so that we have a natural
transformation h : ∆1 × N(J0 ) → N(∆)op from h0 = h|{0} × N(J0 ) to h1 , where h0 is the constant functor
taking the value [n]. For each object x ∈ N(J0 ). , the restriction of h to ∆1 × {x} classifies a morphism
α : [m] → [n] satisfying α(m) = n. It follows that we can lift h to a p-Cartesian transformation e h :
∆1 × N(J0 ) → N with e h|{1} × N(J0 ) = g0 . Let g00 = e
h|{0} × N(J0 ). Using Proposition T.[Link], we obtain
the following reformulations of (a00 ) and (b00 ):
(a000 ) There exists a p-limit diagram g 00 : N(J0 ). → N rendering the following diagram commutative:
g00
/
N(J 0 )
_ s s9 N
s
g 00
p
s
 s s h0 
N(J0 ). / N(∆)op .

(b000 ) Given an arbitrary map g 0 which renders the above diagram commutative, g 00 is a p-limit diagram if
and only if g 00 carries {v}. ⊆ N(J0 ). to an equivalence in N[n] ' M.
We now prove (a000 ). Let us first regard g00 as a functor from N(J0 ) to N[n] ' M. Let M = f1 ([0]) ∈ M.
Unwinding the definitions, we see that the values assumed by g00 can be identified with M , and the morphisms
between these values are given by iterated multiplication by the unit 1C → A. Since f1 is a quasi-unital
module, it follows that g00 carries every morphism in J0 toQan equivalence in N[n] . The simplicial set N(J0 ) is
weakly contractible, since it is isomorphic to the product ( 0≤i<n N(∆nu op nu op
+ ) )×N(∆ ) . Applying Corollary
T.[Link], we deduce that g00 admits a colimit g 00 : N(J0 ). → N[n] , and that g 00 carries {v}. to an equivalence
in N[n] . Since the ∞-category M⊗ ⊗
[n] is equivalent to the product C[n] × M, we conclude that g 0 is a q[n] -limit
0

⊗ ⊗
diagram, where q[n] : M[n] → C[n] denotes the production. Applying Corollary T.[Link] to the diagram

M⊗ I
q
/ C⊗
II vv
II v
II vv
II vvv
$ {v
N(∆)op ,
we conclude that g 00 is a q-limit diagram. Since p is a pullback of q, we deduce that g 00 is a p-limit diagram.
This proves (a000 ). Moreover, the uniqueness of p-limit diagrams implies the “only if” direction of (b000 ).
It remains only to prove the “if” direction of (b000 ). Let ge00 : N(J0 )/ → N[n] be an arbitrary extension of
g0 such that ge00 carries {v}/ to an equivalence. Since g 00 is a limit of g00 , there exists a natural transformation
0

γ : ge00 → g 00 which is an equivalence when restricted to N(J0 ). Since both ge00 and g 00 carry {v}/ to an
equivalence in N[n] , we conclude that γ is also an equivalence at the cone point of N(J0 )/ . It follows that ge00
is equivalent to g 00 , so that ge00 is also a p-limit diagram as desired.

62
2.3 Limits and Colimits of Modules
Let C be a monoidal ∞-category, M an ∞-category which is left-tensored over C, and A an algebra object
of C. Our goal in this section is to construct limits and colimits in the ∞-category ModA (M).
We begin with the study of limits in ModA (M). In fact, we will work a little bit more generally, and
consider relative limits with respect to the forgetful functor Mod(M) → Alg(C) (we refer the reader to §T.4.3
for a discussion of relative limits in general). Our basic result is Proposition 2.3.2, which assert that θ admits
relative limits provided that the corresponding limits exist in M. We begin with a somewhat technical lemma,
whose proof we defer until the end of this section.
Lemma 2.3.1. Let p : C → D be a categorical fibration of ∞-categories, let E and K be simplicial sets, and
suppose given a diagram
K × _ E
f
/
v; C
v
f
v p
v
 v 
K/ × E / D.

Suppose further that for each vertex E of E, there exists an extension f E : K / → C of fE which is compatible
with the above diagram and is a p-limit. Then:
(1) There exists a map f : K / × E → C rendering the above diagram commutative, with the property that
for each vertex E of E, the induced map f E : K / → C is a p-limit diagram.

(2) Let f : K / × E → C be an arbitrary map which renders the above diagram commutative. Then f
satisfies the condition of (1) if and only if the adjoint map K / → Fun(E, C) is a pE -limit diagram,
where pE : Fun(E, C) → Fun(E, D) is given by composition with p.
Proposition 2.3.2. Let C be a monoidal ∞-category, and let M be an ∞-category which is left-tensored
over C. Let K be a simplicial set such that M admits K-indexed limits, and let θ : Mod(M) → Alg(C) be the
forgetful functor. Then:

(1) For every diagram

K _ / Mod(M)
v:
v
v θ
v 
 v
K/ / Alg(C)

there exists a dotted arrow as indicated, which is a θ-limit diagram.

(2) An arbitrary map g : K / → Mod(M) is a θ-limit diagram if and only if the induced map K / → M is
a limit diagram.

Proof. Let n ≥ 0. We first observe that the equivalence M⊗ ⊗ ⊗

[n] ' C[n] × M{n} implies the following:

(10 ) For every diagram

K _ / M⊗
[n]
}>
}
} q[n]
} 
 }
K/ / C⊗
[n]

there exists a dotted arrow as indicated, which is a q[n] -limit diagram.

63
 (20 ) An arbitrary diagram K / → M⊗
[n] is a q[n] -limit diagram if and only the composition

K / → M⊗ ⊗
[n] → M{n} ' M

is a limit diagram.
Combining this observation with Corollary T.[Link], we deduce:

(100 ) For every diagram

K _ / M⊗
[n]
}>
}
} q[n]
} 
 }
K/ / C⊗
[n]

there exists a dotted arrow as indicated, which is a q-limit diagram.

(200 ) An arbitrary diagram K / → M⊗
[n] is a q-limit diagram if and only the composition

K / → M⊗ ⊗
[n] → M{n} ' M

is a limit diagram.
Let Mod0 (M) be the full subcategory of MapN(∆)op (N(∆)op , M⊗ ) spanned by those objects which project
to algebra objects of C. Combining (100 ), (200 ), and Lemma 2.3.1, we deduce:

(1000 ) For every diagram

K _ / Mod0 (M)
v:
v
v 0
θ
v 
 v
K/ / Alg(C)

there exists a dotted arrow as indicated, which is a θ0 -limit diagram.

(2000 ) An arbitrary map p : K / → Mod0 (M) is a θ0 -limit diagram if and only if, for every n ≥ 0, the
composition K / → M⊗ ⊗
[n] → M{n} ' M is a limit diagram.

To deduce (1) from these assertions, it will suffice to show that if if g : K . → Mod0 (M) satisfies (2000 ), and
g = g|K factors through Mod(M), then g factors through Mod(M). Let f : [m] → [n] be a convex morphism
in ∆ such that f (m) = n. Then f induces a natural transformation g [n] → g [m] of functors K / → M⊗ . We
wish to show that this natural transformation is (p ◦ q)-coCartesian. Since f (m) = n, we have a homotopy
commutative diagram
M⊗ / M⊗
[n] [m]
BB {
BBα {
BB {{
BB {{ β
! }{{
M,

and it suffices to show that the associated transformation t : α ◦ g [n] → β ◦ g [m] is an equivalence. Our
hypothesis implies that t restricts to an equivalence t : α ◦ g[n] → β ◦ g[m] . Since g satisfies (200 ), the maps
α ◦ g [n] and β ◦ g [m] are both limit diagrams in M. It follows that t is an equivalence as well, as desired.
We now complete the proof by observing that if g : K / → Mod0 (M) factors through Mod(M), then the
criteria of (2) and (2000 ) are equivalent.

64
Corollary 2.3.3. Let C be a monoidal ∞-category, M an ∞-category which is left-tensored over C, and
θ : Mod(M) → Alg(C) the forgetful functor. Then θ is a Cartesian fibration. Moreover, a morphism f in
Mod(M) is θ-Cartesian if and only if the image of f in M is an equivalence.
Proof. Apply Proposition 2.3.2 in the case K = ∆0 .
Remark 2.3.4. Proposition 2.3.2 has an analogue for nonunital modules, with exactly the same proof. In
particular, we obtain the following nonunital version of Corollary 2.3.3:
(∗) The forgetful functor θ : Modnu (M) → Algnu (C) is a Cartesian fibration. Moreover, a morphism
f : M → N of nonunital module objects is θ-Cartesian if and only if its image in M is an equivalence.
Corollary 2.3.5. Let C be a monoidal ∞-category, M an ∞-category which is left-tensored over C, and
θ : Mod(M) → Alg(C) the forgetful functor. Let A be an algebra object of C. Suppose that M admits
K-indexed limits. Then:
(1) The ∞-category ModA (M) admits K-indexed limits.
(2) A map p : K / → ModA (M) is a limit diagram if and only if the induced map K / → M is a limit
diagram.
(3) Given a morphism φ : B → A of algebra objects of A, the induced functor ModA (M) → ModB (M)
preserves K-indexed limits.
We now turn to the problem of constructing colimits in ∞-categories of modules. We begin with the
following very general principle:
Proposition 2.3.6. Let A ⊆ Cat d∞ be a subcategory of the ∞-category of (not necessarily small) ∞-
categories. Assume that A has the following properties:
(a) The ∞-category A admits small limits, and the inclusion A ⊆ Cat
d∞ preserves small limits.

(b) If X belongs to A, then Fun(∆1 , X) belongs to A.

(c) If X and Y belong to C, then a functor X → Fun(∆1 , Y ) is a morphism in A if and only if, for every
vertex v of ∆1 , the composite functor X → Fun(∆1 , Y ) → Fun({v}, Y ) ' Y is a morphism of A.
Let C be a monoidal ∞-category, A an algebra object of C, M an ∞-category which is left tensored over C.
Suppose M is an object of A, and that the functor A ⊗ • : M → M is a morphism of A. Then:
(1) The ∞-category ModA (M) is an object of A.
(2) For every ∞-category N belonging to A, a functor N → ModA (M) is a morphism in A if and only if
the composite functor N → ModA (M) → M is a morphism in A.
In particular, the forgetful functor ModA (M) → M is a morphism in A.
Proof. Let p : M⊗ → C⊗ exhibit M as left-tensored over C. Form a pullback diagram
X / M⊗

p0 p
 
N(∆)op
A / C⊗ .

We observe that p0 is a locally coCartesian fibration (Lemma 2.1.13), each fiber of p0 is equivalent to M, and
each of the associated functors can be identified with an iterate of the functor A ⊗ • : M → M. Assertion
(1) is now an immediate consequence of Proposition T.[Link].
Now suppose that f : N → ModA (M) is as in (2). Proposition T.[Link] implies that f is a morphism
of A if and only if, for every n ≥ 0, the composite map N → ModA (M) → X[n] belongs to A. We complete
the proof by observing that each of the functors ModA (M) → X[n] is equivalent to the forgetful functor
ModA (M) → M.

65
Corollary 2.3.7. Let C be a monoidal ∞-category, let M be an ∞-category which is left-tensored over C.
Let K be a simplicial set such that M admits K-indexed colimits. Let A ∈ Alg(C), and suppose that the
functor A ⊗ • : M → M preserves K-indexed colimits. Then:
(1) Every diagram f : K → ModA (M) has a colimit.
(2) An arbitrary diagram f : K . → ModA (M) is a colimit diagram if and only if it induces a colimit
diagram K . → M.
Proof. Let A ⊆ Cat
d∞ be the subcategory whose objects are ∞-categories which admit K-indexed colimits,
and whose morphisms are functors which preserve K-indexed colimits. Applying Proposition 2.3.6, we
conclude that ModA (M) belongs to A, and that the forgetful functor ModA (M) → M is a morphism of
A. This proves (1), and the “only if” direction of (2). The “if” direction then follows formally, since the
forgetful functor ModA (M) → M is conservative.

Corollary 2.3.8. Let C be an ∞-category equipped with a monoidal structure and let M be left-tensored
over C. Suppose that M is presentable and that, for each C ∈ C, the functor C ⊗ • : M → M preserves small
colimits. Then:
(1) For every A ∈ Alg(C), the ∞-category ModA (M) is presentable.
(2) For every morphism A → B of algebra objects of C, the associated functor ModB (M) → ModA (M)
preserves small limits and colimits.
(3) The forgetful functor θ : Mod(M) → Alg(C) is a presentable fibration (Definition T.[Link]).
Proof. Assertion (1) is a special case of Proposition 2.3.6. Corollary 2.3.3 implies that the diagram

ModB (M) / ModA (M)

III
III uu
III uuuuu
I$ uz uu
M
commutes up to homotopy. Assertion (2) follows immediately from Corollaries 2.3.5 and 2.3.7. Assertion
(3) follows from (1), Corollaries 2.3.3 and 2.3.5, and Proposition T.[Link].
Remark 2.3.9. Under the hypotheses of Corollary 2.3.8, if A → B is a morphism of algebra objects of
C. then the forgetful functor ψ : ModB (M) → ModA (M) admits both left and right adjoints (Corollary
T.[Link]). In §4.5 we will prove the existence of a left adjoint to ψ under much weaker assumptions (Lemma
4.5.12).
We conclude this section by giving a proof of Lemma 2.3.1.
Lemma 2.3.10. Let p : D → C be a right fibration of ∞-categories, let C0 ⊆ C be a full subcategory, and
let D0 = C0 ×C D. Let q : X → S be a categorical fibration of simplicial sets, and let F : C → X be a map
which is a q-left Kan extension of F | C0 . Then F ◦ p is a q-left Kan extension of F ◦ p| D0 .
Proof. Let D be an object of D, C = p(D), and define
C0/C = C0 ×C C/C D0/D = D0 ×D D/D .
We wish to show that the composition
F
(D0/D ). (C0/C ). → C → X

is a q-colimit diagram. Since F is a q-left Kan extension of F | C0 , it will suffice to show that the map
φ0 : D0/D → C0/C is a trivial Kan fibration. The map φ0 is a pullback of the map φ : D/D → C/C , which is a
trivial Kan fibration by Proposition T.[Link].

66
Lemma 2.3.11. Let p : X → S and q : Y → Z be maps of simplicial sets. Assume that q is a categorical
fibration, and that p is either a Cartesian fibration or a coCartesian fibration.
Define new simplicial sets Y 0 and Z 0 equipped with maps Y 0 → S, Z 0 → S via the formulas

HomS (K, Y 0 ) ' Hom(X ×S K, Y )

HomS (K, Z 0 ) ' Hom(X ×S K, Z).

Let C0 be an ∞-category equipped with a functor f : C0 → Y 0 , and let C be a full subcategory of C0 .
Then:

(1) Composition with q determines a categorical fibration q 0 : Y 0 → Z 0 .

(2) Let F : X ×S C0 → Y be the map classified by f , and suppose that F is a q-left Kan extension of
F |X ×S C. Then f is a q 0 -left Kan extension of f | C.
Proof. We first prove (1). We wish to show that q 0 has the right lifting property with respect to every
inclusion i : A → B of simplicial sets which is a categorical equivalence. For this, it suffices to show that q
has the right lifting property with respect to every inclusion of the form i0 : X ×S A → X ×S B. Since q is
a categorical fibration, it suffices to prove that i0 is a categorical equivalence. This follows from Proposition
T.[Link].
Let C be an object of C0 , and let C/C denote the fiber product C ×C0 C0/C . We wish to show that the
composition
f
C./C → C0 → Y 0
is a q 0 -colimit diagram. Replacing C ⊆ C0 by the inclusion C/C ⊆ C./C (and applying Lemma 2.3.10), we can
reduce to the case C0 = C. . Replacing C by a minimal model if necessary (Proposition T.[Link]), we may
assume that C is minimal. Similarly, we may assume that p is a minimal inner fibration.
Let n > 0, and suppose we are given a diagram

f0
C ? ∂ ∆ n / Y0
_ v:
v
v q0
v 
 v
C ?∆n
g
/ Z 0,

where f 0 | C ?{0} coincides with f . We wish to show that there exists a dotted arrow, as indicated in the
diagram. Composing g with the map Z 0 → S, we obtain a map C ?∆n → S. Let D denote the fiber product
X ×S (C ?∆n ), and let D0 = X ×S (C ? ∂ ∆n ). Since C is minimal, it is strictly unital (Example C.2.4.5), so
that C ?∆n is strictly unital. Similarly, the minimal fibration X → S is strictly unital (Example C.2.4.6),
so that the map D → C ?∆n is strictly unital. It follows that D is a strictly unital ∞-category (Remarks
C.2.4.3 and C.2.4.4). Unwinding the definitions, we are reduced to solving a lifting problem depicted in the
diagram
F0 /
D0 _ Y
q
 
D
G / Z.

Let D0 denote the inverse image of ∆n in D. For every simplicial subset W ⊆ D0 , we let D(W ) denote
the collection of all simplices of D whose intersection with D0 belongs to W . Let P denote the partially
ordered collection of pairs (W, FW ), where W ⊆ D0 contains the inverse image of ∂ ∆n , and FW : D(W ) → Y

67
satisfies FW | D0 = F 0 , and FW fits into a commutative diagram

D(W
FW
/Y
_ )
q
 
D
G / Z.

Applying Zorn’s lemma, we deduce that P has a maximal element. To complete the proof, it will suffice
to show that W = D0 . Assume otherwise, and let σ be a simplex of D0 of minimal dimension which does not
belong to W . We note that σ surjects onto ∆n , so the dimension of σ is necessarily positive. Let E denote
the fiber product D/σ ×C ?∆n C. Using Proposition C.2.4.7, we deduce the existence of a pushout diagram

E ? ∂ _ σ / D(W )

 
E ?σ / D(W 0 ).

In view of the maximality of (W, FW ), it follows that FW does not admit an extension to D(W 0 ). In other
words, there is no solution to the associated lifting problem

E ? ∂ _ σ
h /
w; Y
w
w q
w
 w 
E ?σ / Z.

Let h0 denote the restriction of h to E ?{v}, where v is the initial vertex of σ. Then h0 is not a q-colimit
diagram. However, h0 can be written as a composition
h00
E ?{v} → (D/v ×C ?∆n C). → Y,

where h00 is a q-colimit diagram. The map E → D/v ×C ?∆n C is a pullback of the trivial fibration D/σ → D/v ,
and therefore cofinal; we therefore obtain a contradiction and the proof is complete.
Proof of Lemma 2.3.1. Without loss of generality, we may suppose that E and K are ∞-categories. Let ∞
denote the cone point of K / . For each object E ∈ E, the inclusion K × {idE } ⊆ K × EE/ ' (K × E)(∞,E)/
is left anodyne. Consequently, f satisfies (1) if and only if f is a p-right Kan extension of f 0 . The existence
of f follows from Lemma T.[Link].
The “only if” direction of (2) follows immediately from Lemma 2.3.11. The converse follows from the
uniqueness of limits (up to equivalence).

2.4 Free Modules

Let C be a monoidal ∞-category, and let A be an algebra object of C. Our goal in this section is to construct
a left adjoint to the forgetful functor ModA (C) → C. In fact, we will treat a slightly more general problem,
by allowing modules taking values in an arbitrary ∞-category left tensored over C.
Definition 2.4.1. Let C be a monoidal ∞-category, let M be an ∞-category which is left-tensored over
C. Let A ∈ Alg(C), let θ : ModA (M) → M denote the forgetful functor, and let M0 ∈ M. A free A-
module generated by M0 is an object M ∈ ModA (M) together with a map α : M0 → θ(M ) with the
following universal property: for every N ∈ ModA (M), composition with α induces a homotopy equivalence
MapModA (M) (M, N ) → MapM (M0 , θ(N )).

68
 The goal of this section is to prove the following result:

Proposition 2.4.2. Let C be a monoidal ∞-category, and let M be an ∞-category which is left-tensored
over C. Let A ∈ Alg(C). Then:
(1) The forgetful functor θ : ModA (M) → M admits a left adjoint.
(2) For every M0 ∈ M, there exists a free A-module M generated by M0 .

(3) An arbitrary map M0 → θ(N ) exhibits N as a free A-module generated by M0 if and only if the
composition
A ⊗ M0 → A ⊗ θ(N ) → θ(N )
is an equivalence in M.
Corollary 2.4.3. Let C be a monoidal ∞-category, 1C the unit object of C, A an algebra object of C,
θ : ModA (C) → C the forgetful functor, and M ∈ ModA (C). Then composition with the unit map of A induces
a homotopy equivalence MapModA (C) (A, M ) → MapC (1C , θ(M )). Here we identify A with the corresponding
left A-module (Example 2.1.7).
Proof. Proposition 2.4.2 implies that A is freely generated by the unit map 1C → A as a left A-module.
Corollary 2.4.4. Let C be an ∞-category equipped with a monoidal structure, and let M be an ∞-category
which is left-tensored over C. Let A be an algebra object of Alg(C) such that the unit map 1C → A is an
equivalence in C. Then the forgetful functor θ : ModA (M) → M is an equivalence of ∞-categories.

Proof. Let F be the left adjoint to θ supplied by Proposition 2.4.2, and let

u : idM → θ ◦ F, v : F ◦ θ → idM

be a compatible unit and counit for the adjunction. We wish to prove that u and v are equivalences of
functors.
We first consider the functor u. Proposition 2.4.2 implies that the composition θ ◦ F can be identified
with the functor M 7→ A ⊗ M . The unit map u is given by tensor product with the unit map u0 : 1C → A
of the algebra A. By hypothesis, u0 is an equivalence in C, so that u is an equivalence in Fun(M, M).
We now show that v is an equivalence. Since the functor θ detects equivalences (Corollary 2.3.3), it will
suffice to show that the induced transformation α : θ ◦ F ◦ θ → θ is an equivalence of functors. We now
observe that u provides a right inverse to this α. Since u is an equivalence, we conclude also that α is an
equivalence.
p q
Proof of Proposition 2.4.2. Let M⊗ → C⊗ → N(∆)op realize M as left-tensored over C. Let I be the
subcategory of [1] × ∆op defined as follows:

(a) Every object of [1] × ∆op belongs to I.

(b) Given 0 ≤ i ≤ j ≤ 1 and a map α : [m] → [n] in ∆, the corresponding morphism (i, [n]) → (j, [m])
belongs to I if and only if either j = 1 or α(m) = n.
We let I0 ⊆ ∆op and I1 ' ∆op denote the inverse images in of the objects 0, 1 ∈ [1]. We regard N(I) as
A
an object of (Set∆ )/ C⊗ via the composition N(I) ⊆ N([1] × ∆op ) → N(∆)op → C⊗ . Let N denote the fiber
product N(I) ×C⊗ M⊗ , and let p0 : N → N(I) be the projection. Lemma 2.1.13 implies that p0 is a locally
coCartesian fibration. Let D denote the full subcategory of MapN(I) (N(I), N) spanned by those functors
f : N(I) → M⊗ with the following properties:
(i) The restriction f | N(I0 ) carries each morphism in I0 to a p0 -coCartesian edge of N.

69
 (ii) The restriction f | N(I1 ) is a module object of M⊗ .
Let D0 be the full subcategory of D spanned by those functors f which satisfy the following additional
condition:
(iii0 ) The functor f is a p0 -left Kan extension of f | N(I0 ).
Let D1 be the full subcategory of D spanned by those functors which satisfy the following additional condition:
(iii1 ) The functor f is a p0 -right Kan extension of f | N(I1 ).
We observe that every for every n ≥ 0, the ∞-category N(I1 )(0,[n])/ has a final object (1, [n]), and the
∞-category N(I0 )/(1,[n]) has an initial object (0, [n] ? [0]). Using Lemma T.[Link], we deduce:
(∗) Every functor f0 ∈ MapN(I) (N(I0 ), N) admits a p0 -left Kan extension f ∈ MapN(I) (N(I), N). Moreover,
an arbitrary extension f 0 is a p0 -left Kan extension if and only if p0 carries each morphism (0, [n]?[0]) →
(1, [n]) to a locally p0 -coCartesian edge of N (any such edge is automatically p0 -coCartesian, in view of
Lemma 2.1.16).
(∗0 ) Every functor f1 ∈ MapN(I) (N(I1 ), N) admits a p0 -right Kan extension f ∈ MapN(I) (N(I), N). More-
over, an arbitrary extension f 0 is a p0 -right Kan extension if and only if q carries each morphism
(0, [n]) → (1, [n]) to a p0 -Cartesian edge of N (these are the morphisms which project to equivalences
in M⊗ ).
Using (∗), we deduce that (i) and (iii0 ) imply (ii). Similarly, using (∗0 ), we deduce that (ii) and (iii1 )
imply (i). Let D00 be the full subcategory of MapN(I) (N(I0 ), N) spanned by the coCartesian sections, and
let D01 be the full subcategory of MapN(I) (N(I1 ), N) spanned by those functors which classify A-modules.
Proposition T.[Link] implies that the restriction maps

φ0 : D0 → D00 , φ1 : D1 → D01

are trivial Kan fibrations. The functor φ1 has a canonical section s, given by composition with the projection
I → ∆op ' I1 .
Let N0 = N ×N(I) N(I0 ). Using Proposition T.[Link], we deduce that the projection p00 : N0 → N(I0 ) is
a coCartesian fibration. Each fiber of p00 is equivalent to M, and each of the associated functors between the
fibers is an equivalence. Since N(I0 ) is weakly contractible (the category I0 has a zero object (0, [0])), we
deduce that N0 is equivalent to a product N(I0 ) × M. It follows also that the ∞-category D00 is equivalent,
via evaluation at the zero object of I0 , to M.
The forgetful functor ModA (M) → M is equivalent to the composition
s
θ : ModA (M) ' D01 → D1 → D00 → M,

where the final pair of maps are given by restriction and evaluation at the zero object of I0 . Using Proposition
T.[Link], we deduce that θ has a left adjoint, given by the composition
s0 s00
M → D00 → D0 ⊆ D → D01 ' ModA (M)

where the penultimate map is given by restriction, and the maps s0 and s00 are sections of the trivial fibrations
D0 → D00 → M. This proves (1).
The equivalence (1) ⇔ (2) follows from Proposition T.[Link]. More precisely, given M0 ∈ M, a free
A-module M generated by M0 can be constructed as follows. Choose a morphism α : (s00 ◦ s0 )(M0 ) → D
in D, where D ∈ D1 . Then D restricts to an object M ∈ ModA (M), and the natural transformation α
determines a map

ψ : M0 = ((s00 ◦ s0 )(M0 ))(0, [0]) → ((s00 ◦ s0 )(M0 ))(1, [0]) ' D(1, [0]) = θ(M ).

70
Since (s00 ◦ s0 )(M0 ) is a q-left Kan extension of D| I0 , (∗) implies that the edge

((s00 ◦ s0 )(M0 ))(0, [1]) → ((s00 ◦ s0 )(M0 ))(1, [0])

is q-coCartesian. It follows that the canonical map A ⊗ M0 → θ(M ) is an equivalence. This proves the “only
if” direction of (3), since a free A-module generated by M0 is uniquely determined up to equivalence.
We now prove the “if” direction of (3). Suppose given an A-module N and a map ψ 0 : M0 → θ(N ) which
induces an equivalence A ⊗ M0 ' N . Let M ∈ ModA (M) be defined as above. Then ψ 0 is equivalent to a
ψ θ(g)
composition M0 → θ(M ) → θ(N ) for some map g : M → N . We have a commutative diagram

θ(M )
uu: GG
GG θ(g)
uuu GG
GG
uu G#
uu
A ⊗ M0 / θ(N ).

It follows from the two-out-of-three property that θ(g) is an equivalence in M. Corollary 2.3.5 now implies
that g is an equivalence, so that ψ 0 exhibits N as an A-module freely generated by M0 .

2.5 Modules in a Monoidal Model Category

Let A be a simplicial model category, equipped with a compatible monoidal structure. In §1.6, we saw that
the underlying ∞-category N(Ao ) inherits the structure of a monoidal ∞-category. Moreover, in good cases
we can show that every algebra object A0 ∈ Alg(N(Ao )) arises (up to equivalence) from a strictly associative
algebra object A ∈ Alg(A) (Theorem 1.6.17). Our goal in this section is to compare the ∞-categorical
theory of A0 -modules in N(Ao ) with the classical theory of A-modules in A.
We begin by observing that if A is a monoidal category equipped with a compatible simplicial structure,
and A is a (strictly associative) algebra object of A, then ModA (A) inherits a simplicial structure in a
canonical way. Moreover:
Proposition 2.5.1. Let A be a combinatorial monoidal model category, and let A be an algebra object of A
which is cofibrant as an object of A. Then ModA (A) has the structure of a combinatorial model category,
where:
(W ) A morphism f : M → N is a weak equivalence in ModA (A) if and only if it is a weak equivalence in
A.
(F ) A morphism f : M → N is a fibration in ModA (A) if and only if it is a fibration in A.

The forgetful functor ModA (A) → A is both a left Quillen functor and a right Quillen functor. Moreover,
if A is equipped with a compatible simplicial structure, then the induced simplicial structure on ModA (A)
endows ModA (A) with the structure of a simplicial model category.
Proof. The proof is similar to that of Proposition 1.6.15. We first observe that ModA (A) is presentable
(Corollary 2.3.8). Let T : ModA (A) → A be the forgetful functor. Then T admits a left adjoint given by
the formula F (X) = A ⊗ X, and a right adjoint given by the formula G(X) =A X. Since A is combinatorial,
there exists a (small) collection of morphisms I = {iα : C → C 0 } which generates the class of cofibrations in
A, and a (small) collection of morphisms J = {jα : D → D0 } which generates the class of trivial cofibrations
in A. Let F (I) be the weakly saturated class of morphisms in ModA (A) generated by {F (i) : i ∈ I}, and
let F (J) be defined similarly. Unwinding the definitions, we see that a morphism in ModA (A) is a trivial
fibration if and only if it has the right lifting property with respect to F (i), for every i ∈ I. Invoking the
small object argument, we deduce that every morphism f : M → N in ModA (A) admits a factorization
f0 f 00
M → N 0 → N where f 0 ∈ F (I) and f 00 is a trivial fibration. Similarly, we can find an analogous factorization

71
where f 0 ∈ F (J) and f 00 is a fibration. Using standard arguments, we are reduced to the problem of showing
that each morphism belonging to F (J) is a weak equivalence in ModA (A). Let S be the collection of all
morphisms f : M → N in ModA (A) such that T (f ) is a trivial cofibration in A. We wish to prove that
F (J) ⊆ S. Since T preserves colimits, we conclude that S is weakly saturated; it will therefore suffice to
show that for each j ∈ J, F (j) ∈ S. In other words, we must show that if j : X → Y is a trivial cofibration
in A, then the induced map A ⊗ X → A ⊗ Y is again a trivial cofibration. This follows immediately from
the definition of a monoidal model category, in view of our assumption that A is a cofibrant object of A.
This completes the proof that ModA (A) is a model category.
The forgetful functor T : ModA (A) → A is a right Quillen functor by construction. To see that T is also
a left Quillen functor, it suffices to show that the right adjoint G : A → ModA (A) preserves fibrations and
trivial fibrations. In view of the definition of fibrations and trivial fibrations in ModA (A), this is equivalent
to the assertion that the composition T ◦ G : A → A preserves fibrations and trivial fibrations. This follows
immediately from the definition of a monoidal model category, since A ∈ A is cofibrant.
Now suppose that A is equipped with a compatible simplicial structure. We claim that ModA (A)
inherits the structure of a simplicial model category. For this, we suppose that f : M → N is a fibration in
ModA (A) and that g : X → Y is a cofibration of simplicial sets. We wish to show that the induced map
M Y → M X ×N X N Y is a fibration in ModA (A), which is trivial if either f or g is trivial. This follows
immediately from the analogous statement in the simplicial model category A.
Remark 2.5.2. Proposition 2.5.1 admits the following generalization: suppose that A is a combinatorial
model category, and that F : A → A is a left Quillen functor which is equipped with the structure of a
monad. Then the category of algebras over F inherits a model structure, where the fibrations and weak
equivalences are defined at the level of the underlying objects of A. The proof is exactly as given above.
Remark 2.5.3. If A is a symmetric monoidal model category satisfying the monoid axiom, then the hy-
pothesis that the algebra A be cofibrant is superfluous; see [60].
Theorem 2.5.4. Let A be a combinatorial simplicial model category equipped with a compatible monoidal
structure, let θ : N(Alg(A)o ) → Alg(N(Ao )) be as in Theorem 1.6.17, and let A ∈ Alg(A)o . Then the
canonical map N(ModA (A)o ) → Modθ(A) (N(Ao )) is an equivalence of ∞-categories.
Proof. Consider the diagram

N(ModA (A)o ) / Modθ(A) (N(Ao ))

NNN oo
NNNG ooo
NNN oo
o 0
NN' wooo G
o
N(A ).

We will show that this diagram satisfies the hypotheses of Corollary 3.5.7:
(a) The ∞-categories N(ModA (A)o ) and Modθ(A) (N(Ao )) admit geometric realizations of simplicial ob-
jects. In fact, both of these ∞-categories are presentable. For N(Alg(A)o ), this follows from Proposi-
tion T.A.3.7.6 and 2.5.1. For Modθ(A) (N(A)o ), we first observe that N(A)o is presentable (Proposition
T.A.3.7.6) and that the functor A ⊗ • preserves colimits, and then apply Corollary 2.3.8.
(b) The functors G and G0 admit left adjoints F and F 0 . The existence of a left adjoint to G follows from
the fact that G is determined by a right Quillen functor. The existence of a left adjoint to G0 follows
from Proposition 2.4.2.
(c) The functor G0 is conservative and preserves geometric realizations of simplicial objects. This follows
from Corollaries 2.3.3 and 2.3.7.
(d) The functor G is conservative and preserves geometric realizations of simplicial objects. The first
assertion is immediate from the definition of the weak equivalences in ModA (A), and the second
follows from the fact that G is also a left Quillen functor.

72
 (e) The natural map G0 ◦ F 0 → G ◦ F is an equivalence. Unwinding the definitions, we are reduced to
proving that if N ∈ Ao , then the natural map N → A ⊗ N induces an equivalence F 0 (N ) ' A ⊗ N .
This is a special case of Proposition 2.4.2.

2.6 The ∞-Category of Tensored ∞-Categories

According to Remark 1.2.15, there are at least three possible ways to define a monoidal ∞-category:
(A) As a coCartesian fibration C⊗ → N(∆)op ; this is the point of view we have chosen to adopt.
(B) As a monoid object in the ∞-category Cat∞ .

(C) As an algebra object of Cat∞ , where we view Cat∞ as endowed with the Cartesian monoidal structure.
The purpose of this section is to obtain a similar picture for the theory of ∞-categories tensored over
C. Our first step is to establish an analogue of Proposition 1.2.14, which will allow us to describe module
objects in an ∞-category endowed with a Cartesian monoidal structure.

Definition 2.6.1. Let C be an ∞-category. A left monoid action in C is a functor F : ∆1 × N(∆)op → C

with the following properties:
(1) The restriction F |{1} × N(∆)op → C is a monoid object of C (Definition 1.2.12).
(2) For n ≥ 0, the maps F (0, {n}) ← F (0, [n]) → F (1, [n]) exhibit F (0, [n]) as a product F (0, {n}) ×
F (1, [n]) in C.

We let MonL (C) denote the full subcategory of Fun(∆1 × N(∆)op , C) spanned by the left monoid actions.
We now have the following analogue of Proposition 1.2.14:
Proposition 2.6.2. Let C be an ∞-category which admits finite products, let p : C× → N(∆)op be the
associated (Cartesian) monoidal ∞-category (see Proposition 1.2.8), and let C×,L → C× exhibit C as left
tensored over itself (Example 2.1.3). Then composition with the Cartesian structure C× → C induces an
equivalence of ∞-categories Mod(C) → MonL (C).
Proof. Let t : ∆ → ∆ be the functor [n] 7→ [n] ? [0], and α : id → t the canonical transformation, which we
may identify with a functor [1] × ∆ → ∆. Let I = ∆× ×∆ ([1] × ∆). Objects of I can be identified with
triples (a, [n], i ≤ j), where 0 ≤ a ≤ 1 and 0 ≤ i ≤ j ≤ n + a. Let I0 be the full subcategory (equivalent to
[1] × ∆) spanned by objects of the form (a, [n], 0 ≤ n + a).
Unwinding the definitions, we see that Mod(M) is equivalent to the full subcategory of Fun(N(I)op , C)
spanned by those functors F : N(I)op → C which have the following properties:
(i) For 0 ≤ a ≤ 1 and 0 ≤ i ≤ jQ ≤ n + a, the natural maps F (a, [n], i ≤ j) → F (a, [n], k ≤ k + 1) exhibit
F (a, [n], i ≤ j) as a product i≤k<j F (a, [n], k ≤ k + 1) in the ∞-category C.

(ii) For every convex map α : [m] → [n], and every 0 ≤ i ≤ j ≤ m, the associated map F (0, [n], α(i) ≤
α(j)) → F (0, [m], i ≤ j) is an equivalence.
(iii) For every 0 ≤ i ≤ j ≤ n, the canonical map F (1, [n], i ≤ j) → F (0, [n], i ≤ j) is an equivalence in C.
(iv) For every morphism α : [m] → [n] such that α(m) = n, the map

F (1, [n], n ≤ n + 1) → F (1, [m], m ≤ m + 1)

is an equivalence.

73
 On the other hand, MonL (C) can be identified with the ∞-category of functors F0 : N(I0 )op → C with
the following properties:
(i0 ) For each n ≥ 0, the natural map F0 (0, [n], 0 ≤ n) → 0≤i<n F0 (0, {i, i + 1}, i ≤ i + 1) is an equivalence.
Q

(ii0 ) For n ≥ 0, the inclusion [0] ' {n} ⊆ [n] induces an equivalence
F0 (1, [n], 0 ≤ n + 1) ' F0 (0, [n], 0 ≤ n) × F0 (1, [0], 0 ≤ 1).

The map Mod(M) → MonL (C) is given by restriction from I to I0 . In virtue of Proposition T.[Link],
it will suffice to verify the following:
(a) Every functor F0 : N(I0 )op → C which satisfies (i0 ) and (ii0 ) admits a right Kan extension to N(I).
(b) A functor F : N(I)op → C satisfies conditions (i) through (iv) if and only if the restriction F0 =
F | N(I0 )op satisfies (i0 ) and (ii0 ), and F is a right Kan extension of F0 .
Assertion (a) is obvious, since for every object X = (a, [n], i ≤ j), the category I0 ×I IX/ has a final
object. Moreover, we see that a functor F : N(I)op → C is a right Kan extension of F | N(I0 )op if and only if
the following condition is satisfied:
(iii0 ) For every 0 ≤ a ≤ 1 and every 0 ≤ i ≤ j ≤ n, the natural map
F (a, [n], i ≤ j) → F (0, {i, i + 1, . . . j}, i ≤ j)
is an equivalence.
(iv 0 ) For 0 ≤ i ≤ n + 1, the map F (1, [n], i ≤ n + 1) → F (1, {i, . . . , n}, i ≤ n + 1) is an equivalence.
It remains only to show that conditions (i) through (iv) are equivalent to conditions (i0 ) through (iv 0 ).
This is straightforward and left to the reader (we observe that each of the relevant conditions depends only
on the underlying functor Iop → hC).
Our next goal is to relate monoid actions in Cat∞ with tensored ∞-categories. For this, we will need to
introduce a bit of notation.
Notation 2.6.3. Let A = (Set+ ∆ )/ N(∆)op denote the category of marked simplicial sets (X, E) equipped
with a map X → N(∆)op . We will regard A as endowed with the coCartesian model structure of Remark
T.[Link]. The fibrant objects of A are of the form X \ , where p : X → N(∆)op is a coCartesian fibration
and X \ = (X, E), where E is the collection of p-coCartesian edges of X.
We let A[1] denote the category of functors from [1] to A, endowed with the injective model structure
(see §T.A.3.3). Then A[1] inherits from A the structure of a simplicial model category. We let Cat ^ Mod
∞
denote the underlying ∞-category N((A[1] )o ), obtained by taking the simplicial nerve of the category of
fibrant-cofibrant objects of A[1] .
Note that every object of A[1] is cofibrant. Consequently, we may identify (A[1] )o with the category of
fibrant objects of A[1] . These, in turn, can be described as fibrations between fibrant objects of A. The
following result describes the structure of such a fibration:
Proposition 2.6.4. Suppose given a commutative triangle of ∞-categories

X@
f
/Y
@@ p 
@@ 
@@ q
 
S
where p and q are coCartesian fibrations, and f carries p-coCartesian morphisms in X to q-coCartesian
morphisms in Y . The following conditions are equivalent:

74
 (1) The map f is a categorical fibration of simplicial sets.
(2) The map f induces a fibration X \ → Y \ in the category (Set+ ∆ )/S , endowed with the opposite of
the model structure defined in §T.3.1 (here X \ and Y \ denote the marked simplicial sets obtained by
marking all of the coCartesian edges of X and Y , respectively).
Proof. According to Theorem T.[Link], the forgetful functor (Set+
∆ )/S → (Set∆ )/S is a right Quillen functor,
where (Set∆ )/S is endowed with the Joyal model structure. Thus (2) ⇒ (1). Conversely, suppose that f is
a categorical fibration. We must show that every lifting problem

A _
f
/ X\
}>
G } p
i }
 } 
B
g
/ Y\

in (Set+ +
∆ )/S has a solution, provided that i is a trivial cofibration in (Set∆ )/S . We first invoke Proposition
T.[Link] to deduce that X \ and Y \ are fibrant objects of (Set+ ∆ )/S . Consequently, f extends to a map
F : B → X \ . The maps p ◦ F, g : B → Y \ agree on A. Since Y \ is fibrant, we conclude that there is a
homotopy from p ◦ F to g. Thus, there exists a contractible Kan complex K containing a pair of vertices v
and v 0 , and a map h : B × K [ → Y \ such that h|B × {v} = p ◦ F and h|B × {v 0 } = g, and h|A × K [ coincides
f
with the composition A × K [ → A → Y \ . The inclusion
a
(B × {v}) (A × K [ ) ⊆ B × K [
A×{v}

is a categorical equivalence (on the level of the underlying simplicial sets). Invoking assumption (1), we
conclude that the homotopy h lifts to a map of simplicial sets H : B × K → X such that H|A × K coincides
f
with the composition A × K → A → X and H|B × {v} = F . Since the collection of p-coCartesian edges of
X is stable under equivalence, we deduce that H underlies a map of marked simplicial sets B × K [ → X \ .
The restriction H|B × {v 0 } is therefore a map B → X \ with the desired properties.
p
Suppose given a diagram M⊗ → C⊗ → N(∆)op which exhibits an ∞-category M as left-tensored over a
monoidal ∞-category C. According to Proposition 2.6.4, we can view p as a fibration between fibrant objects
of A; that is, as an object of the ∞-category Cat
^ Mod
∞ . Of course, not every object of Cat
^ Mod
∞is of this form.

Definition 2.6.5. Let CatMod

∞ denote the full subcategory of Cat
^ Mod
∞ corresponding to those diagrams
M → C → N(∆)op which exhibit C as a monoidal ∞-category and M⊗ as an ∞-category left-tensored
⊗ ⊗ ⊗

over C⊗ .
Corollary 2.6.6. There are canonical equivalences of ∞-categories

CatMod
∞ ' MonL (Cat∞ ) ' Mod(Cat∞ ),

where Cat∞ is endowed with the Cartesian monoidal structure and regarded as left-tensored over itself.
Proof. Let A denote the category of marked simplicial sets, endowed with the model structure of Proposition
T.[Link]. According to Proposition T.[Link], there is a (canonical) equivalence of ∞-categories Cat
^ Mod
'∞
Fun(∆1 , N(Ao )). Theorem T.[Link] allows us to identify N(Ao ) with Fun(N(∆)op , Cat∞ ). Consequently,
we obtain an equivalence Cat
^ Mod
∞ ' Fun(∆1 × N(∆)op , Cat∞ ). Unwinding the definitions, we see that under
this equivalence, Cat∞ corresponds to the full subcategory MonL (Cat∞ ) ⊆ Fun(∆1 × N(∆)op , Cat∞ ). This
Mod

proves the existence of the first equivalence. The existence of the second follows from Proposition 2.6.2.

75
Corollary 2.6.7. Let p : C⊗ → N(∆)op be a monoidal ∞-category, let q : C⊗,L → C⊗ exhibit C as
left-tensored over itself, let q 0 : M⊗ → C⊗ be an arbitrary ∞-category left-tensored over C, and let E ⊆
MapC⊗ (C⊗,L , M⊗ ) be the full subcategory spanned by those functors which carry q ◦ p-coCartesian edges
to q 0 ◦ p-coCartesian edges. Then evaluation on the unit object 1C induces an equivalence of ∞-categories
θ : E → M = M⊗ [0] .

Proof. It will suffice to prove that, for every simplicial set K, the map θ induces a bijection from the set
of equivalence classes of objects of Fun(K, E) to the set of equivalence classes of objects of Fun(K, M).
Replacing M⊗ by the fiber product C⊗ ×Fun(K,C⊗ ) Fun(K, M⊗ ), we may reduce to the case where K = ∆0 .
In this case, it suffices to show that θ induces an equivalence from the largest Kan complex contained in E
to the largest Kan complex contained in M. The latter Kan complex is equivalent to the mapping space
MapCat∞ (∗, M). Using Corollary 2.6.6, we can identify the former with MapModC⊗ (Cat∞ ) (C⊗,L , M⊗ ). The
desired result is now a special case of Corollary 2.4.3 (applied to the monoidal ∞-category Cat∞ , endowed
with the Cartesian monoidal structure).

2.7 Endomorphism Algebras

Let C be a monoidal category containing an object X, and suppose that there exists an endomorphism
object End(X): that is, an object of C equipped with evaluation map e : End(X) ⊗ X → X which induces
a bijection HomC (Y, End(X)) → HomC (Y ⊗ X, X) for every Y ∈ C (such an object always exists if the
monoidal category C is closed). Then End(X) is automatically equipped with the structure of an algebra
object of C: the unit map 1C → End(X) classifies the identity map idX : X → X, while the multiplication
End(X) ⊗ End(X) → End(X) classifies the composition

End(X) ⊗ (End(X) ⊗ X) → End(X) ⊗ X → X.

Our goal in this section is to obtain an ∞-categorical generalization of the above construction. With an
eye toward later applications, we will treat a slightly more general problem. Suppose that C is a monoidal
∞-category, M an ∞-category which is left-tensored over C, and M is an object of M. Let us say that an
object C ∈ C acts on M if we are given a map C ⊗ M → M . We would like to extract an object End(M ) ∈ C
which is universal among objects which act on M , and show that End(M ) is an algebra object of C (and
further, that M can be promoted to an object of ModEnd(M ) (M)).
Our first goal is to give a careful formulation of the universal property desired of End(M ). We would like to
have, for every algebra object A ∈ Alg(C), a homotopy equivalence of MapAlg(C) (A, End(M )) with a suitable
classifying space for actions of A on M . The natural candidate for this latter space is the fiber product
{A} ×Alg(C) Mod(M) ×M {M }, which can be viewed as a fiber of the projection map θ : Mod(M) ×M {M } →
Alg(C). We will see below that θ is a right fibration (Corollary 2.7.8). Our problem is therefore to find an
object End(M ) ∈ Alg(C) which represents the right fibration θ, in the sense that we have a an equivalence
Mod(M) ×M {M } ' Alg(C)/ End(M ) of right fibrations over Alg(C) (see §T.4.4.4).
The next step is to realize the ∞-category Mod(M) ×M {M } as (equivalent to) the ∞-category of algebra
objects in a suitable monoidal ∞-category C[M ]. Roughly speaking, we will think of objects of C[M ] as pairs
(C, η), where C ∈ C and η : C ⊗ M → M is a morphism in M. The monoidal structure on C[M ] may be
described informally by the formula (C, η) ⊗ (C 0 , η 0 ) = (C ⊗ C 0 , η 00 ), where η 00 denotes the composition

id ⊗η 0 η
C ⊗ C 0 ⊗ M → C ⊗ M → M.

The desired object End(M ) can be viewed as a final object of C[M ]. Provided that this final object exists, it
automatically has the structure of an algebra object of C[M ] (Corollary 1.5.5). The image of End(M ) under
the (monoidal) forgetful functor C[M ] → C will therefore inherit the structure of an algebra object of C.
We are now ready to begin with a detailed definition of the monoidal ∞-category C[M ].
Notation 2.7.1. We define a category J as follows:

76
 (a) An object of J is either pair ([n], i ≤ j), where [n] ∈ ∆ and 0 ≤ i ≤ j ≤ n, or a pair ([n], ∗), where
[n] ∈ ∆ and ∗ is a fixed symbol.
(b) Morphisms in J are given as follows:

HomJ (([m], i ≤ j), ([n], i0 ≤ j 0 )) = {α ∈ Hom∆ ([m], [n]) : i0 ≤ α(i) ≤ α(j) ≤ j 0 }

HomJ (([m], ∗), ([n], i ≤ j)) = {(α, k) : α ∈ Hom∆ ([m], [n]), i ≤ k ≤ j}

HomJ (([m], i ≤ j), ([n], ∗)) = ∅ HomJ (([m], ∗), ([n], ∗)) = Hom∆ ([m], [n]).

Let ∆0 denote a new copy ∞-category ∆ (to avoid confusion below), and define functors

ψ:J→∆ ψ 0 : J → ∆0

by the formulas

ψ([n], i ≤ j) = ψ([n], ∗) = [n] ψ 0 ([n], i ≤ j) = {i, i + 1, . . . , j} ψ 0 ([n], ∗) = [0].

We will identify ∆ with the full subcateory of J spanned by the objects ([n], ∗) (note that the remainder of
J can be identified with the category ∆× introduced in Notation 1.2.5). Similarly, we will identify ∆0 with
the full subcategory of J spanned by the objects ([n], 0 ≤ n).
q p
Let M⊗ → C⊗ → N(∆0 )op exhibit M = M⊗ ⊗
[0] as left-tensored over the monoidal ∞-category C = C[1] , and
⊗ ⊗
let M be an object of M. We define a simplicial set C[M
]] equipped with a map C[M
]] → N(∆)op as follows.
⊗
Let K be an arbitrary simplicial set equipped with a map e : K → N(∆)op . Then HomN(∆)op (K, C[M
]] ) is
in bijection with the set of commutative diagrams

K ×N(∆)op N(∆)op / {M }
_ _

 
K ×N(∆)op N(J)op / M⊗

p◦q
 
N(∆0 )op N(∆0 )op .

For each [n] ∈ ∆, we let J[n] denote the fiber J ×∆ {[n]}. An object of J[n] can be identified with either a
⊗
pair of integers i and j such that 0 ≤ i ≤ j ≤ n, or the symbol ∗. A vertex of C[M
]] can be identified with
the following data:
(i) An object [n] ∈ ∆.

(ii) A functor f : N(J[n] )op → M⊗ , covering the map N(J[n] )op → N(∆0 )op induced by the functor ψ 0 .
⊗
We let C[M ]⊗ denote the full simplicial subset of C[M
]] spanned by those objects which classify functors
f : N(J[n] )op → M⊗ satisfying the following additional conditions:

(1) For every morphism α in J[n] , the edge (q ◦ f )(α) ∈ Hom(∆1 , C⊗ ) is p-coCartesian.

(2) Let α : ([n], ∗) → ([n], i ≤ j) be the morphism in J[n] corresponding to the element j ∈ {i, . . . , j}. Then
f (α) is (p ◦ q)-coCartesian.
We let C[M ] denote the fiber C[M ]⊗
[1] .

77
 Our first main result is:
Proposition 2.7.2. Let C be a monoidal ∞-category, M an ∞-category which is left-tensored over C, and
M ∈ M an object. Then:
(1) The map C[M ]⊗ → N(∆)op constructed in Notation 2.7.1 is a monoidal ∞-category.
(2) Restriction to ∆0 ⊆ J induces a monoidal functor C[M ]⊗ → C⊗ .
We will give the proof at the end of this section. First, we would like to have a better understanding of
the category C[M ]. An object of C[M ] can be identified with a diagram

M0 Co N / M1
CC {{
CC∼ {
CC {{
C!   }{{{ ∼
M

in M⊗ . Here N ∈ M⊗ [1] ' C × M. We may therefore identify N with a pair of objects C ∈ C, M ∈ M. The
0

triangle on the right determines equivalences M 0 ' M1 ' M in M, while the triangle on the left determines
an equivalence M0 ' M and a map C ⊗ M 0 → M0 . Consequently, every object of C[M ] determines a
morphism α : C ⊗ M → M in M, which is well-defined up to homotopy. In fact, we have the following more
precise statement:
Proposition 2.7.3. Let C be a monoidal ∞-category, M an ∞-category which is left-tensored over C, and
M ∈ M an object. Then:
(1) Consider the functor C → M given by C 7→ C ⊗ M . There exists an equivalence of ∞-categories
e : C ×M M/M → C[M ] such that the composition of e with the forgetful functor C[M ] → C is equivalent
to the projection C ×M M/M → C.
(2) The forgetful functor C[M ] → C is a right fibration.
(3) Let N ∈ C[M ] be an object which classifies a morphism α : C ⊗ M → M (see the discussion preceding
Lemma 2.7.14). Then N is a final object of C[M ] if and only if α exhibits C as a morphism object
MorM (M, M ) (Definition 2.1.9).
Again, we will defer the proof until the end of this section.
Remark 2.7.4. Proposition 2.7.3 implies that the fiber C[M ] ×C {1C } can be identified with the Kan
complex M/M ×M {M } ' MapM (M, M ). The monoidal structure on C[M ] induces a coherently associative
multiplication on MapM (M, M ), which simply encodes the composition in the ∞-category M. In particular,
1C is an object of C[M ] ×C {1C } which classifies an equivalence from M to itself, then e
if e 1C is an invertible
object of C[M ] (see Example 1.3.2).
Corollary 2.7.5. Let C be a monoidal ∞-category, M an ∞-category which is left-tensored over C, and
M ∈ M an object. Then:
(1) For each n ≥ 0, the forgetful functor f : C[M ]⊗ → C⊗ induces a right fibration f[n] : C[M ]⊗ ⊗
[n] → C[n] .

(2) Consider the diagram

C[M ]⊗
f
/ C⊗
JJ ww
JJ q ww
JJ
JJ wwwp
J% {ww
N(∆)op .
A morphism α in C[M ]⊗ is q-coCartesian if and only if f (α) is p-coCartesian.

78
 (3) Let A be a section of q. Then A ∈ Alg(C[M ]) if and only if f ◦ A ∈ Alg(C). Similarly, an object
A0 ∈ MapN(∆)op (N(∆nu )op , C[M ]⊗ ) belongs to Algnu (C[M ]) if and only if f ◦ A0 ∈ Algnu (C).
(4) Composition with f induces isomorphisms of simplicial sets

Alg(C[M ]) → Alg(C) ×MapN(∆)op (N(∆)op ,C⊗ ) MapN(∆)op (N(∆)op , C[M ]⊗ ).

Algnu (C[M ]) → Algnu (C) ×MapN(∆)op (N(∆nu )op ,C⊗ ) MapN(∆)op (N(∆nu )op , C[M ]⊗ ).

Proof. Each of the functors f[n] is equivalent to the nth power of the forgetful functor C[M ] → C. Conse-
quently, (1) follows immediately from Proposition 2.7.3 and Lemma 2.7.14. In particular, we deduce that a
morphism α in C[M ]⊗ [n] is an equivalence if and only if f[n] (α) is an equivalence. Since f preserves coCartesian
edges (Proposition 2.7.2), assertion (2) follows. The implications (2) ⇒ (3) and (3) ⇒ (4) are obvious.
In the situation of Proposition 2.7.2, we obtain an induced map Alg(C[M ]) → Alg(C). We may therefore
think of an algebra object of C[M ] as an algebra object of C equipped with some kind of additional structure.
The following result makes this idea precise:

Proposition 2.7.6. Let C be a monoidal ∞-category, M an ∞-category which is left-tensored over C, and
M ∈ M an object. Then composition with the functor ψ 0 of Notation 2.7.1 induces categorical equivalences

Mod(M) ×M {M } → Alg(C[M ]) Modnu (M) ×M {M } → Algnu (C[M ]).

In other words, we can identify algebra objects of C[M ] with algebra objects A ∈ Alg(C), together with
an action of A on the fixed object M ∈ M. The proof will be given at the end of this section.
Proposition 2.7.7. Suppose given a diagram

C⊗ G
f
/ D⊗
GG v
GGq vv
GG vv
GG vvv p
# {v
N(∆)op

which exhibits C = C⊗ ⊗
[1] and D = D[1] as monoidal ∞-categories. Suppose further that f is a categorical
fibration, a monoidal functor, and that f induces a right fibration C → D. Then composition with f induces
right fibrations
Alg(C) → Alg(D) Algnu (C) → Algnu (D).
Corollary 2.7.8. Let C be a monoidal ∞-category, M an ∞-category which is left-tensored over C, and
M ∈ M an object. Then:
(1) The forgetful functors
Alg(C[M ]) → Alg(C) Algnu (C[M ]) → Algnu (C)
are right fibrations of simplicial sets.
(2) The forgetful functors

Mod(M) ×M {M } → Alg(C) Modnu (M) ×M {M } → Algnu (C)

are right fibrations of simplicial sets.

Proof. Assertion (1) follows from Propositions 2.7.7, 2.7.3, and 2.7.2. Assertion (2) follows from (1), Propo-
sition 2.7.6, and Lemma 2.7.14.

79
Corollary 2.7.9. Let C be a monoidal ∞-category, M an ∞-category which is left-tensored over C, and let
M ∈ Mod(M) a (left) module object having images M ∈ M and A ∈ Alg(C). Suppose that the multiplication
map A⊗M → M exhibits A as a morphism object MorM (M, M ). Then, for every algebra object B ∈ Alg(C),
we have a canonical isomorphism MapAlg(C) (B, A) ' ModB (M) ×M {M } in the homotopy category H of
spaces.
Proof. Consider the diagram Alg(C)/A ← (Mod(M) ×M {M })/M → Mod(M) ×M {M }. Propositions 2.7.6
and 2.7.3 imply that M is a final object of Mod(M) ×M {M }, so that the right map is a trivial Kan fibration.
Since the map Mod(M) ×M {M } → Alg(C) is a right fibration (Corollary 2.7.8), the left map is also a trivial
Kan fibration. Passing to fibers over the object B ∈ Alg(C), we obtain the desired result.
Remark 2.7.10. In the situation of Corollary 2.7.9, let A0 ∈ Algnu (C) denote the image of A under the
forgetful functor. Then the same argument shows that, for every nonunital algebra object B0 ∈ Algnu (C),
B0 (M) ×M {M } in the homotopy category H
we have a canonical isomorphism MapAlgnu (C) (B0 , A0 ) ' Modnu
of spaces.
The proof of Proposition 2.7.7 is based on the following easy lemma:
Lemma 2.7.11. Consider a diagram of simplicial sets

XA
f
/Y
AA q ~~
AA ~
AA ~~p
~~ ~
S.
Suppose that:
(i) The maps p and q are coCartesian fibrations.
(ii) The map f carries q-coCartesian edges of X to p-coCartesian edges of Y .
(iii) The map f is a categorical fibration.
(iv) For every vertex s of S, the induced map Xs → Ys is a right fibration.
Then composition with f induces a right fibration MapS (S, X) → MapS (S, Y ).
Proof. According to Corollary T.[Link], it will suffice to show that the map

φ : Fun(∆1 , MapS (S, X)) → Fun(∆1 , MapS (S, Y )) ×Fun({1},MapS (S,Y )) Fun({1}, MapS (S, X))

is a trivial Kan fibration. Set

X 0 = Fun(∆1 , X) ×Fun(∆1 ,S) S, Y 0 = Fun(∆1 , Y ) ×Fun(∆1 ,S) S, Z = Y 0 ×Fun({1},Y ) Fun({1}, X),

so that f determines a map φ : X 0 → Z, and φ can be identified with the induced map MapS (S, X 0 ) →
MapS (S, Z). It will therefore suffice to show that φ is a trivial Kan fibration. Since φ is clearly a categorical
fibration, it will suffice to show that φ is a categorical equivalence.
We have a commutative diagram

X0 B
φ
/ Z ψ / Y0
BB q0 }
BB }}
BB r }}} 0
B  ~}} p
S.
Proposition T.[Link] and (i) implies that p0 and q 0 are coCartesian fibrations, and (ii) implies that ψ◦φ carries
q 0 -coCartesian edges to p0 -coCartesian edges. Combining (ii), (iii), and Proposition 2.6.4, we deduce that r

80
is also a coCartesian fibration, and that φ carries q 0 -coCartesian edges to r-coCartesian edges. According to
Proposition T.[Link], the map φ is a categorial equivalence if and only if it induces a categorical equivalence
φs : Xs0 → Zs , for each vertex s ∈ S. We now observe that (iv) and Corollary T.[Link] imply that φs is a
trivial Kan fibration.
Proof of Proposition 2.7.7. We will give the proof in the unital case; the assertion for nonunital algebras
follows using the same argument. According to Corollary 2.7.5, we have a pullback diagram

Alg(C)
θ / Alg(D)

 
θ0 / MapN(∆)op (N(∆)op , D⊗ ).
MapN(∆)op (N(∆)op , C⊗ )

Corollary 2.7.5 and Lemma 2.7.11 imply that θ0 is a right fibration. It follows that θ is also a right fibration.

We now return to the proofs of Propositions 2.7.2, 2.7.3, and 2.7.6. We first need to establish some
technical preliminaries.
Lemma 2.7.12. Suppose given a pullback square of ∞-categories

X0
q
/X

p0 p
 0 
S0
q
/ S,

where p is a coCartesian fibration. Let f : K / → X 0 be an arbitrary diagram. Then f is a p0 -limit diagram

if and only if q ◦ f is a p-limit diagram.
Proof. Replacing the above diagram by
0
X/f / X/qf

 
0
S/p 0f / S/pqf

(and invoking Proposition T.[Link]), we can reduce to the case where K = ∅. In this case, we can identify
f with an object x0 ∈ X 0 . Let s0 = p0 (x0 ) and s = q 0 (s0 ). Corollary T.[Link] implies that x0 is p0 -final if
and only if x0 is a final object of the fiber Xs0 0 . Similarly, x = q(x0 ) is p-final if and only if x is a final object
of the fiber Xs . The desired result now follows from the observation that Xs0 0 ' Xs .
q p
Lemma 2.7.13. Let M⊗ → C⊗ → N(∆)op exhibit M = M⊗ ⊗
[0] as left-tensored over C = C[1] . Let n ≥ 2, and
suppose we are given a diagram σ0 :
XL E XR
EE x x
EEα xx
EE xx
E" |xx β
X LR
in M⊗ , lifting the diagram τ0 :

{0, . . . , n − 1} {1, . . . , n}
hQQQ oo7
QQQ ooo
QQQ
QQQ ooo
ooo
{1, . . . , n − 1}

81
in ∆. Suppose furthermore that α is a (p◦q)-coCartesian morphism in M⊗ , and that q(β) is a p-coCartesian
morphism in C⊗ . Then:
(1) The diagram τ :
6 [n] gOOO
m mmmm OOO
mmmm OOO
mmm OOO
mm O
{0, . . . , n − 1} {1, . . . , n}
hQQQ oo7
QQQ ooo
QQQ o o
QQQ o
ooo
{1, . . . , n − 1}
in ∆ can be lifted to a (p ◦ q)-limit diagram σ:

XF
xx FF
x FFα
xxx FF
FF
|xx β #
L
X E XR
EE xx
EEα xx
EE xx β
E" |xx
X LR

in M⊗ .
(2) Consider an arbitrary diagram σ as in (1), which is compatible with both σ0 and τ . Then σ is a (p ◦ q)-
limit diagram if and only if α is a (p ◦ q)-coCartesian morphism in M⊗ and q(β) is p-coCartesian
morphism in C⊗ .
Proof. We first prove (1). Let us identify τ with a map ∆1 × ∆1 → N(∆)op , and form a pullback diagram

N / M⊗

r p◦q
 
∆1 × ∆1
τ / N(∆)op .

We will regard N as a simplicial subset of M⊗ containing the diagram σ0 . In view of Lemma 2.7.12 and
the fact that (p ◦ q) is a coCartesian fibration, it will suffice to prove that σ0 can be extended to an r-limit
diagram σ : ∆1 × ∆1 → N.
Unwinding the definitions, we see that the coCartesian fibration r : N → ∆1 × ∆1 is classified by the
following diagram of ∞-categories:

C⊗ ⊗ ⊗
{0,1} × C{1,...,n−1} × M{n−1,n}
h U UUUU
hhh
hhhhhhh UUUU
UUUU
hhhhhh T UUUU
shh *
C⊗
{0,1} × C⊗
{1,...,n−1} × M ⊗
C⊗ ⊗
{1,...,n−1} × M{n−1,n}
V{n−1}
VVVV ii
VVVV iiii
VVVV
VVVV iiiiiii
VV+ tiiii T
⊗ ⊗
C{1,...,n−1} × M{n−1} .

Here
⊗
T : M⊗ ⊗
{n−1,n} ' C × M → M ' M{n−1}

82
denotes the left action of C on M. Let N0 denote the relative nerve of this diagram (see §3.1), so that we have
an equivalence f : N0 → N of coCartesian fibrations over ∆1 × ∆1 , and r0 : N0 → ∆1 × ∆1 the projection.
The diagram σ0 is equivalent to a composition f ◦ σ00 . It will suffice to show that the analogues of (1)
and (2) hold for σe0 :
(10 ) The map σ00 can be extended to an r0 -limit diagram σ 0 : ∆1 × ∆1 → N0 (which is simultaneously a
section of r0 ).
(20 ) Let σ 0 be an arbitrary section of r0 which extends σ00 , and denote the diagram f ◦ σ 0 as follows:
X0 G
w GG 0
ww GGα
www0 GG
GG
w
w{ w β #
L R
X0 G X0
GG 0 ww
GGα ww
GG ww β
G# {ww
LR
X0 .
0
Then σ 0 is an r0 -limit diagram if and only if α0 is a (p ◦ q)-coCartesian morphism in M⊗ and q(β ) is
p-coCartesian morphism in C⊗ .
The diagram σ00 determines objects
(C, D, N ) ∈ C⊗ ⊗ ⊗
{0,1} × C{1,...,n−1} × M{n−1}

(D0 , M ) ∈ C⊗ ⊗
{1,...,n−1} × M{n−1,n} (D00 , N 0 ) ∈ C⊗ ⊗
{1,...,n−1} × M{n−1} .
and morphisms
γ0 γ1 δ δ
D → D00 ← D0 , 0
N→ N0 ←
1
T (M )
Since α is (p ◦ q)-coCartesian, we conclude that γ0 and δ0 are equivalences in C⊗ {1,...,n−1} . Similarly, since
q(β) is p-coCartesian, we deduce that δ0 is an equivalence in M⊗ {n−1} . Modifying σ00 if necessary, we may
0 00 0
assume that D = D = D , N = N , and the maps γ0 , γ1 and δ0 are all identities. In this case, we can take
σ 0 to be the diagram
(C, D, M )
q q LLL
qqq LLL
qq q LLL
xqq id × id ×δ 1 L&
(C, D, N ) (D, M )
MMM rr
MMM rr
MMM rrr
M& xrrr id ×δ1
(D, N ).
An easy calculation shows that σ 0 is an r0 -limit diagram. This proves (1). Moreover, σ 0 satisfies the criterion
of (20 ). Since r0 -limit diagrams extending σ00 are uniquely determined up to equivalence, we deduce the “only
if” direction of (20 ).
To prove the “if” direction of (20 ), let us suppose given an arbitrary section σ 0 of r0 which extends σ00 ,
depicted below:
(C0 , D0 , M0 )
oo NNN
ooo NNgN1 ×g2
o
ofo0 ×f1 ×f2 NNN
wooo N&
(C, D, N ) (D, M )
OOO pp
OOO pp
OOO ppp
OO' xppp id ×δ1
(D, N ).

83
Since σ 0 is an r0 -limit diagram, there exists a natural transformation s : σ 0 → σ 0 which is the identity,
except possibly on the initial object. Suppose that σ 0 satisfies the criterion of (20 ). Then the maps f0 ,
f1 , and g2 are equivalences. Using the two-out-of-three property, we see that s induces an equivalence
(C0 , D0 , M0 ) → (C, D, M ), so that s is itself an equivalence. Since σ 0 is equivalent to σ 0 , it is also an r0 -limit
diagram, as we wished to prove.
q p
Proof of Proposition 2.7.2. Let M⊗ → C⊗ → N(∆)op exhibit M = M⊗ ⊗
[0] as left-tensored over C = C[1] . The
functor ψ : J → ∆ is an op-fibration of categories, so the induced map N(J)op → N(∆)op is a Cartesian
fibration. We now define a simplicial set X equipped with a map f : X → N(∆)op , characterized by the
following universal property: for every map of simplicial sets K → N(∆)op , we have a canonical bijection
HomN(∆)op (K, X) ' HomN(∆0 )op (K ×N(∆)op N(J)op , M⊗ ). Invoking Corollary T.[Link], we deduce:
(i) The map f is a coCartesian fibration.
(ii) Let α be a morphism in X, covering a map α : [m] → [n] in ∆. Then α is f -coCartesian if and only if
the following conditions are satisfied:
(a) For every 0 ≤ i ≤ j ≤ m, the induced map α([n], α(i) ≤ α(j)) → α([m], i ≤ j) is a (q ◦ p)-
coCartesian morphism in M⊗ .
(b) The map α([n], ∗) → α([m], ∗) is a (q ◦ p)-coCartesian morphism in M⊗ .
Restriction to the full subcategory ∆ ⊆ J (spanned by the objects of the form ([n], ∗)) gives rise to a
commutative diagram
XG
s / M × N(∆)op
GG pp
GGf ppp
GG pp
GG p 0
# xppp f
op
N(∆) .

The map f 0 is obviously a coCartesian fibration, and (ii) implies that s carries f -coCartesian edges to f 0 -
⊗
coCartesian edges. Moreover, we have a canonical isomorphism C[M ]] ' X ×M × N(∆)op ({M } × N(∆)op ).
Combining Proposition 2.6.4 with Proposition T.[Link], we deduce:
⊗
(i0 ) The projection g : C[M
]] → N(∆)op is a coCartesian fibration.
⊗
(ii0 ) Let α be a morphism in C[M]] , covering a map α : [m] → [n] in ∆. Then α is g-coCartesian if and only
if, for every 0 ≤ i ≤ j ≤ m, the induced map α([n], α(i) ≤ α(j)) → α([m], i ≤ j) is a (q ◦ p)-coCartesian
morphism in M⊗ .

It follows easily from (ii0 ) that the map g restricts to a coCartesian fibration g0 : C[M ]⊗ → N(∆)op , and
that a morphism in C[M ]⊗ is g0 -coCartesian if and only if it is g-coCartesian.
We now prove (1). We must show that, for each n ≥ 0, the coCartesian fibration g0 induces an equivalence
⊗
C[M ]⊗ ⊗ ⊗
[n] ' C[M ]{0,1} × . . . × M{n−1,n} . For n = 0 this is easy: the ∞-category C[M ][0] is isomorphic to the
full subcategory of M/M spanned by the final objects, and therefore a contractible Kan complex (Proposition
T.[Link]). For each object [m] ∈ ∆, let J[m] denote the fiber product J ×∆ {[m]}. Let P be the set of pairs
of integers i, j ∈ [n] satisfying i ≤ j. We endow P with the following partial ordering:

(i ≤ j) ≤ (i0 ≤ j 0 )

if and only if the interval {i, . . . , j} is contained in the interval {i0 , . . . , j 0 }; in other words, if and only
if 0 ≤ i0 ≤ i ≤ j ≤ j 0 ≤ n. We regard P as equipped with a functor P → ∆0 , given by (i ≤ j) 7→

84
{i, i + 1, . . . , j} ' [j − i]. For every downward-closed subset Q ⊆ P , let JQ
[n] denote the full subcategory of
J[n] spanned by the objects ([n], ∗) and {([n], i ≤ j)}(i≤j)∈Q , and let E(Q) denote the full subcategory of
⊗
[n] ) , M )
MapN(∆0 )op (N(JP 0 op

spanned by those functors F : N(J0[n] )op → M⊗ which satisfy the following conditions:
(a) Let (i ≤ j) ∈ Q and let 0 ≤ i ≤ i0 ≤ j ≤ n. Then the induced map F ([n], i ≤ j) → F ([n], i0 ≤ j) is a
(p ◦ q)-coCartesian morphism in M⊗ .
(b) Let (i ≤ j) ∈ Q and let 0 ≤ i ≤ i0 ≤ j 0 ≤ j ≤ n. Then q carries the induced map F ([n], i ≤ j) →
F ([n], i0 ≤ j 0 ) to a p-coCartesian morphism in C⊗ .
(c) Let (i ≤ j) ∈ Q and let α : ([n], ∗) → ([n], i ≤ j) be the morphism in J[n] classified by the element
j ∈ {i, i + 1, . . . , j}. Then F (α) is a (p ◦ q)-coCartesian morphism in M⊗ .
It is convenient to reformulate the conditions (a), (b), and (c). Using a transitivity argument, it is easy
to see that (a) is equivalent to the following apparently weaker condition:

(a0 ) Let (i < j) ∈ Q. Then the induced map F ([n], i ≤ j) → F ([n], i + 1 ≤ j) is a (p ◦ q)-coCartesian
morphism in M⊗ .

Assuming (a0 ), another transitivity argument allows us to reformulate (b) as follows:

(b0 ) Let (i < j) ∈ Q. Then q carries the induced map F ([n], i ≤ j) → F ([n], i ≤ j − 1) to a p-coCartesian
morphism in C⊗ .

Finally, assuming that (a0 ) is satisfied, another transitivity argument gives the following reformulation of
(c):

(c0 ) Let (i ≤ i) ∈ Q. Then the induced map F ([n], i ≤ i) → F ([n], ∗) is (p ◦ q)-coCartesian.

Evaluation at the object ([n], ∗) induces a map E(Q) → M; let E0 (Q) denote the fiber E(Q) ×M {M }.
We observe that C[M ]⊗ [n] is canonically isomorphic to E0 (P ). Let P≤1 ⊆ P be the collection of all pairs
(i ≤ j) ∈ P such that j ≤ i + 1. The simplicial set E0 (P≤1 ) is canonically isomorphic to the fiber product

C[M ]⊗
{0,1} ×C[M ]
⊗ . . . ×C[M ]⊗ C[M ]⊗
{n−1,n} ,
{1} {n−1}

and
Q is therefore (since this fiber product is also a homotopy fiber product) equivalent to the product
⊗
0≤i<n C[M ]{i,i+1} . To complete the proof of (1), it will suffice to show that the restriction map r0 :
E0 (P ) → E0 (P≤1 ) is a trivial Kan fibration.
We have a Cartesian rectangle of simplicial sets

E0 (P )
r0
/ E0 (P≤1 ) / {M }

  
E(P )
r / E(P≤1 ) / M.

It will therefore suffice to show that the map r is a trivial Kan fibration. We will prove the following more
general statement: for every P≤1 ⊆ Q ⊆ Q0 ⊆ P , the restriction map E(Q0 ) → E(Q) is a trivial Kan fibration.
Using a transitivity argument, we can reduce the case where Q0 is obtained from Q by adjoining a single
element (i ≤ j) ∈ P − Q. Since P≤1 ⊆ Q, we have j ≥ i + 2.
In view of Proposition T.[Link], it will suffice to prove the following, for every F0 ∈ E(Q):

85
 0
⊗
(I) There exists a functor F ∈ MapN(∆0 )op (N(JQ
[n] ) , M ) which is a (p ◦ q)-right Kan extension of F0 .
op

0
⊗
(II) Let F ∈ MapN(∆0 )op (N(JQ [n] ) , M ) be an arbitrary extension of F0 . Then F is a (p ◦ q)-right Kan
op

extension of F0 if and only if F ∈ E(Q0 ).

Let I = JQ
[n] ×J[n] (J[n] )/([n],i≤j) , and let I0 denote the full subcategory of I spanned by the objects

X L : ([n], i ≤ j − 1) → ([n], i ≤ j), X R : ([n], i + 1 ≤ j) → ([n], i ≤ j)

X LR : ([n], i + 1 ≤ j − 1) → ([n], i ≤ j).

F
Let F00 denote the composition N(I)op → N(JQ [n] ) → M⊗ . Using Lemma T.[Link] (and the equivalence
op 0

between conditions (a), (b), and (c) with their analogues (a0 ), (b0 ), and (c0 )) we are reduced to proving:

(I 0 ) There exists a (p ◦ q)-limit diagram F 0 , rendering following diagram commutative:

F00
N(I) op / ⊗
_ r8M
F0 r
r
r r p◦q
 r 
(N(I)op )/ / N(∆0 )op .

(II 0 ) Let F 0 be an arbitrary map which renders the above diagram commutative. Then F 0 is a (p ◦ q)-limit
diagram if and only if F 0 ({X R }/ ) is (p ◦ q)-coCartesian morphism in M⊗ , and (q ◦ F 0 )({X L }/ ) is
p-coCartesian morphism in C⊗ .

We observe that the inclusion I0 ⊆ I has a left adjoint, so the induced map N(I0 ) → N(I) is cofinal.
Consequently, it suffices to prove the analogues of (I 0 ) and (II 0 ) obtained by replacing I by I0 . In this case,
the desired result is an immediate conseqence of Lemma 2.7.13. This completes the proof of (1).
We now prove (2). We wish to show that the restriction map C[M ]⊗ → C⊗ carries g0 -coCartesian edges
to p-coCartesian edges. Let α be a g0 -coCartesian edge of C[M ]⊗ , covering a map α : [m] → [n] in ∆. Let
β : α([n], 0 ≤ n) → α([m], 0 ≤ m) be the induced map in M⊗ . We wish to show that q(β) is a p-coCartesian
morphism in C. We observe that q(β) factors as a composition

q(β 0 ) q(β 00 )
q(α([n], 0 ≤ n)) → q(α([n], α(0) ≤ α(m))) → q(α([m], 0 ≤ m)).

Condition (ii0 ) implies that β 00 is (p ◦ q)-coCartesian, so that q(β 00 ) is p-coCartesian. Moreover, since the
domain of α belongs to C[M ]⊗ , the map q(β 0 ) is p-coCartesian. It follows that q(β) is p-coCartesian, as
desired.
We now turn to the proof of Proposition 2.7.3. Once again, we will need a lemma.

Lemma 2.7.14. Suppose given a diagram of simplicial sets

X0 /X

p0 p
 
S0 /S

where p is a right fibration, p0 a categorical fibration, and the horizontal arrows are categorical equivalences.
Then p0 is also a right fibration.

86
Proof. Replacing X by X ×S 0 S (and invoking Proposition T.[Link]), we can reduce to the case where S = S 0
and the bottom horizontal map is the identity. We wish to show that every lifting problem of the form

A _ / X0
}>
} p0
i }
 } 
B / S0

admits a solution, provided that i is right anodyne. Applying Proposition T.[Link] (in the model category
(Set∆ )/S 0 , we can reduce to the problem of solving the associated mapping problem

A _ /X
~?
~
i ~ p
 ~ 
B / S,

which is possibly in virtue of our assumption that p is a right fibration.

Proof of Proposition 2.7.3. We first prove (1). Let N denote the fiber product M⊗ ×N(∆0 )op N(J[1] )op , and
p : N → N(J[1] )op the projection. Then p is classified by a functor F from N(J[1] )op to the ∞-category
Cat∞ . Unwinding the definitions, we see that F is equivalent to the functor described by the diagram of
∞-categories
M Fo FF ⊗ C × M /
π
FFFF x
xxx M
FFFF x
FFFF xxxxx
FFF   xxxxxxx
x
M.
Here ⊗ : C × M → M denotes the tensor product functor, and π : C × M → M denotes the projection onto
0
the second factor. This diagram is described by a functor F 0 : Jop
[1] → Set∆ . Let N denote the relative nerve
op 0
NF 0 (J[1] ) (see §3.1), so that we have an equivalence N → N of coCartesian fibrations over N(J[1] )op . This
equivalence induces an equivalence of ∞-categories C[M ]0 → C[M ], where C[M ]0 denotes the fiber product

{M } ×Fun({2},M) Fun0 (∆2 , M) ×Fun({0},M) Fun({0}, C × M) ×Fun({0},M) Fun00 (∆2 , M) ×Fun({2},M) {M };

here Fun0 (∆2 , M) denotes the full subcategory of Fun(∆2 , M) spanned by those diagrams

N0 C
}}> CC γ
β
}} CC
}} CC
}} C!
N / N 00

where γ is an equivalence, and Fun00 (∆2 , M) the full subcategory spanned by those diagrams where β
and γ are both equivalences. Since ∆2 is weakly contractible, the diagonal map M → Fun00 (∆2 , M) is a
categorical equivalence. An easy argument shows that this diagonal map induces a categorical equivalence
C[M ]00 → C0 [M ], where C[M ]00 denotes the fiber product

{M } ×Fun({2},M) Fun0 (∆2 , M) ×Fun({0},M) Fun({0}, C);

here C maps to M via the functor C 7→ C ⊗ M . We now observe that evaluation along the long edge of ∆2
induces a trivial Kan fibcation Fun0 (∆2 , M) → Fun(∆{0,2} , M). We therefore obtain a trivial Kan fibration
C[M ]00 → C ×M M/M . Let s denote a section to this trivial fibration (for example, the section given by
composition with a degeneracy map ∆2 → ∆1 ), and define e to be the composition
s
C ×M M/M → C[M ]00 ⊆ C[M ]0 → C[M ].

87
It is easy to see that e satisfies the requirements of (1). Assertion (2) follows from (1) and Lemma 2.7.14,
since the projection M/M → M is a right fibration.
We now prove (3). Note that the objects of C[M ]00 can be identified with pairs (C, α), where C ∈ C and
α : C ⊗ M → M is a morphism in M. In view of (1), it will suffice to show that an object (C, α) ∈ C[M ]00
is final if and only if α exhibits C as a morphism object MorM (M, M ). Fix another object In view of (1), it
will suffice to show that an object (C 0 , α0 ) ∈ C[M ]00 . Since the projection C[M ]00 → C is a right fibration, the
induced map q : C[M ]00/(C,α) → C/C is a trivial Kan fibration. We therefore obtain a homotopy fiber sequence
γ
MapC[M ]00 ((C 0 , α0 ), (C, α)) → MapC (C 0 , C) → C[M ]00 ×C {C 0 }.

It follows that (C, α) is an initial object of C[M ]00 if and only if γ is a homotopy equivalence, for every choice
of object C 0 ∈ C. We now observe that the codomain of γ is canonically identified with MapM (C 0 ⊗M, M ), so
that the γ is an equivalence for every C 0 ∈ C if and only if α exhibits C as a morphism object MorM (M, M ).
This proves (3).
Proof of Proposition 2.7.6. We will give the proof for unital algebras; the nonunital case can be established
q p
using the same argument. Let M⊗ → C⊗ → N(∆0 )op exhibit M = M⊗ [0] as left-tensored over the monoidal
∞-category C = C⊗ [1] . In what follows, we will abuse notation by identifying the object M ∈ M with the
constant functor N(I) → M taking the value M , for a variety of categories I.
op

Let r : C[M ]⊗ → N(∆)op denote the projection map. Using the definition of C[M ]⊗ and the description
of the r-coCartesian edges given in the proof of Proposition 2.7.2, we see that the ∞-category Alg(C[M ])
can be identified with the full subcategory of MapN(∆0 )op (N(J)op , M⊗ ) ×Fun(N(∆)op ,M) {M } spanned by those
functors f with the following properties:
(a) Given a collection of integers 0 ≤ i ≤ i0 ≤ j 0 ≤ j ≤ n, let α denote the induced map f ([n], i ≤ j) →
f ([n], i0 ≤ j 0 ). THen q(α) is p-coCartesian.
(b) Given a collection of integers 0 ≤ i ≤ j ≤ n, let β : ([n], ∗) → ([n], i ≤ j) be the map in J corresponding
to the element j ∈ {i, . . . , j}. Then f (β) is (p ◦ q)-coCartesian.
(c) Given a convex map γ : [m] → [n] in ∆ and a pair of integers 0 ≤ i ≤ j ≤ m, the induced map
f ([n], α(i) ≤ α(j)) → f ([m], i ≤ j) is an equivalence in M⊗ .

We let X denote the full subcategory of MapN(∆0 )op (N(J)op , M⊗ ) spanned by those functors f which
satisfy. We observe that if f ∈ X, then f satisfies the following additional condition:

(d0 ) The functor f carries every morphism α : ([m], ∗) → ([n], ∗) to an equivalence in M.

To prove this, we first use the two-out-of-three property to reduce to the case where m = 0. The map α
is then classified by an element i ∈ [n]. We have a commutative diagram in Jop :

([n], i ≤ i)
β
/ ([0], 0 ≤ 0)

γ γ0
 
([n], ∗)
α / ([0], ∗).

Assumptions (b) and (c) guarantees that f carries β, γ, and γ 0 to equivalences in M, so that f carries β 0 to
an equivalence in M as well.
Let J+ denote the category obtained from J by adjoining a new element ([−1], ∗), with

HomJ+ (([−1], ∗), ([n], ∗)) = {∗} HomJ+ (([−1], ∗), ([n], i ≤ j)) = {i, . . . , j}

88
 (
{∗} if u = ([−1], ∗)
HomJ+ (u, ([−1], ∗)) =
∅ otherwise.

The functor ψ 0 : J → ∆0 extends canonically to a functor ψ+ 0

: J+ → ∆0 , with ψ+0
([−1], ∗) = [0]. Let Xe
⊗
denote the full subcategory of MapN(∆0 )op (N(J+ ) , M ) spanned by those functors f which satisfy (a), (b),
op

(c), and the following stronger form of (d0 ):

(d) The functor f carries every morphism α : ([m], ∗) → ([n], ∗) in J+ to an equivalence in M.

Recall that we have identified ∆ with the full subcategory of J spanned by the objects ([n], ∗). We
observe that the inclusion ∆ ' ∆ ×J+ (J+ )([−1],∗)/ ⊆ J ×J+ (J+ )([−1],∗)/ has a right adjoint, given by the
formula
(α : ([−1], ∗) → ([n], •)) 7→ ([n], ∗).
Combining this observation with (d), Lemma T.[Link], Corollary T.[Link], and the observation that
N(∆0 )op is weakly contractible, we deduce:

(i) Every functor f0 ∈ X admits a (p ◦ q)-left Kan extension f ∈ MapN(∆0 )op (N(J+ )op , M⊗ ).

(ii) Let f ∈ MapN(∆0 )op (N(J+ )op , M⊗ ) be arbitrary. Then f ∈ X

e if and only if f0 = f | N(J)op belongs to
X and f is a (p ◦ q)-left Kan extension of f0 .

Recall that we have identified ∆0 with the full subcategory of J spanned by the objects {([n], 0 ≤ n)}n≥0 .
Similarly, we will category ∆× of Notation 1.2.5 with the full subcategory of J spanned by the objects
{([n], i ≤ j)}0≤i≤j≤n . Finally, let J0 denote the full subcategory of J+ spanned by ∆× together with the
object ([−1], ∗), so that we have a sequence of inclusions

∆0 ⊆ ∆× ⊆ J0 ⊆ J+ .

We make the following observations:

(1) For every object ([n], i ≤ j) ∈ ∆0 , the category ∆0 ×∆× (∆× )/([n],i≤j) has a final object, given by the
map ({i, . . . , j}, i ≤ j) → ([n], i ≤ j).
(2) The category ∆× ×J0 J0([−1],∗)/ has an initial object, given by the map ([−1], ∗) → ([0], 0 ≤ 0).

(3) For every object ([n], ∗) ∈ J+ , the category J0 ×J+ J/([n],∗) has a unique (final) object, given by
([−1], ∗) → ([n], ∗).
Applying Lemma T.[Link], we deduce:
(10 ) Every functor f0 ∈ MapN(∆0 )op (N(∆0 )op , M⊗ ) admits a (p ◦ q)-right Kan extension

f ∈ MapN(∆0 )op (N(∆× )op , M⊗ ).

Moreover, an arbitrary functor f ∈ MapN(∆0 )op (N(∆× )op , M⊗ ) is a (p ◦ q)-right Kan extension of the
restriction f | N(∆0 )op if and only if the following condition is satisfied:
(a0 ) For every object ([n], i ≤ j) ∈ ∆× , the canonical map f ([n], i ≤ j) → f ({i, . . . , j}, i ≤ j) is an
equivalence in M⊗ .
(20 ) Every functor f0 ∈ MapN(∆0 )op (N(∆× )op , M⊗ ) admits a (p ◦ q)-left Kan extension

f ∈ MapN(∆0 )op (N(J0 )op , M⊗ ).

Moreover, an arbitrary functor f ∈ MapN(∆0 )op (N(J0 )op , M⊗ ) is a (p ◦ q)-left Kan extension of the
restriction f | N(∆× )op if and only if the following condition is satisfied:

89
 (b0 ) The canonical map f ([0], 0 ≤ 0) → f ([−1], ∗) is an equivalence in M.

(30 ) Every functor f0 ∈ MapN(∆0 )op (N(J0 )op , M⊗ ) admits a (p ◦ q)-right Kan extension

f ∈ MapN(∆0 )op (N(J+ )op , M⊗ ).

Moreover, an arbitrary functor f ∈ MapN(∆0 )op (N(J+ )op , M⊗ ) is a (p ◦ q)-right Kan extension of the
restriction f | N(J0 )op if and only if the following condition is satisfied:

(c0 ) For each n ≥ 0, the canonical map f ([n], ∗) → f ([−1], ∗) is an equivalence in M.

Let Y denote the full subcategory of MapN(∆0 )op (N(J+ )op , M⊗ ) spanned by those functors f which satisfy
(a0 ), (b0 ), and (c0 ), and the following additional condition:
(d0 ) The restriction f | N(∆0 )op belongs to Mod(M).

Applying Proposition T.[Link] repeatedly, we deduce that the restriction map Y → Mod(M) is a
trivial Kan fibration. Our next goal is to show that Y = X. e In other words, if we fix a functor f ∈
⊗
MapN(∆0 )op (N(J+ ) , M ), then f satisfies (a), (b), (c), and (d) if and only if f satisfies (a0 ), (b0 ), (c0 ) and
op

(d0 ).
Suppose first that f satisfies (a), (b), (c), and (d). Condition (a0 ) follows immediately from (c), and
conditions (b0 ) and (c0 ) follow from (d). To verify (d0 ), we first observe that (a) and (c) imply that q ◦
f | N(∆0 )op is an algebra object of C. It remains only to show that if α : ([m], 0 ≤ m) → ([n], 0 ≤ n) satisfies
α(m) = n, then the induced map f ([n], 0 ≤ n) → f ([m], 0 ≤ m) is (p ◦ q)-coCartesian. Using the equivalence
M⊗ ⊗
[m] ' C[m] × M and the fact that (q ◦ f )| N(∆ )
0 op
∈ Alg(C), we may reduce to the case m = 0. In this
case, the desired result follows from (b).
Now let us suppose that f satisfies (a0 ), (b0 ), (c0 ) and (d0 ). Condition (c) follows immediately from (a0 )
Condition (d) follows from (c0 ) using a two-out-of-three argument. To prove (b), we observe that the map
β : ([n], ∗) → ([n], i ≤ j) corresponding to j ∈ {i, . . . , j} fits into a commutative diagram

([−1], ∗)
β1
/ ([0], 0 ≤ 0) β2
/ ({i, . . . , j}, i ≤ j)

β0 β3
 
([n], ∗)
β
/ ([n], i ≤ j).

Condition (c0 ) implies that f (β0 ) and f (β1 ) are equivalences, and condition (a0 ) guarantees that f (β3 ) is an
equivalence. It follows that f (β) is (p ◦ q)-coCartesian if and only if f (β2 ) is (p ◦ q)-coCartesian, which follows
from (d0 ). The verification of (a) is similar: let 0 ≤ i ≤ i0 ≤ j 0 ≤ i ≤ n, and let α : ([n], i0 ≤ j 0 ) → ([n], i ≤ j)
be the induced morphism in J. Then α fits into a commutative diagram

({i0 , . . . , j 0 }, i0 ≤ j 0 )
α1
/ ({i, . . . , j}, i ≤ j)

α0 α2
 
([n], i0 ≤ j 0 )
α / ([n], i ≤ j).

Condition (a0 ) guarantees that f (α0 ) and f (α2 ) are equivalences, so that (q ◦ f )(α) is p-coCartesian if and
only if (q ◦ f )(α2 ) is p-coCartesian. Since α2 is a convex morphism, this follows from assumption (d0 ).
We have a diagram of trivial Kan fibrations X ← X e = Y → Mod(M). For every full subcategory I ⊆ J+ ,
let E(I) denote the full subcategory of Fun(N(I) , M) spanned by those functors which carry each morphism
op

in I to an equivalence in M. Let J00 be the full subcategory of J+ spanned by the objects of ∆+ and the
object ([0], 0 ≤ 0), and J000 the full subcategory spanned by the single object ([0], 0 ≤ 0).

90
 Consider the diagram
φ0 φ1 φ2 φ3
e ×E(J00 ) {M } → X
e ×E(∆) {M } ← X
Alg(C[M ]) ' X ×E(∆) {M } ← X e ×E(J00 ) {M } → Mod(M) ×M {M }.
0

The above arguments imply that φ0 and φ3 are trivial Kan fibrations. We now observe that the categories
J00 , ∆, and J000 all have weakly contractible nerves (for J00 , this follows from the observation that ([0], 0 ≤ 0)
is a final object). Consequently, the restriction maps E(∆) ← E(J00 ) → E(J000 ) ' M are trivial Kan fibrations.
Using this, we deduce that φ1 and φ2 are categorical equivalences.
The functor Mod(M) ×M {M } → Alg(C[M ]) induced by composition with ψ 0 factors as a composition
φ φ1 φ0
Mod(M) ×M {M } → X e ×E(∆) {M } → Alg(C[M ]).
e ×E(J00 ) {M } → X

Consequently, it will suffice to show that φ is a categorical equivalence. We now observe that φ is a section
of φ3 ◦ φ2 , hence a categorical equivalence by the two-out-of-three property.

2.8 Application: Existence of Units

Recall that a nonunital ring is an abelian group (A, +) equipped with an associative multiplication A⊗A → A.
Every associative ring determines a nonunital ring, simply by forgetting the multiplicative identity element.
On the other hand, if A is an associative ring, then the ring structure on A is uniquely determined by
underlying nonunital ring of A. In other words, if A is a nonunital ring which admits a multiplicative
identity 1, then 1 is uniquely determined. The proof is simple: if 1 and 10 are both identities for the
multiplication on A, then 1 = 110 = 10 . Our goal in this section is to prove an ∞-categorical analogue of this
result. More precisely, we will show that if A is a nounital algebra object of a monoidal ∞-category C which
admits a quasi-unit, then A can be extended to an algebra object of C in an essentially unique way.
In ordinary category theory, this is a tautology. However, in the ∞-categorical setting the result is
not quite as obvious; the unit in an algebra object A of C is required to satisfy a hierarchy of coherence
conditions with respect to the multiplication on A, while the definition of a quasi-unit involves only the
homotopy category hC. Nevertheless, we have the following result:
Theorem 2.8.1. Let C be a monoidal ∞-category. Then the restriction map Alg(C) → Algnu (C) induces a
trivial Kan fibration θ : Alg(C) → Algqu (C).
Before giving the proof, let us sketch the main idea. Suppose that A is a nonunital ring, and we wish
to promote A to an associative ring. Let M = A, regarded as a (nonunital) right module over itself. Left
multiplication induces a homomorphism of nonunital algebras φ : A → HomA (M, M ). If A admits a left
unit 1, then A is freely generated by 1 as a right A-module, so that evaluation at 1 induces an isomorphism
HomA (M, M ) ' M . Under this isomorphism, φ corresponds to the map a 7→ a1. If the element 1 ∈ A is
also a right unit, then φ is an isomorphism. On the other hand, EndA (M, M ) is manifestly an associative
ring. To translate this sketch into the setting of higher category theory, we will need the following lemma,
which will be proven at the end of this section:
Lemma 2.8.2. Let C be a monoidal ∞-category, and let A ∈ Algqu (C). There exists an ∞-category M which
is left-tensored over C and an object M ∈ Modqu
A (M) which exhibits A as a morphism object MorM (M, M ).

Proof of Theorem 2.8.1. The map θ is evidently a categorical fibration. It will therefore suffice to show that
θ is a categorical equivalence. We first show that θ is essentially surjective. Let A0 be a quasi-unital algebra
object of C. According to Lemma 2.8.2, we can find an ∞-category M which is left-tensored over C and a
quasi-unital module
M 0 ∈ ModquA0 (M) ⊆ Mod (M)
nu

91
which exhibits A0 as a morphism object MorM (M, M ); here we let M denote the image of M 0 under the
forgetful functor Modqu (M) → M. We have a commutative diagram

Mod(M) ×M {M } / Alg(C[M ]) / Alg(C)

θ0 θ
  
Modnu (M) ×M {M } / Algnu (C[M ]) / Algnu (C).

Let A0 be the image of M 0 in Algnu (C[M ]). To prove that A0 belongs to the essential image of θ, it will
suffice to show that A0 belongs to the essential image of θ0 .
Proposition 2.7.3 and Remark 1.5.6 imply that A0 is a final object of Algnu (C[M ]). On the other hand,
Proposition 2.7.3 an Corollary 1.5.5 imply that Alg(C[M ]) admits a final object A, and Remark 1.5.6 implies
that θ0 (A) is a final object of Algnu (C[M ]). It follows that θ0 (A) ' A0 , so that A0 belongs to the essential
image of θ0 as desired.
We now prove that θ is fully faithful. Fix objects A, B ∈ Alg(C), and set A0 = θ(A) ∈ Algqu (C), B0 =
θ(B) ∈ Algqu (C). We wish to prove that the map MapAlg(C) (B, A) → MapAlgqu (C) (B0 , A0 ) is a homotopy
equivalence. Use Lemma 2.8.2 to choose M 0 ∈ Modqu A0 (M) as above. Consider the diagram

ModA (M) o ModA (M) ×M {M } / {A} ×Alg(C) Alg(C[M ])

ψ
  
Modqu
A0 (M)
o Modqu
A0 (M) ×M {M } / {A0 } ×Algqu (C) Algqu (C[M ]).

The left square is a pullback, and the left vertical map is a trivial Kan fibration (Proposition 2.2.16). The
horizontal maps on the right are both categorical equivalences (Proposition 2.7.6). Using the two-out-of-
three property, we deduce that ψ is a categorical equivalence. Since ψ is also a categorical fibration, it is a
trivial Kan fibration; we may therefore choose M ∈ ModA (M) lifting M 0 .
According to Corollary 2.7.9 and Remark 2.7.10, we have canonical homotopy equivalences

φ : MapAlg(C) (B, A) ' ModB (M) ×M {M }

φnu : MapAlgnu (C) (B0 , A0 ) ' Modnu

B0 (M) ×M {M }.

Let f : B0 → A0 be a map of nonunital algebras, and let N 0 be the corresponding object of Modnu
B0 (M) ×M
f
{M }. Then f is quasi-unital if and only if the composition u : 1C → B → A is homotopic to the unit
of A; here we abuse notation by identifying A and B with their images in C. Using the equivalence A '
MorM (M, M ), we can identify u ∈ π0 MapC (1C , MorM (M, M )) with a point u0 ∈ π0 MapM (M, M ); then f is
quasi-unital if and only if u0 is homotopic to the identity. The map u0 can be identified with the composition
M ' 1C ⊗ M → B ⊗ M → M, so that f is quasi-unital if and only if N 0 is quasi-unital. It follows that φnu
restricts to a homotopy equivalence

φqu : MapAlgqu (C) (B0 , A0 ) ' Modqu

B0 (M) ×M {M }.

We wish to prove that θ induces a homotopy equivalence MapAlg(C) (B, A) → MapAlgqu (C) (B0 , A0 ). In
view of the above identifications, it will suffice to show that the restriction map

g : ModB (M) ×M {M } → Modqu

B0 (M) ×M {M }

is a homotopy equivalence. Proposition 2.2.16 implies that g is a trivial Kan fibration.

Definition 2.8.3. Let C be a monoidal ∞-category, let M be an ∞-category which is left-tensored over C.
We let Modqu (M) denote the full subcategory of Modnu (M) ×Algnu (C) Algqu (C) spanned by those nonunital
A-modules M satisfying the following condition:

92
 (∗) Let u : 1C → A be a quasi-unit for A. Then the composition
u
M ' 1C ⊗ M → A ⊗ M → M
is an equivalence.
Remark 2.8.4. Let C be a monoidal ∞-category, let M be an ∞-category which is left-tensored over C, let
A be an algebra object of C and let A0 denote the underlying nonunital algebra. Then the fiber
Modqu (M) ×Algqu (C) {A0 } ⊆ Modnu
A (M)

coincides with the full subcategory Modqu nu

A (M) ⊆ ModA (M) introduced in Definition 2.8.3.
Corollary 2.8.5. Let C be a monoidal ∞-category, let M be an ∞-category which is left-tensored over C.
Then:
(1) The restriction functor θ : Mod(M) → Modqu (M) ×Algqu (C) Alg(C) is a trivial Kan fibration.
(2) The restriction functor Mod(M) → Modqu (M) is a trivial Kan fibration.
Proof. It is clear that (2) follows from (1) and Theorem 2.8.1. It will therefore suffice to prove (1). Consider
the diagram
Modqu (M) / Modnu (M)

p p0
 
Algqu (C) / Algnu (C).

According to Remark 2.3.4, the map p0 is a Cartesian fibration; moreover, a morphism f : M → M 0 in

Modnu (M) is p0 -Cartesian if and only if it induces an equivalence in M. Suppose that f is the image of
0
a morphism f : M → M in Modnu (M) ×Algqu (C) Algnu (C). If f is p0 -Cartesian, we deduce easily that
0
M ∈ Modqu (M) if and only if M ∈ Modqu (M). It follows that p is also a Cartesian fibration, and that a
0
morphism f : M → M in Modqu (M) is p-Cartesian if and only if it induces an equivalence in M.
Consider next the diagram
θ0 / Modqu (M) ×Algqu (C) Alg(C)
Mod(M)
q q0
 
Alg(C) Alg(C).

The map q 0 is a pullback of p. It follows that q 0 is a Cartesian fibration, and that a morphism of
Modqu (M) ×Algqu (C) Alg(C) is q 0 -Cartesian if and only if its image in M is an equivalence. Corollary 2.3.3
implies that q is also a Cartesian fibration, and that θ0 carries q-Cartesian edges to q 0 -Cartesian edges. Ac-
cording to Corollary T.[Link], θ is a categorical equivalence provided that it induces a categorical equivalence
of fibers ModA (M) → Modqu A (M) over every object A ∈ Alg(C). The desired result now follows immediately
from Proposition 2.2.16.
Remark 2.8.6. Theorem 2.8.1 can be considered as a special case of Corollary 2.8.5. To see this, let
q : C⊗ → N(∆)op be a monoidal ∞-category. The identity map C⊗ → C⊗ exhibits M = C⊗[0] as left-tensored
over C = C⊗
[1] . We have a commutative diagram

Mod(M) / Modqu (M) / Modnu (M)

  
Alg(C) / Algqu (C) / Algnu (C)

where the vertical maps are isomorphisms.

93
 We now return to the proof of Lemma 2.8.2. The idea is to take M to be the ∞-category of nonunital
right A-modules in C, and M ∈ M to be A itself, regarded as a right A-module.
Proof of Lemma 2.8.2. Let ∆nu nu
+ be the category obtained by adjoining an initial object [−1] to ∆ . Let
nu
ψ : ∆ × ∆+ → ∆ denote the join functor

ψ([m], [n]) = [m] ? [n] ' [m + n + 1],

and let ψ0 = ψ| ∆ × ∆nu . We observe that there is a canonical natural transformation of functors α : π2 →
ψ0 , where π2 : ∆ × ∆nu → ∆ denotes projection onto the second factor (followed by the inclusion ∆nu ⊆ ∆).
Set a
T = (N(∆nu+ )
op
× {0}) (N(∆nu )op × ∆1 )
N(∆nu )op ×{0}

so that α determines a map of simplicial sets h : N(∆)op × T → N(∆)op .

⊗
Let p : C⊗ → N(∆)op exhibit C = C⊗
[1] as a monoidal ∞-category. We define a simplicial set M equipped
e
⊗
with a map M e → N(∆)op by the following universal property: for every simplicial set K equipped with a
map K → N(∆)op , we have a canonical bijection of HomN(∆)op (K, Me ⊗ ) with the set of maps f : K ×T → C⊗
for which the diagram
K × (N(∆nu )op × {1}) / N(∆nu )op

A
 
K ×T
f
/ C⊗

p
 
N(∆)op × T
h / N(∆)op

is commutative. ⊗
e classifies a map of simplicial sets f : T → C⊗ . Let M⊗ be the full simplicial subset of
A vertex of M
⊗
Me spanned by those vertices corresponding to functors f which satisfy the following additional conditions:

(i) For every object [n] ∈ ∆nu , the restriction f |{[n]} × ∆1 is a p-coCartesian edge of C⊗ .

(ii) For every convex morphism α : [m] → [n] in ∆nu such that α(0) = 0, the morphism f (α × {0}) is
p-coCartesian.
(iii) For every object [n] ∈ ∆nu , if α : [−1] → [n] denotes the unique morphism in ∆nu
+ , then f (α × {0}) is
q-coCartesian.

It is not difficult to see that the projection M⊗ → N(∆)op is a coCartesian fibration. Restriction to
the object {[−1]} × {0} determines a map q : M⊗ → C⊗ , which exhibits M = M⊗ [0] as left-tensored over C.
Restriction to the simplicial subset N(∆nu )op × ∆1 ⊆ T determines a trivial Kan fibration M → Modnu,R
A (C).
h A
Finally, we observe that the composition N(∆nu )op × T → N(∆nu )op → C⊗ determines a quasiunital left
A-module M ∈ Modnu A (M).
In what follows, we will abuse notation by identifying the nonunital algebra A and the A-module M with
their images in C and M, respectively. To complete the proof, it will suffice to show that the (left) action of A
on M θ : A⊗M → M exhibits A as a morphism object MorM (M, M ). In other words, we must show that for
every object C ∈ C, composition with θ induces a homotopy equivalence MapC (C, A) → MapM (C ⊗ M, M ).
We must show that for every Kan complex K, θ induces a bijection

[K, MapC (C, A)] → [K, MapM (C ⊗ M, M )];

94
here [K, X] denotes the set of maps from K to X in the homotopy category H of spaces. Replacing C by
Fun(K, C) and A by the nonunital algebra A0 ∈ Algnu (Fun(K, C)) ' Fun(K, Algnu (C)) corresponding to the
constant map K → {A} ⊆ Algnu (C), we can reduce to the case where K = ∅. In other words, it will suffice
to show composition with θ induces a bijection q : π0 MapC (C, A) → π0 MapM (C ⊗ M, M ).
Our next step is to construct an inverse to q. Let u : 1C → A be a quasi-unit. Let φ : C ⊗ M → M be
an arbitrary morphism in M. Then φ determines a map C ⊗ A → A in C. Let q 0 (φ) denote the composition
u
C ' C ⊗ 1C → C ⊗ A → A.
We may view q 0 as a map of sets from π0 MapM (C ⊗ M, M ) to π0 MapC (C, A). The composition q 0 ◦ q :
π0 MapC (C, A) → π0 MapC (C, A) is induced by the map ru : A → A given by right multiplication by u.
Since u is a right unit of A, we deduce that q 0 ◦ q is the identity. In particular, q is injective.
To complete the proof, it will suffice to show that q is surjective. For this, we use the results of §2.7.
According to Proposition 2.7.3, if C is an object of C, then giving a map φ : C ⊗ M → M is equivalent to
lifting C to an object C e ∈ C[M ]. In particular, the left action of A on M gives rise to a canonical element
A ∈ C[M ]. Then φ belongs to the image of q if and only if there exists a map C
e e→A e in C[M ]. Consequently,
the surjectivity of q is equivalent to the following assertion:
e ∈ C[M ], there exists a morphism C
(∗) For every object C e in C[M ].
e→A

Proposition 2.7.3 asserts that the projection C[M ] → C is a right fibration. Consequently, the quasi-unit
u : 1C → A can be lifted to a map u e:e 1C → Ae in C[M ]. Since u is a left unit of A, the object e
1C classifies a
map v : 1C ⊗ M → M which determines an equivalence in C. Using Remark 2.3.4, we conclude that v is an
equivalence in M. It follows from Remark 2.7.4 that e 1C is an invertible object of C[M ], so that the functor
C
e 7→ C
e⊗e 1C is an equivalence from C[M ] to itself. Consequently, condition (∗) is equivalent to:
e ∈ C[M ], there exists a morphism C
(∗0 ) For every object C e ⊗e e in C[M ].
1C → A
In view of the existence of u 1C → A,
e:e e it will suffice to prove the following slightly stronger assertion:
e ∈ C[M ], there exists a morphism C
(∗00 ) For every object C e⊗A e in C[M ].
e→A

Applying Proposition 2.7.3 again, we see that (∗00 ) is equivalent to the following assertion: for every map
φ : C ⊗ M → M , there exists a commutative diagram
C ⊗A⊗M / C ⊗M

φ
 
A⊗M / M;

in M, where the horizontal arrows are given by the canonical action of A on M . This is a straightforward
consequence of our construction of M and M ; we leave the details to the reader.

3 Monads and the Barr-Beck Theorem

Suppose given a pair of adjoint functors C o
F / D between ordinary categories. Then:
G

(A) The composition T = G◦F has the structure of a monad on C; that is, an algebra object of the category
Fun(C, C) of endofunctors of C. Here the unit map idC → T given by the unit of the adjunction between
F and G, and the product is given by the composition
T ◦ T = G ◦ (F ◦ G) ◦ F → G ◦ idD ◦F = T
where the second map is given by a compatible counit v for the adjunction between F and G.

95
(B) For every object D ∈ D, the object G(D) has the structure of a module over the monad T , given by
v
the map T G(D) = ((G ◦ F ) ◦ G)(D) = (G ◦ (F ◦ G))(D) → G(D). This construction determines a
functor θ from D to the category of T -modules in C.
(C) In many cases, the functor θ is an equivalence of categories. The Barr-Beck theorem provides necessary
and sufficient conditions on the functor G to guarantee that this is the case. We refer the reader to
[52] for a detailed statement (or to Theorem 3.5.1 for our ∞-categorical version, which subsumes the
classical statement).
Our goal in this section is to obtain ∞-categorical generalizations of assertions (A) through (C). Our
first step is to define the notion of a monad on an ∞-category C. We will do so by introducing a monoidal
structure on the ∞-category Fun(C, C), which is determined by the (strictly) associative composition product
on Fun(C, C). The details of this construction are outlined in §3.1, using a relative version of the nerve
construction which is of some independent interest. We can then define a monad on C to be an algebra
object of Fun(C, C).
The next step is to establish an analogue of assertion (A). Suppose given a pair of adjoint functors
F /
Co D between ∞-categories. The composition T = G ◦ F is equipped with natural transformations
G
id → T , T ◦ T → T , given by composition with unit and counit maps for the adjunction between C and
D. Moreover, the classical proof of the associative law can be carried through unchanged in the homotopy
category, so that T can be viewed as an algebra object of the monoidal category hFun(C, C). However,
endowing T with the structure of a monad on C is more difficult. To carry it out, we need to construct an
elaboration of the monoidal structure on Fun(C, C), which involves not only C but also D. We will outline
this construction in §3.2. Using this construction, we will define an adjunction datum: roughly speaking, this
is a structure relating the ∞-categories C and D that simultaneously determines a pair of adjoint functors
F /
Co D and a monad T on C, where T is equivalent to G ◦ F . Our main result, Theorem 3.2.10, asserts
G
that every functor F : C → D which admits a right adjoint can be extended, in an essentially unique way,
to an adjunction datum. Surprisingly, this turns out to be quite a bit harder to prove than the Barr-Beck
theorem itself; we will therefore postpone the proof until §3.6.
Once we have extracted the relevant monad T ∈ Alg(Fun(C, C)), we can consider the ∞-category of
T -modules ModT (C). In §3.3, we will show that there is a canonical functor D → ModT (C), which is
well-defined up to contractible ambiguity. More precisely, we will construct a diagram
φ ψ
D ← ADatU (C, D) → ModT (C),

and prove that φ is a trivial Kan fibration (Corollary 3.3.6). This is our ∞-categorical analogue of (B).
In §3.5, we will prove an ∞-categorical version of the Barr-Beck theorem (Theorem 3.5.1), which gives a
necessary and sufficient condition for the functor ψ appearing above to be an equivalence of ∞-categories.
The formulation of Theorem 3.5.1 rests on the notion of a split simplicial object in an ∞-category, which we
will discuss in §3.4.

3.1 ∞-Categories of Endofunctors

Let M be an ∞-category. In §2, we introduced the definition of a monoidal ∞-category C equipped with a
left action of C on M. In this section, we will construct the universal example of such a left action, given by
C = Fun(M, M). It is clear that C is a monoid object in the ordinary category of simplicial sets, and that M
is equipped with a (left) action of C in the classical sense. Our first goal is to convert this data into a left
action of a monoidal ∞-category, in the sense of Definition 2.1.1.
Notation 3.1.1. Let M be a simplicial set. We define functors E, E : ∆op → Set∆ as follows:

96
 (1) If n ≥ 0 and K is a simplicial set, then elements HomSet∆ (K, E([n])) can be identified with collections
of maps {σij ∈ HomK (K × M, K × M)}0≤i≤j≤n such that each σii is the identity, and σij ◦ σjk = σik
for 0 ≤ i ≤ j ≤ k ≤ n.
(2) If n ≥ 0 and K is a simplicial set, then elements HomSet∆ (K, E([n])) can be identified with collections
of maps {σij ∈ HomK (K × M, K × M)}0≤i≤j≤n , {τi ∈ HomK (K, K × M)}0≤i≤n such that each σii is
the identity, σij ◦ σjk = σik for 0 ≤ i ≤ j ≤ k ≤ n, and τi = σij ◦ τj for 0 ≤ i ≤ j ≤ n.

(3) For each morphism α : [m] → [n] in ∆, the associated maps

E([n]) → E([m]), E([n]) → E([m])

are given by composition with α.

⊗
We set End⊗ (M) = NE (∆op ) and End (M) = NE (∆op ) (see Definition T.[Link]). We observe that
there is a natural transformation of functors E → E, given by forgetting the morphisms τi ; this natural
⊗
transformations induces a map of simplicial sets End (M) → End⊗ (M).

Proposition 3.1.2. Let M be an ∞-category. Then:

(1) The map p : End⊗ (M) → N(∆)op determines a monoidal structure on the ∞-category Fun(M, M) '
End⊗
[1] (M).

⊗ ⊗
(2) The map q : End (M) → End⊗ (M) exhibits M ' End[0] (M) as left-tensored over Fun(M, M).

Proof. Proposition T.[Link] implies that p and p ◦ q are coCartesian fibrations, while Lemma T.[Link]
impies that q is a categorical fibration. We observe that the fiber of p over a vertex [n] ∈ N(∆)op is
isomorphic to Fun(M, M)n , while the fiber of q over [n] is isomorphic to Fun(M, M)n × M. It is easy to see
that these identifications are compatible with the associated functors, so that p and q satisfies the hypotheses
of Definitions 1.1.2 and 2.1.1, respectively.

We will refer to the monoidal structure on Fun(M, M) supplied by Proposition 3.1.2 as the composition
monoidal structure.
Definition 3.1.3. Let M be an ∞-category. A monad on M is an algebra object of Fun(M, M) (with respect
to the composition monoidal structure). If T is a monad on M, we let ModT (M) denote the associated ∞-
category of T -modules in M.

Remark 3.1.4. More informally, a monad on an ∞-category M consists of an endofunctor T : M → M

equipped with maps 1 → T and T ◦ T → T which satisfy the usual unit and associativity conditions, up
to coherent homotopy. A T -module is then an object M ∈ M equipped with a structure map T (M ) → M
which is compatible with the algebra structure on T , again up to coherent homotopy.

Remark 3.1.5. Let M be an ∞-category, which we regard as an object of Cat∞ , and regard Cat∞ as endowed
with the Cartesian monoidal structure. According to Corollary 2.6.6, giving a monoidal ∞-category C over
which M is left-tensored is equivalent to lifting M to a (left) module object of Cat∞ . In view of Proposition
2.7.6, this is equivalent to producing an algebra object of the monoidal ∞-category Cat∞ [M]. In particular,
⊗
End (M) determines an algebra object of A ∈ Alg(Cat∞ [M]).
For every ∞-category D, the action of End(M) on M determines a homotopy equivalence (even an
isomorphism of simplicial sets) MapCat∞ (D, End(M)) → MapCat∞ (D × M, M). Combining Proposition 2.7.3
with Corollary 1.5.5, we deduce that A is a final object of Alg(Cat∞ [M]). In other words, End⊗ (M) is
universal among monoidal ∞-categories which act on M.

97
3.2 Adjunction Data

Suppose given a pair of adjoint functors C o

F / D between ∞-categories C and D. Our objective in this
G
section is to associate to the pair (F, G) a monad T ∈ Alg(Fun(C, C)), given informally by the formula
T = G ◦ F.
The basic obstacle we need to overcome is that an adjunction is overdetermined by the pair of functors
F and G. Specifying only the functor F : C → D determines a right adjoint G to F up to contractible
ambiguity, provided that G exists. If we also specify the functor G, we should really include additional data
which identifies G with an adjoint to F . Such data is provided by either a unit u : idC → G ◦ F or a counit
v : F ◦ G → idD for the adjunction. However, if we specify both a unit and a counit, then the adjunction is
again overdetermined. In classical category theory, one imposes a condition on a unit and counit: they are
said to be compatible if the composite transformations
u v
α : F → F ◦ (G ◦ F ) = (F ◦ G) ◦ F → F
u v
β : G → (G ◦ F ) ◦ G = G ◦ (F ◦ G) → G
coincide with the identity. In the higher categorical setting, we should instead require the existence of
homotopies h : α ' idF , h0 : β ' idG . Once again, we obtain the correct theory if we specify either h or
h0 , but specifying both will overdetermine the adjunction. This leads us to formulate further compatibilities
between h and h0 , and so forth. There are two different strategies for dealing with the situation:

(1) Specify a minimal amount of data: for example, the single functor F .
(2) Specify all of the relevant data: the functors F and G, the unit u and counit v, the homotopies h and
h0 , and all of their higher-dimensional relatives.

For most applications, approach (1) is entirely sufficient. However, if we wish to extract a monad from
the adjoint pair (F, G), then we are forced to adopt approach (2). Our primary goal in this section is to
make precise sense of (2), by introducing the notion of an adjunction datum between the ∞-categories C and
D. Our main result Theorem 3.2.10, which asserts that (1) and (2) are actually equivalent to one another:
in other words, if F : C → D is a functor which admits a right adjoint, then F can be promoted (in an
essentially unique way) to an adjunction datum. The proof of this result is very technical (probably the
most difficult result in this paper) and will be given in §3.6.
We begin with some generalities. Recall that the classical theory of monoidal categories can be regarded
as a special case of the theory of (weak) 2-categories. More precisely, giving a monoidal category (C, ⊗)
is essentially the same thing as giving a 2-category D having only a single, fixed object X ∈ D. The
correspondence assigns to X ∈ D the category HomD (X, X), with monoidal structure given by composition
in D. We can obtain a generalization of the theory of monoidal categories by allowing D to have many
objects. We now describe the analogous generalization of the theory of monoidal ∞-categories.
Definition 3.2.1. Let S be a set of symbols. We define a category ∆S as follows. An object of ∆S
consists of a pair ([n], c), where [n] ∈ ∆ and c : [n] → S is an arbitrary map. Given a pair of objects
([m], c), ([n], c0 ) ∈ ∆S , we set

Hom∆S (([m], c), ([n], c0 )) = {α ∈ Hom∆ ([m], [n]) : (∀0 ≤ i ≤ m)[c(i) = (c0 ◦ α)(i)]}.

We will think of an object of ∆S as a (nonempty) finite sequence [c(0), c(1), . . . , c(n)] of elements of S;
in other words, as a nonempty word in the alphabet S.
One can define an S-colored monoidal ∞-category to be a coCartesian fibration C⊗ → N(∆S )op , satisfying
an appropriate analogue of the condition (∗) of Definition 1.1.2. We can think of an S-colored monoidal
∞-category as encoding the structure of an (∞, 2)-category whose objects are in bijection with the set S.
We will refrain from going into the details, since we are primarily interested in only a single example.

98
Notation 3.2.2. Let C and D be simplicial sets, and let S = {C, D}. We define a functor E : ∆op
S → Set∆
as follows:
(1) Let n ≥ 0 be a finite nonempty linearly ordered set, and let c : [n] → {C, D} be a map, and let
K ∈ Set∆ be arbitrary. Then elements of HomSet∆ (K, E([n], c)) can be identified with collections of
maps {σij ∈ HomK (K × c(j), K × c(i))}0≤i≤j≤n such that σii = id for i ∈ J, and σij ◦ σjk = σik for
i ≤ j ≤ k.

(2) Given a map f : ([m], c) → ([n], c0 ) in ∆S , the associated map E([n], c0 ) → E([m], c) is given by
pullback along f .
Remark 3.2.3. In Notation 3.2.2, and all that follows, we will implicitly assume that C 6= D, so that the
set S = {C, D} contains exactly two symbols. (This is no loss of generality, since we can always achieve this
state of affairs by replacing C or D by an isomorphic simplicial set.) This is purely a matter of notational
convenience.
Definition 3.2.4. Let C and D be ∞-categories, and let S = {C, D}. We let End⊗ (C, D) denote the relative
nerve NE (∆op
S ).

Remark 3.2.5. The projection map q : End⊗ (C, D) → N(∆S )op is an example of an S-colored monoidal
∞-category. One can think of this monoidal ∞-category as follows: the collection of all ∞-categories really
constitutes an (∞, 2)-category, since we can associate to every pair of ∞-categories E and E0 an ∞-category
Fun(E, E0 ). Then q encodes the full subcategory spanned by the pair of ∞-categories C and D.
Definition 3.2.6. Let C and D be ∞-categories, and let S = {C, D}. We will say that a morphism
α : ([m], c) → ([n], c0 ) in ∆S is C-convex if, whenever j ∈ [n] satisfies c0 (j) = C and α(i) ≤ j ≤ α(i0 ) for
suitably chosen i, i0 ∈ [m], there exists a unique j0 ∈ [m] such that α(j0 ) = j. We will say that a section U
of the projection q : End⊗ (C, D) → N(∆S )op is an adjunction datum if it carries every C-convex morphism
in ∆S to a q-coCartesian morphism of End⊗ (C, D). We let ADat(C, D) denote the full subcategory of
MapN(∆S )op (N(∆S )op , End⊗ (C, D)) spanned by the adjunction data.
Remark 3.2.7. We can identify ∆ ' ∆{C} with the full subcategory of ∆S spanned by the objects ([n], c),
where c : [n] → {C, D} takes the constant value C. Under this identification, a morphism in ∆ is C-convex
if and only if it is convex in the sense of Definition 1.1.7.
The fiber product End⊗ (C, D) ×N(∆S )op N(∆)op is canonically isomorphic to the monoidal ∞-category
End⊗ (C) defined in §3.1. Restriction determines a functor ADat(C, D) → Alg(End(C)), so that every ad-
junction datum determines a monad on C.
We can use the same argument to produce a restriction map ADat(C, D) → Alg(End(D)). However, this
map is less interesting: the monad on D induced by an adjunction datum is always equivalent to the initial
object of Alg(End(D)) (since every morphism of ∆{D} is C-convex in ∆S .
We now proceed to unwind the details of Definition 3.2.6.
Remark 3.2.8. Let C and D be ∞-categories, let S = {C, D}, and let U ∈ ADat(C, D). Let us agree to
denote an object of ∆S by a finite (nonempty) string of elements of S. Then:

(i) Evaluation of U at the object [D, C] ∈ ∆S determines a functor F : C → D.

(ii) Evaluation of U at the object [C, D] ∈ ∆S determines another functor G : D → C.
(iii) Evaluation of U at the object [D, D] ∈ ∆S determines a functor i : D → D. The morphism [D, D] →
[D] determines a natural transformation idD → i in Fun(D, D). Since the morphism [D, D] → D is
C-convex, this natural transformation is an equivalence, so that i is (canonically) equivalent to the
identity functor on D.

99
(iv) Evaluation of U at the object [C, C] ∈ ∆S determines a functor T : C → C. The morphism [C, C] → [C]
determines a natural transformation idC → T . This transformation is generally not an equivalence.
Rather, it is the unit map of a monad structure on T (see Remark 3.2.7).
(v) Evaluation of U at the object [D, C, D] ∈ ∆S determines a diagram

F 0 ◦G0 /D
DA >
AA 0 }
AAG F 0 }}}
AA }}
A }}
C.

Moreover, applying s to the inclusions of [D, C], [D, C], and [D, D] into [D, C, D] determines natural
transformations α : F 0 → F , β : G0 → G, and F 0 ◦ G0 → i. Since the first two of these inclusions are
C-convex, we deduce that α and β are equivalences. Combining these three transformations with (iii),
we obtain a natural transformation v : F ◦ G → idD , well-defined up to homotopy.
(vi) Evaluation of U at the object [C, D, C] ∈ ∆S determines a diagram

G00 ◦F 00 /C
CA
AA 00 }}>
AAF G }}
00
AA }
A }}}
D.

Moreover, applying s to the inclusions of [C, D], [C, C] and [D, C] into [C, D, C] yields natural trans-
formations F 00 → F , G00 → G, and G00 ◦ F 00 → T . All three of these inclusions are C-convex, so the
induced natural transformations are all equivalences. In particular, the transformation idC → T of (iv)
determines a natural transformation u : idC → G ◦ F , which is well-defined up to homotopy.

Lemma 3.2.9. Let C and D be ∞-categories, and suppose given an adjunction datum U ∈ ADat(C, D). Let
F : C → D, G : D → C, u : idC → G ◦ F , and v : F ◦ G → idD be defined as in Remark 3.2.8. Then u
is the unit of an adjunction between F and G, and v is the counit of an adjunction between F and G. In
particular, F and G are adjoint to one another.
Proof. Let S = {C, D}. We will show that u and v determine an adjunction between the underlying functors
between the homotopy categories hC and hD. For this, it suffices to show that the compositions
u v
F → F ◦ (G ◦ F ) = (F ◦ G) ◦ F → F
u v
G → (G ◦ F ) ◦ G = G ◦ (F ◦ G) → G
are homotopic to the identity.
To prove the first claim, we consider the commutative diagram

[D, C]
qq MMM
qqq MMM
qqq MMM
x q
q M&
[D, D, C] [D, C, C]
MMM q JJ
ttt MM qqq JJ
t M q JJ
ttt MMM qqq JJ
ztt M& xqq J$
[D, C] [D, C, D, C] [D, C]

100
in ∆S , where each of the corresponding maps of linearly ordered sets preserves both the initial and final
objects. Applying the functor U and evaluating at those initial and final objects, we obtain a diagram in
the homotopy category hFun(C, D), which is equivalent to

: F gOOO
tt OOO
id ttt OO
tt α OOO
tt O
t
F d J F ◦ G ◦dJF
~? JJ
JJ ooo7 JJ
id ~~ JJ oo JJ
~ o
id
J
~~ v JJJ ooo u JJJ
~ oo J
F F ◦G◦F F.

The desired result now follows from the observation that the composition of α with u is (homotopic to) the
identity.
The proof of the second claim is similar, but we consider instead the diagram

[C, D]
q MMM
qqqqq MMM
MMM
q
x qq
q M&
[C, D, D] [C, C, D]
t MMM qq JJ
t MMM qq JJ
ttt M qq JJ
tt MM qq JJ
ztt M& xqq J%
[C, D] [C, D, C, D] [C, D].

Applying s and evaluating at the initial and final objects, we obtain a diagram in hFun(D, C) which is
equivalent to
: G gOOO
tt OOO
id ttt OOO
t OOO
ttt β
t
G dII G 7 ◦ F ◦dJG
? II pp JJ
id  II id ppp JJ
J
 III pp p u JJJ

 I ppp J
v

G G◦F ◦G G,
and the desired result follows from the observation that the composition of u with β is the identity on G.
In the situation of Lemma 3.2.9, either the unit u or the counit v are sufficient to identify F with the
left adjoint of G. An adjunction datum U therefore supplies two such identifications. However, these two
identifications are compatible with one another, up to coherent homotopy: this compatibility is encoded in
the values U on more complicated objects of ∆S . The result is that, up to a contractible space of choices,
the entire diagram U is determined by the functor F :
Theorem 3.2.10. Let C and D be ∞-categories, let S = {C, D}, and let θ : ADat(C, D) → Fun(C, D) be
given by evaluation at [D, C] ∈ ∆S . Then θ induces a trivial Kan fibration ADat(C, D) → Fun0 (C, D), where
Fun0 (C, D) denotes the subcategory of Fun(C, D) whose objects are functors from C to D which admit right
adjoints, and whose morphisms are equivalences of functors. In particular, ADat(C, D) is a Kan complex.
In other words, if a functor F : C → D admits a right adjoint G, then F can be extended (in an essentially
unique way) to an adjunction datum U ∈ ADat(C, D). Invoking Remark 3.2.7, we see that F determines a
monad on C, whose underlying functor from C to C can be identified with G ◦ F (Remark 3.2.8). We may
therefore regard Theorem 3.2.10 as an ∞-categorical analogue of assertion (A) (see the introduction to §3).
The proof will be given in §3.6.

101
3.3 Pointed Adjunction Data
Our goal in this section is to establish an ∞-categorical analogue of assertion (B) of §3. Let C and D
be a pair of ∞-categories, fixed throughout this section, and let S = {C, D}. For each adjunction datum
U ∈ ADat(C, D), we will construct a diagram D ← ADatU (C, D) → ModT (C), where T is the monad on C
determined by U . Moreover, we will show that the left map is a trivial Kan fibration (Corollary 3.3.6), so
that U determines a map D → ModT (C), which is well-defined up to a contractible space of choices.
Our first step is to introduce an elaboration of the simplicial set End⊗ (C, D).
Notation 3.3.1. We define a functor E : ∆op
S → Set∆ as follows:

(1) Let n ≥ 0, let c : [n] → {C, D} be a map, and let K be a simplicial set. Then elements of
HomSet∆ (K, E([n], c)) can be identified with collections of maps

{σij ∈ HomK (K × c(j), K × c(i))}0≤i≤j≤n , {τi ∈ HomK (K, K × c(i)}0≤i≤n

such that σii = id for 0 ≤ i ≤ n, σij ◦ σjk = σik for 0 ≤ i ≤ j ≤ k ≤ n, and σij ◦ τj = τi for
0 ≤ i ≤ j ≤ n.
(2) Given a map f : ([m], c) → ([n], c0 ) in ∆S , the associated map E([n], c0 ) → E([m], c) is given by
pullback along the induced map [m] → [n].
⊗
We let End (C, D) = NE (∆op
S ) (see Definition T.[Link]).

Remark 3.3.2. Let C and D are ∞-categories. Proposition T.[Link] implies that the projection
⊗
End (C, D) → N(∆S )op
⊗
is a coCartesian fibration of simplicial sets. We observe that there is a canonical projection End (C, D) →
End⊗ (C, D), given by forgetting the maps τi . Moreover, if we identify ∆ with the full subcategory of ∆S
spanned by those functions which take the constant value C, then the fiber product
⊗
End (C, D) ×N(∆S )op N(∆)op
⊗
is canonically isomorphic to the simplicial set End (C) defined in §3.1.
⊗ p q
Definition 3.3.3. Consider the maps End (C, D) → End⊗ (C, D) → N(∆S )op . A pointed adjunction datum
⊗
is a map U : N(∆S )op → End (C, D) with the following properties:
(1) The map U is a section of q ◦ p; that is, q ◦ p ◦ U = id.
(2) The map p ◦ U : N(∆S )op → End⊗ (C, D) is an adjunction datum.
(3) Suppose that α : ([m], c) → ([n], c0 ) is a morphism in ∆S such that α(m) = n. Then U (α) is a locally
⊗
p-coCartesian morphism in End (C, D).
(4) Let α be the unique morphism from [C] to [C, D] in ∆S . Then U (α) is a locally p-coCartesian morphism
⊗
in End (C, D).
⊗
We let ADat(C, D) denote the full subcategory of MapN(∆S )op (N(∆S )op , End (C, D)) spanned by the pointed
adjunction data.
Remark 3.3.4. Restriction to the subcategory N(∆{C} )op ⊆ N(∆S )op determines a functor ADat(C, D) →
⊗
Mod(C), where we regard C as left-tensored over End⊗ (C). Composition with the projection End (C, D) →
End⊗ (C, D) determines another functor ADat(C, D) → ADat(C, D). Finally, evaluation on the object [D] ∈
∆S determines a functor ADat(C, D) → D.

102
 Our main result can be stated as follows:
Proposition 3.3.5. Let C and D be ∞-categories, and let θ : ADat(C, D) → ADat(C, D) × D be the product
of the maps described in Remark 3.3.4. Then θ is a trivial fibration of simplicial sets.
Corollary 3.3.6. Let C and D be ∞-categories, let s ∈ ADat(C, D), and let ADats (C, D) denote the fiber
product ADat(C, D) ×ADat(C,D) {s}. Then the evaluation map ADats (C, D) → D is a trivial fibration of
simplicial sets.
The proof will require the following somewhat technical lemma:
⊗
Lemma 3.3.7. Consider the projection map p : End (C, D) → End⊗ (C, D). Then:
(1) The map p is a locally coCartesian fibration.
(2) Suppose that α : ([m], c) → ([n], c0 ) is a morphism in ∆S which induces a convex morphism [m] → [n],
⊗
and let f be a morphism in End (C, D) which projects to α. Then f is p-coCartesian if and only if f
is locally p-coCartesian.
(3) Suppose that α : ([m], c) → ([n], c0 ) is a morphism in ∆S such that α(m) = n, and let f be a morphism
⊗
in End (C, D) which projects to α. The following conditions are equivalent:
(i) The morphism f is locally p-coCartesian.
(ii) The morphism f is locally p-Cartesian.
(iii) The morphism f is p-Cartesian.
Proof. Assertion (1) follows from Proposition T.[Link]. The map p is generally not a coCartesian fibration,
because the class of locally p-coCartesian morphisms is not stable under composition. Nevertheless, suppose
given a 2-simplex
Y
~> @@@ 00
e0 ~~ @@e
~~~ @@
~ 
X
e /Z
⊗
in End (C, D) where the morphisms e0 and e00 are locally p-coCartesian, covering a diagram

([m], c0 )
s dJJ
s JJ
sss JJ
s
s 0 JJ
yss α α00 J
00 o
([n], c ) α
([k], c)

in the category ∆S . Using the description of the class of locally p-coCartesian morphisms provided by
Proposition T.[Link], we conclude that e is locally p-coCartesian provided that the following condition is
satisfied:
(∗) The map α determines a bijection
{i ∈ [m] : i > α00 (k)} → {j ∈ [n] : α(k) < j ≤ α0 (m)}.
(This implies, in particular, that if α00 (k) < i ≤ m, then α(k) < α(i).)
We note that condition (∗) is always satisfied if the map α0 induces a convex map [m] ,→ [n]. Assertion
(2) now follows from Lemma T.[Link]. Similarly, condition (∗) is automatically satisfied if α00 (k) = m. Using
Lemma T.[Link], we deduce the equivalence (ii) ⇔ (iii) of (3). The equivalence (i) ⇔ (ii) follows from the
⊗ ⊗
observation that if f : X → Y is as in (3), then the associated functor End (C, D)X → End (C, D)Y is an
equivalence of ∞-categories.

103
Remark 3.3.8. In the situation of Definition 3.3.3, condition (4) is can be replaced by the following
apparently stronger condition:
(40 ) Let n > 0, let c : [n] → {C, D} satisfy c(n) = D, let c0 = c|{0, . . . , n − 1}, and let α : ([n − 1], c0 ) →
⊗
([n], c) be the inclusion. Then s(α) is a p-coCartesian morphism in End (C, D).
It is clear that (40 ) ⇒ (4). Conversely, suppose that U satisfies the conditions of Definition 3.3.3, and let α
be as in the statement of (40 ). We wish to prove that U (α) is p-coCartesian. Lemma 3.3.7 shows that U (α)
is p-coCartesian if and only if U (α) is locally p-coCartesian. Using assumption (2), we deduce that U (α) is
locally p-coCartesian if and only if s(α) is locally p-Cartesian.
If c(n − 1) = D, then we have a commutative diagram

([n − 1], c0 )
r8 LLL
β rrr LLLα
rrr LLL
rrr L&
([n], c)
id / ([n], c)

in ∆S . In view of Proposition T.[Link], it will suffice to show that U (id) and U (β) are p-Cartesian. For
U (id) this is obvious; for U (β) we combine Lemma 3.3.7 with assumption (3).
Suppose instead that c(n − 1) = C. Let c0 = c|{n − 1, n} and let c00 = c|{n − 1}. We have a commutative
diagram
α0 /
({n − 1}, c00 ) ({n − 1, n}, c0 )
β β0
 
([n − 1], c0 )
α / ([n], c)

in the category ∆S . We wish to prove that U (α) is locally p-Cartesian. Using (2) and Lemma 3.3.7, we
deduce that U (β) is p-Cartesian. According to Lemma T.[Link], it will suffice to show that U (α ◦ β) is
locally p-Cartesian. Using (2), we reduce to showing that U (α ◦ β) is locally p-coCartesian. Conditions (3)
and (4) imply that U (α0 ) and U (β 0 ) are locally p-coCartesian. Invoking Lemma 3.3.7, we deduce that U (β 0 )
is p-coCartesian. Lemma T.[Link] now implies that U (β 0 ◦α0 ) ' U (α ◦β) is locally p-coCartesian, as desired.
Remark 3.3.9. In the situation of Definition 3.3.3, condition (3) is equivalent to the following apparently
weaker condition:
(30 ) For every object ([n], c) ∈ ∆S , let c0 = c|{n} and let α : ({n}, c0 ) → ([n], c) denote the inclusion. Then
U (α) is locally p-coCartesian.
It is clear that (3) ⇒ (30 ). Conversely, suppose that (30 ) is satisfied, and consider an arbitrary morphism
β : ([m], c) → ([n], c0 ) such that β(m) = n. Let c0 = c|{m}, and form a commutative diagram

([m], c)
α sss
ss9 JJ
JJ β
JJ
ss JJ
sss J%
({m}, c0 ) / ([n], c0 ).

We wish to show that U (β) is locally p-coCartesian. In view of Lemma 3.3.7, it will suffice to show U (β) is
p-Cartesian. Using Lemma T.[Link], it will suffice to show that U (α) and U (β ◦ α) are p-Cartesian. This
follows immediately from (30 ) and Lemma 3.3.7.
⊗
Remark 3.3.10. Suppose we are in the situation of Definition 3.3.3, and that U : N(∆S )op → End (C, D)
satisfies conditions (1), (2), and (40 ) (see Remark 3.3.8). Then (3) is equivalent to the following apparently
weaker condition:

104
(300 ) Let ([n], c) be an object of ∆S such that c(n) = D, let c0 = c|{n} and let α : ({n}, c0 ) → ([n], c) be
the inclusion. Then U (α) is locally p-coCartesian.
It is clear that (3) ⇒ (300 ). To prove the converse, it will suffice to show that (300 ) implies condition (30 ) of
Remark 3.3.9. Suppose given an object ([n], c) ∈ ∆S , let c0 = c|{n} and let α : ({n}, c0 ) → ([n], c) be the
inclusion. We wish to prove that U (α) is locally p-coCartesian. If c(n) = C this follows immediately from
(300 ). Otherwise, define c0 : [n + 1] → {C, D} by the formula
(
0 c(i) if 0 ≤ i ≤ n
c (i) =
D if i = n + 1,

and let c00 = c0 |{n, n + 1}. We have a commutative diagram

({n}, c0 )
α / ([n], c)

β0 β
 
α0 / ([n + 1], c0 )
({n, n + 1}, c00 )

in the category ∆S . Condition (40 ) implies that U (β) is p-coCartesian. Lemma T.[Link] now asserts that
U (α) is locally p-coCartesian if and only if U (β ◦ α) is locally p-coCartesian. For this, it will suffice to
show that U (α0 ) and U (β 0 ) are p-coCartesian. For U (β 0 ), this follows from (40 ). To show that U (α0 ) is
p-coCartesian, we first invoke Lemma 3.3.7 to reduce to the problem of showing that U (α0 ) is locally p-
coCartesian, then (2) to reduce to the problem of showing that U (α) is locally p-Cartesian. We now set
c01 = c0 |{n + 1} and consider the commutative diagram

({n, n + 1}, c00 )

γ mmm
mm6 PPP
PPPα0
mmm PPP
mm PPP
mm (
0
({n + 1}, c1 )) / ([n + 1], c0 ).

In view of Lemma T.[Link], it will suffice to show that U (γ) and U (α0 ◦ γ) are p-Cartesian. This follows
immediately from (300 ) and Lemma 3.3.7.
⊗
Proof of Proposition 3.3.5. Lemma T.[Link] implies that the projection End (C, D) → End⊗ (C, D) is a
categorical fibration. It follows easily that θ is a categorical fibration. It will therefore suffice to show that
θ is an equivalence of ∞-categories.
Theorem 3.2.10 implies that ADat(C, D) is a Kan complex. Since the projection θ0 : ADat(C, D) →
ADat(C, D) is a categorical fibration, Proposition T.[Link] implies that θ0 is a coCartesian fibration. Invoking
Corollary T.[Link], we deduce that θ is a categorical equivalence if and only if, for every adjunction datum
U ∈ ADat(C, D), the induced map ADatU (C, D) → D is a categorical equivalence. In other words, we are
reduced to proving a slightly weaker version of Corollary 3.3.6.
Form a pullback diagram
N / End⊗ (C, D)

p
 
N(∆S )op
U / End⊗ (C, D).

Using Lemma 3.3.7, we deduce:

(a1) The map p is a locally coCartesian fibration.

105
(a2) Suppose that α : ([m], c) → ([n], c0 ) is a morphism in ∆S which induces a convex morphism [m] → [n],
and let f be a morphism in N which projects to α. Then f is p-coCartesian if and only if f is locally
p-coCartesian.
Let I be the subcategory of [1] × ∆op
S defined as follows:

• An object (i, [n], c) of [1] × ∆op

S belongs to I if and only if either i = 1 or c(n) = D.

• Given a pair of objects (i, [m], c), (j, [n], c0 ) ∈ I, a morphism (i, [m], c) → (j, [n], c0 ) in [1] × ∆op
S belongs
to I if and only if either j = 1, or the induced map α : [n] → [m] satisfies α(n) = m.
Let I0 , I1 ⊆ I be the preimages of the objects 0, 1 ∈ [1], respectively, so we have canonical identifications
I0 ⊆ ∆op
S ' I1 . Let
E = MapN(∆S )op (N(I), N)
E0 = MapN(∆S )op (N(I0 ), N) E1 = MapN(∆S )op (N(I1 ), N).
Then we have restriction maps E0 ← E → E1 . Moreover, E1 can be identified with the ∞-category of sections
of p. Consequently, Remarks 3.3.8 and 3.3.10 allow us to identify ADatU (C, D) with the full subcategory of
E1 spanned by those sections u : N(∆S )op → N which have the following properties:
(b1) Let ([n], c) ∈ ∆S be such that c(n) = D, let c0 = c|{n}, and let α : ({n}, c0 ) → ([n], c) be the inclusion.
Then u(α) is a locally p-coCartesian morphism of N.
(b2) Let n > 0, let c : [n] → {C, D} satisfy c(n) = D, let c0 = c|{0, . . . , n − 1}, and let α : ([n − 1], c0 ) →
([n], c) be the inclusion. Then u(α) is a p-coCartesian morphism of N.
Note that, for every object (1, [n], c) ∈ I1 , the ∞-category N(I0 )/(1,[n],c) has a final object, given by
(0, [n + 1], c0 ), where c0 is defined by the formula
(
c(i) if i ≤ n
c0 (i) =
D if i = n + 1.

Likewise, for every object (0, [n], c) ∈ I0 , the ∞-category N(I1 )(0,[n],c)/ has an initial object, given by (1, [n], c).
Combining this observation, (a1), (a2), and Proposition T.[Link], we deduce the following:

(c1) Every functor f0 ∈ E0 admits a p-left Kan extension f : N(I) → N. Moreover, an arbitrary extension
f 0 : N(I) → N of f0 is a p-left Kan extension of f0 if and only if it is compatible with the projection
to N(∆S )op and carries each morphism α : (0, [n + 1], c0 ) → (1, [n], c) to a locally p-coCartesian edge
of N, where c0 is defined as above. Moreover, the projection E0 → E0 is a trivial Kan fibration, where
E0 ⊆ E denotes the full subcategory spanned by those functors f which are p-left Kan extensions of
f | N(I0 ).
(c2) Every functor f1 ∈ E1 admits a p-right Kan extension f : N(I) → N. Moreover, an arbitrary extension
f 0 : N(I) → N of f1 is a p-right Kan extension of f1 if and only if it is compatible with the projection
to N(∆S )op and carries each morphism β : (0, [n], c) → (1, [n], c) to an equivalence in N. Moreover,
the projection E1 → E1 is a trivial Kan fibration, where E1 ⊆ E denotes the full subcategory spanned
by those functors f which are p-right Kan extensions of f | N(I1 ).

Applying Proposition T.[Link], we obtain a pair of adjoint functors E0 o

F / E , where:
1
G

ψ
• The functor F is given by a composition E0 → E0 ⊆ E → E1 where ψ is a section of the trivial Kan
fibration E0 → E0 (a left Kan extension functor) and the last map is given by restriction.

106
 ψ0
• The functor G is given by a composition E1 → E1 ⊆ E → E0 where ψ 0 is a section of the trivial Kan
fibration E1 → E1 (a right Kan extension functor) and the last map is given by restriction. If we
choose ψ 0 to be given by composition with the retraction of I onto I1 , then we can identify G with the
restriction map E1 → E0 .

Unwinding the definitions, and using the fact that U ([D, D]) is equivalent to the identity map from D to
itself (see Remark 3.2.8), we deduce that the unit idE0 → G ◦ F is an equivalence. Consequently, the functor
F is fully faithful. The essential image of F can be identified with the full subcategory E01 ⊆ E1 spanned by
those sections of p which satisfy condition (b2).
Form a pullback diagram
N0 /N

p0 p
  
N(I0 )  / N(∆S )op .

If u ∈ E0 , then F (u) ∈ E01 satisfies (b1) if and only if u satisfies the obvious analogue of (b1). In view of
Lemma 3.3.7, this can be reformulated as follows:

(b10 ) Let ([n], c) ∈ I0 , let c0 = c|{n}, and let α : ({n}, c0 ) → ([n], c) be the inclusion. Then u(α) is a
p0 -Cartesian morphism of N0 .

Let E000 ⊆ E0 be the full subcategory spanned by those functors u which satisfy condition (b10 ). Let
I00⊆ I0 be the full subcategory spanned by the object [D] ∈ ∆S . We observe that condition (b10 ) is simply
a reformulation of the requirement that u be a p-right Kan extension of u| N(I00 )op . Moreover, Lemma
3.3.7 implies that every partial section u0 : N(I00 )op → N0 admits a p-right Kan extension. It follows from
Proposition T.[Link] that the restriction map E000 → MapN(I0 )op (N(I00 )op , N0 ) ' D is a trivial Kan fibration,
hence a categorical equivalence as desired.

3.4 Split Simplicial Objects

Notation 3.4.1. The category ∆−∞ is defined as follows: the objects of ∆−∞ are integers n ≥ −1, and
Hom∆−∞ ([m], [n]) is the set of monotone maps [m] ∪ {−∞} → [n] ∪ {−∞} which preserve the base point
−∞ (which is regarded as a least element of both [m] ∪ {−∞} and [n] ∪ {−∞}). We have inclusions of
subcategories ∆ ⊆ ∆+ ⊆ ∆−∞ , where the latter identifies ∆+ with the subcategory of ∆−∞ having the
same objects, and a map f : [m] ∪ {−∞} → [n] ∪ {−∞} belongs to ∆+ if and only if f −1 (−∞) = {−∞}.
Definition 3.4.2. Let C be an ∞-category. We will say that an augmented simplicial object U : N(∆+ )op →
C is split if U extends to a functor N(∆−∞ )op → C. We will say that a simplicial object U : N(∆)op → C
is split if it extends to a split augmented simplicial object. Given a functor G : D → C, we will say that an
(augmented) simplicial object U of D is G-split if G ◦ U is split, when regarded as a simplicial object of C.
Remark 3.4.3. Using the terminology of Definition 3.4.2, we can formulate Lemma T.[Link] as follows:
every split augmented simplicial object is a colimit diagram. (In [40], we used a slightly different notation,
employing a category ∆∞ in place of the category ∆−∞ defined above. However, these two categories are
equivalent to one another, via the functor which reverses order.)

Remark 3.4.4. Let F : C → D be a functor between ∞-categories, and let V be a split simplicial object of
C. Then F ◦ V is a split simplicial object of D. Lemma T.[Link] implies that V admits a colimit in C, and
that F preserves that colimit.
Example 3.4.5. Let C be an abelian category. Then Dold-Kan correspondence establishes an equivalence
of categories between the category Fun(∆op , C) of simplicial objects of C and the category Ch≥0 (C) of

107
nonnegatively graded chain complexes in C (see, for example, [75]). Let X• be a simplicial object of C, and
let
d d
· · · → C2 →2 C1 →1 C0
be the corresponding (normalized) chain complex. Then extending X to an augmented simplicial object
∆op
+ → C is equivalent to providing an augmentation on the chain complex C• : that is, providing a map
d0 : C0 → C−1 such that d0 ◦ d1 = 0. Extending X• to a split augmented simplicial object ∆op −∞ → C is
equivalent to providing a contracting homotopy {hn : Cn−1 → Cn }n≥0 for the augmented chain complex C• :
that is, a family of maps satisfying d0 h0 = idC−1 and dn+1 hn+1 + hn dn = idCn for n ≥ 0. The existence
of such a contracting homotopy is equivalent to the requirement that the chain complex C• be a split exact
resolution of C−1 .
The construction [n] 7→ [0] ? [n] ' [n + 1] determines a functor ρ from ∆+ to itself. There is an evident
natural transformation α : id∆+ → ρ. Let j denote the inclusion ∆+ ,→ ∆−∞ . There is another natural
transformation of functors β : jρ → j, which carries an object [n] ∈ ∆+ to the map f : [n + 1] ∪ {−∞} →
[n] ∪ {−∞} such that f (0) = ∞ and f (k) = k − 1 for 0 < k ≤ n + 1. The composition of j(α) and β is the
identity natural transformation from the functor j to itself. We can view j(α) and β as determining a map
of simplicial sets N(∆+ ) × ∆2 → N(∆∞ ).
Proposition 3.4.6. The diagram

N(∆+ ) × Λ20 / N(∆+ )

 
N(∆+ ) × ∆2 / N(∆−∞ )

is a homotopy pushout square of simplicial sets (with respect to the Joyal model structure).
Proof. Let f : [m] ∪ {−∞} → [n] ∪ {−∞} be a morphism in ∆−∞ . We define the complexity of c(f ) to be
the cardinality of the set {i ∈ [m] : f (i) = ∞}. If σ is an k-simplex of N(∆−∞ ) corresponding to a sequence
of maps
f1 fk
[n0 ] ∪ {−∞} → · · · → [nk ] ∪ {−∞},
then we define the complexity c(σ) to be the sum c(f1 ) + · · · + c(fk ). For every nonnegative integer m, let
X(m) denote the simplicial subset of N(∆−∞ ) spanned by those S simplices σ such that c(σ) ≤ m. We note
that X(0) can be identified with N(∆+ ), and that N(∆−∞ ) = m X(m).
Proposition 3.4.6 follows immediately from the following pair of assertions:

(a) The diagram

N(∆+ ) × Λ20 / N(∆+ )

 
N(∆+ ) × ∆2 / X(1)

is a homotopy pushout square of simplicial sets (with respect to the Joyal model structure).
(b) For n ≥ 2, the inclusion X(n − 1) ,→ X(n) is a categorical equivalence.
The proofs of both (a) and (b) will require some auxiliary constructions. Let F1 , . . . , Fk be a sequence of
functors from ∆+ to itself (in practice, each of these functors will be either ρ or the identity). We define a
category C[F1 , . . . , Fk ] as follows:

(i) An object of C[F1 , . . . , Fk ] is a pair ([n], i), where [n] ∈ ∆+ and 0 ≤ i ≤ k.

108
 (ii) If i ≤ j, a morphism from ([m], i) → ([n], j) in C[F1 , . . . , Fk ] is a map [m] → Fi+1 · · · Fj [n] in ∆+ . If
i > j, there are no morphisms from ([m], i) to ([n], j) in C[F1 , . . . , Fk ].
We now prove (a). The natural transformation β : jρ → j determines a map N(C[ρ]) → N(∆∞ ), which
factors through the simplicial subset X(1). This map fits into a commutative diagram

N(∆+ ) × Λ20 / N(C[id]) / N(∆+ )

  
N(∆+ ) × ∆2 / N(C[ρ]) / X(1)

where the right hand square is a pushout square and therefore (since the vertical maps are all cofibrations)
a homotopy pushout square. It will therefore suffice to show that the left square is a homotopy pushout
diagram. This square fits into a larger diagram

N(∆+ ) × ∆{0,1}
θ / N(∆+ )

 
N(∆+ ) × Λ20 / N(C[id])

 
N(∆+ ) × ∆2 / N(C[ρ])

where the upper square is a pushout diagram and therefore (since the vertical maps are again cofibrations)
a homotopy pushout square; here the map θ classifies the natural transformation α : id → ρ. It therefore
suffices to show that the outer square is a homotopy pushout. This square fits into a larger diagram

N(∆+ ) × {1} / N(∆+ ) × ∆{0,1} θ / N(∆+ )

  
N(∆+ ) × ∆{1,2} / N(∆+ ) × ∆2 / N(C[ρ]).

The left square is obviously a homotopy pushout diagram, and the outer rectangle is a homotopy pushout
diagram by Proposition T.[Link]. It follows that the right square is a homotopy pushout diagram, as desired.
We now prove (b). Fix n ≥ 2. For every subset S ⊆ {1, . . . , n}, let CS = C[F1 , . . . , Fn ] where
(
ρ if i ∈ S
Fi =
id∆+ if i ∈ / S.

The natural transformation α : id∆+ → ρ determines inclusion functors CS → CS 0 for S ⊆ S 0 . Let

Y = N(C{1,...,n} ) and let YS = N(CS ) for S ⊆ {1, . . . , n}, so that we can identify each YS with a simplicial
subset of Y . Let P be the collectionS of all subsets S ⊆ {1, . . . , n} and P0 ⊂ P the collection of all proper
subsets S ⊂ {1, . . . , n}. Let Y 0 = S∈P0 YS ⊆ Y . The diagram S 7→ YS is projectively cofibrant, so we
can identify Y 0 with a homotopy colimit of the diagram {YS }S∈P0 . There is an evident pushout diagram of
simplicial sets
Y0 /Y

 
X(n − 1) / X(n).

109
Consequently, to prove that the inclusion X(n − 1) ,→ X(n) is a categorical equivalence, it will suffice to
show that the inclusion Y 0 ,→ Y is a categorical equivalence. In other words, we are reduced to proving that
the diagram {YS }S∈P is a homotopy colimit (with respect to the Joyal model structure).
There is an obvious forgetful functor Y → ∆n . For every simplicial subset K ⊆ ∆n , let ZK : P → Set∆
be the functor given by the formula ` ZK (S) = ` YS ×∆` n K. We wish to prove that Z∆n is a homotopy
colimit diagram. Let K0 = ∆{0,1} {1} ∆{1,2} {2} · · · {n−1} ∆{n−1,n} , so that the inclusion K0 ,→ ∆n is
a categorical equivalence. Since each of the projection maps YS → ∆n is a Cartesian fibration, we deduce
that the inclusion ZK0 → Z∆n is a weak equivalence of diagrams (Proposition T.[Link]). It will therefore
suffice to show that ZK0 is a homotopy colimit diagram. Since the collection of homotopy colimit diagrams
in Set∆ is stable under the formation of homotopy pushout squares, we can reduce further to proving that
ZK is a homotopy colimit diagram when K = {i} ⊆ ∆n for 0 ≤ i ≤ n or K = ∆{i−1,i} ⊆ ∆n for 0 < i ≤ n.
In the first case, this is obvious: the diagram Z{i} is isomorphic to the constant diagram taking the value
N(∆+ ), which is a homotopy colimit diagram because the simplicial set N(P0 ) is weakly contractible (since
P0 has a least element ∅ ⊂ {1, . . . , n}). To handle the second, let us suppose that K = ∆{i−1,i} , so that ZK
is given by the formula (
N(C[ρ]) if i ∈ S
ZK (S) =
N(C[id]) otherwise.
Let P1 be the subset of P0 consisting of the subsets ∅, {i} ⊂ {1, . . . , n}. We observe that ZK |P0 is a homotopy
left Kan extension of ZK |P1 . Consequently, to prove that ZK is a homotopy colimit diagram, it suffices to
show that ZK exhibits ZK ({1, . . . , n}) as a homotopy colimit of the restriction ZK |P1 . Since P1 has a
final object {i}, this follows from the observation that the map ZK ({i}) → ZK ({1, . . . , n}) is a categorical
equivalence (in fact, an isomorphism) of simplicial sets.
For any ∞-category C, composition with the functor ρ : ∆+ → ∆+ determines a functor from the
∞-category of augmented simplicial objects of C to itself. We will denote this functor by T . The natural
transformation id → ρ determines a natural map T X• → X• , for every augmented simplicial object X• of
C. Proposition 3.4.6 immediately implies the following:
Corollary 3.4.7. Let C be an ∞-category, and let X• be an augmented simplicial object of C. The following
conditions are equivalent:
(1) The augmented simplicial object X• is split.
(2) The canonical map T X• → X• admits a right homotopy inverse.
Corollary 3.4.8. Let C be an ∞-category, and let X denote the full subcategory of Fun(N(∆op + ), C) be the
full subcategory spanned by the split augmented simplicial objects of C (that is, X is the essential image of the
restriction functor Fun(N(∆op op
−∞ ), C) → Fun(N(∆+ ), C)). Then X is stable under retracts in Fun(N(∆+ ), C).
op

Proof. Suppose we have a commutative diagram

> X• BB
||| BB β
||
α BB
| BB
||
Y•
id / Y•

of augmented simplicial objects of C, where X• is split. We wish to prove that Y• is split. According to
Corollary 3.4.7, there exists a map of augmented simplicial objects γ : X• → T X• such that the composition
γ
X• → T X• → X• is homotopic to the identity. Let γ 0 denote the composition
α γ Tβ
Y• → X• → T X• → T Y• .
γ0
A simple diagram chase shows that the composition Y• → T Y• → Y• is homotopic to the identity, so that
Y• splits as required.

110
Remark 3.4.9. The functor ρ carries ∆+ into the subcategory ∆ ⊆ ∆+ . Consequently, the translation
functor T factors through a functor Fun(N(∆op ), C) → Fun(N(∆op
+ ), C). We will abuse notation and denote
this functor also by T .
Corollary 3.4.10. Let C be an ∞-category, and let X• be an augmented simplicial object of C. The following
conditions are equivalent:
(1) The augmented simplicial object X• is split.

(2) There exists a simplicial object U• of C such that X• is a retract of T U• (in the ∞-category of augmented
simplicial objects of C).
Proof. The implication (1) ⇒ (2) follows from Corollary 3.4.7. We will prove that (2) ⇒ (1). Using Corollary
3.4.8, we may assume that X• = T U• . The desired result in this case follows from the observation that the
shift functor ρ : ∆+ → ∆ extends naturally to a functor ∆−∞ → ∆, determined by the natural identification
of linearly ordered sets [n + 1] = [0] ? [n] ' [n] ∪ {−∞}.

3.5 The Barr-Beck Theorem

Let F : C → D be a functor between ∞-categories which admits a right adjoint G. In §3.2, we saw that
F determines an adjunction datum U ∈ ADat(C, D), which restricts to give a monad T ∈ Alg(Fun(C, C)).
G0
Moreover, the functor G factors canonically (up to homotopy) as a composition D → ModT (C) → C . Our
goal in this section is to establish a criterion which can be used to test whether G0 is an equivalence of
∞-categories.
Theorem 3.5.1 (∞-Categorical Barr-Beck Theorem). Let C and D be ∞-categories, and suppose given an
adjunction datum U ∈ ADat(C, D). Let G : D → C be the induced functor, T the induced monad on C, and
let ψ : ADatU (C, D) → ModT (C) be the forgetful functor (here ADatU (C, D) is canonically equivalent to D,
in view of Corollary 3.3.6). Then ψ is an equivalence of categories if and only if the following conditions are
satisfied:
(1) The functor G : D → C is conservative; that is, a morphism f : D → D0 in D is an equivalence if and
only if G(f ) is an equivalence in C.
(2) Let V be a simplicial object of D which is G-split. Then V admits a colimit in D, and that colimit is
preserved by G.
We will give the proof at the end of this section.
Remark 3.5.2. Hypotheses (1) and (2) of Theorem 3.5.1 can be rephrased as the following single condition:
(∗) Let V : N(∆)op → D be a simplicial object of D which is G-split. Then V has a colimit in D. Moreover,
an arbitrary extension V : N(∆)op → D is a colimit diagram if and only if G ◦ V is a colimit diagram.
It is clear that (1) and (2) imply (∗), and that (∗) implies (2). To prove that (∗) implies (1), let us consider
an arbitrary morphism f : D0 → D in D. The map f determines an augmented simplicial object V of D,
with (
D0 if n ≥ 0
V ([n]) =
D if n = −1.

The underlying simplicial object V = V | N(∆)op is constant, and therefore G-split. Since N(∆) is con-
tractible, V is a colimit diagram if and only if f is an equivalence, and G ◦ V is a colimit diagram if and only
if G(f ) is an equivalence. If (∗) is satisfied, then these conditions are equivalent, so that G is conservative
as desired.

111
Remark 3.5.3. Let C be an ∞-category, and regard C as left-tensored over End(C). We have a commutative
diagram
Mod(C)
θ / C × Alg(End(C))
NNN
NNNp mmm
NNN mmmmm
NN' mm 0
vmmm p
Alg(End(C)),

where θ is determined by the p and the forgetful functor Mod(C) → C. The functor p0 is obviously a Cartesian
fibration, and Corollary 2.3.3 implies that p is a Cartesian fibration as well. Proposition 2.3.2 implies that
θ carries p-Cartesian edges to p0 -Cartesian edges. Consequently, θ classifies a natural transformation of
functors F → F 0 , where F : Alg(End(C))op → Cat∞ is the functor classified by p (so that F (A) ' ModT (C)
for every monad T ∈ Alg(End(C))) and F 0 is the constant functor taking the value C ∈ Cat∞ . We may
identify this transformation with a functor α : Alg(End(C))op → Cat/∞C . We can interpret Theorem 3.5.1 as
describing the essential image of the functor α: namely, a functor G : D → C belongs to the essential image
of α if and only if G admits a left adjoint and satisfies conditions (1) and (2) of Theorem 3.5.1. In a future
paper, we will show that the functor α is fully faithful. In other words, we may identify monads on C with
(certain) ∞-categories lying over C.

In practice, it is often easier to apply the following slightly weaker version of the Barr-Beck theorem:
Corollary 3.5.4. Let G : D → C and ψ : D → ModT (C) be as in Theorem 3.5.1. Suppose that:
(1) The functor G is conservative.

(2) The ∞-category D admits geometric realizations of simplicial objects.

(3) The functor G preserves geometric realizations of simplicial objects.
Then the functor ψ is an equivalence.
The Barr-Beck theorem is an extremely useful tool. However, for many applications (such as our con-
struction of coproducts of algebras in §1.5) it will suffice to know the following less precise statement (which
makes no mention of the theory of monads):

Proposition 3.5.5. Let C o

F / D be a pair of adjoint functors between ∞-categories which satisfies condi-
G
tions (1) and (2) of Theorem 3.5.1. For every object D ∈ D, there exists a simplicial object D• : N(∆)op → D
having colimit D, such that each Dn lies in the essential image of F .
Proof. Using Theorem 3.2.10 and Proposition 3.3.5, we may assume that the functors F and G are determined
by an adjunction datum U ∈ ADat(C, D), and that D ∈ D is the image of [D] ∈ ∆S under a pointed
⊗
adjunction datum U : N(∆S )op → End (C, D) which lies over U .
We define a functor φ : ∆+ → ∆S by the formula

[n] 7→ ([1] ? [n], c),

where c is given by the equation 

C if i = 0

c(i) = D if i = 1
C if 1 < i ≤ n + 2.



Composing with s and evaluating at 1 ∈ [n + 2] ' [1] ? [n], we obtain an augmented simplicial object
D• : N(∆+ )op → D. It follows immediately from Definition 3.3.3 that D−1 ' D. To complete the proof, it

112
will suffice to show that D• is a colimit diagram. In view of Remark 3.5.2, it will suffice to show that D• is
G-split.
Let C• : N(∆+ )op → C be the functor obtained from ψ by composing with U and evaluating at 0 ∈ [n+2].
Then C• is the image of D• under a functor G• : N(∆+ )op → Fun(D, C) determined by the augmentation
datum s. Invoking the definition of an augmentation datum, we conclude that G• is equivalent to the
constant augmented simplicial object taking the value G ∈ Fun(D, C). It will therefore suffice to show that
the augmented simplicial object C• is split.
Let φ0 : ∆+ → ∆S be defined by the formula [n] 7→ ([0] ? [n], c0 ), where c0 takes the constant value
C. Let C•0 be the augmented simplicial object of C obtained by composing φ0 with s and evaluating at
0 ∈ [n + 1] ' [0] ? [n]. There is an evident natural transformation of functors φ0 → φ, which induces a map
of simplicial objects C• → C•0 . Invoking the definition of a pointed adjunction datum, we deduce that this
transformation is an equivalence. It will therefore suffice to prove that C•0 is split. For this, it suffices to
show that φ0 can be extended to a functor ∆−∞ → ∆S , which is evident.
Remark 3.5.6. The proof of Proposition 3.5.5 actually establishes a stronger result: namely, the simplicial
resolution of an object D ∈ D can be chosen functorially in D. Informally, we can describe this functorial
choice as follows. Let V = F ◦ G. A dual to Theorem 3.2.10 implies that V underlies a comonad on the
∞-category D. Consequently, given D ∈ D we can construct an augmented simplicial object E• , with
Dn ' V n+1 D. This augmented simplicial object is G-split, and therefore a colimit diagram. Consequently,
D ' D−1 can be obtained as the colimit of the underlying simplicial object D• , and for n ≥ 0 we have
Dn ' F ((G ◦ V n )D), so that Dn belongs to the essential image of F .
We conclude with few applications of Proposition 3.5.5:
Corollary 3.5.7. Suppose given a commutative diagram of ∞-categories

C@
U / C0
@@ }}
@@G }
@@ }}
@ ~}}} G0
D.

Suppose furthermore that:

(a) The ∞-categories C and C0 admit geometric realizations for simplicial objects.
(b) The functors G and G0 admit left adjoints, which we will denote by F and F 0 , respectively.
(c) The functor G0 is conservative and preserves geometric realizations of simplicial objects.
Then:
(1) The functor U admits a left adjoint T .
(2) The functor U is an equivalence if and only if the following additional conditions are satisfied:
(d) The functor G is conservative and preserves geometric realizations of simplicial objects.

(e) The unit of the adjunction C0 o

T / C induces an equivalence of functors
U

G0 ◦ F 0 → G0 ◦ U ◦ T ◦ F 0 ' G ◦ F.

Proof. We first prove (1). Let j 0 : C0 → Fun(C0 , S)op denote the Yoneda embedding, and consider the
composition
0 j0 ◦U
j : C0 → Fun(C0 , S)op → Fun(C, S)op .

113
 0
Let C00 ⊆ C be the full subcategory spanned by those objects C 0 ∈ C0 for which j (C 0 ) belongs to the essential
image of the Yoneda embedding j : C → Fun(C, S)op . We wish to prove that C00 = C0 .
Proposition T.[Link] and assumption (a) guarantee that the essential image of j is stable under the
formation of geometric realizations of simplicial objects. Moreover, Proposition T.[Link] also implies that
0
j preserves geometric realizations of simplicial objects. It follows that C00 is stable under the formation of
geometric realizations in C0 . In view of Proposition 3.5.5, it will suffice to show that C00 contains the essential
0
image of F 0 . This is clear, since j (F 0 (D)) ' j(F (D)). This completes the proof of (1).
We now prove (2). If U is an equivalence, then (e) is obvious (since the unit itself is an equivalence) and
(d) follows from (c). To prove the converse, choose compatible unit and counit transformations

u : id → U ◦ T v : T ◦ U → id .

We wish to prove that u and v are equivalences. Since G is conservative, the functor U is also conservative.
Consequently, the transformation v is an equivalence if and only if U (v) : U ◦ T ◦ U → U is an equivalence.
Since U (v) has a section determined by u, it will suffice to prove that u is an equivalence.
Let C01 be the full subcategory of C0 spanned by those objects C for which the map u(C) : C → (U ◦T )(C)
is an equivalence. Since G = G0 ◦ U and G0 both preserve geometric realizations and G0 is conservative, we
conclude that U preserves geometric realizations. The functor T is a left adjoint, and therefore preserves
geometric realizations. It follows that U ◦ T preserves geometric realizations, so that C01 is stable under
geometric realizations in C0 . In view of Propsition 3.5.5, it will suffice to show that C01 contains the essential
image of F 0 . In other words, we must show that u induces an equivalence F 0 → U ◦ T ◦ F 0 . Using our
assumption that G0 is conservative, we reduce immediately to (e).
Our next result makes use of the notation and terminology of §T.5.5.8.

Corollary 3.5.8. Suppose given a pair of adjoint functors between ∞-categories C o

F / D . Assume that:
G

(i) The ∞-category D admits filtered colimits and geometric realizations, and G preserves filtered colimits
and geometric realizations.
(ii) The ∞-category C is projectively generated (Definition T.[Link]).

(iii) The functor G is conservative.

Then:
(1) The ∞-category D is projectively generated.

(2) An object D ∈ D is compact and projective if and only if there exists a compact projective object C ∈ C
such that D is a retract of F (C).
(3) The functor G preserves all sifted colimits.
Proof. Let C0 denote the full subcategory of C spanned by the compact projective objects. Let D0 denote
the essential image of F | C0 . Using assumption (i), we deduce that D0 consists of compact projective objects
of D. Without loss of generality, we may assume that the ∞-category D is minimal, so that D0 is small.
Moreover, since C0 is stable under finite coproducts in C, and F preserves finite coproducts, we conclude
that D0 admits finite coproducts (which are also finite coproducts in D).
Let D0 = PΣ (D0 ) (see Definition T.[Link]). Using Proposition T.[Link], we deduce that the inclusion
D ⊆ D is homotopic to a composition
0
j f
D0 → D0 → D,
where the functor f preserves filtered colimits and geometric realizations. Combining Proposition T.[Link]
with Proposition 3.5.5 and assumption (ii), we conclude that f is an equivalence of ∞-categories. This

114
proves (1). Moreover, the proof shows that D0 is spanned by a set of compact projective generators for D
(Definition T.[Link]), so that assertion (2) follows from Proposition T.[Link]. Assertion (3) now follows
from Proposition T.[Link].
We now return to the proof of Theorem 3.5.1. One direction is essentially contained in the following
Lemma:
Lemma 3.5.9. Let C be a monoidal ∞-category, M an ∞-category which is left-tensored over C, A an
algebra object, and θ : ModA (M) → M the forgetful functor. Then:
(1) Every θ-split simplicial object of ModA (M) admits a colimit in M.
(2) The functor θ preserves colimits of θ-split simplicial objects.
Proof. Note that a simplicial object of ModA (M) can be viewed as a bisimplicial object of M⊗ . In order to
avoid confusion, we let K denote the simplicial set N(∆)op when we wish to emphasize the role of N(∆)op as
indexing simplicial objects, while we use the notation N(∆)op to denote the base of the coCartesian fibration
M⊗ → N(∆)op .
Form a pullback diagram
N / M⊗

p
 
N(∆)op
A / C⊗ ,

so that p is a locally coCartesian fibration (Lemma 2.1.13). Let u : K → ModA (M) be a θ-split simplicial
object, corresponding to a map V : K × N(∆)op → N. We observe that every fiber of p can be identified with
M, and each of the induced maps V[n] : K × {[n]} → N[n] can be identified with the composition of V with
the forgetful functor θ. It follows that each of the simplicial objects V[n] splits. Using Lemma T.[Link], we
deduce that each V[n] admits a colimit V [n] : K . → N[n] , and that these colimits are preserved by each of the
associated functors N[n] → N[m] (Remark 3.4.4). Applying Proposition T.[Link], we conclude that each V [n]
is a p-colimit diagram. We now invoke Lemma 2.3.1 to deduce the existence of a map V : K . × N(∆)op → N
which induces a p-colimit diagram on each fiber K . × {[n]}. Moreover, Lemma 2.3.1 implies that V defines
a colimit diagram v : K . → MapN(∆)op (N(∆)op , N). Since v = v|K factors through the full subcategory
ModA (M) ⊆ MapN(∆)op (N(∆)op , N), it is easy to see that v factors through ModA (M). It follows that v is
a colimit of v in ModA (M). This proves (1), and (2) follows from the observation that θ ◦ v = V |K . × {[0]}
is a colimit diagram in M by construction.
Proof of Theorem 3.5.1. The “only if” direction follows from Corollary 2.3.5 and Lemma 3.5.9. For the
converse, let us suppose that (1) and (2) are satisfied. Set S = {C, D}, and form a pullback diagram

N / End⊗ (C, D)

p
 
N(∆S )op
s / End⊗ (C, D).

We define full subcategories ∆C ⊆ ∆0S ⊆ ∆S as follows:

• An object ([n], c) ∈ ∆S belongs to ∆0S if and only if c(n) = C.

• An object ([n], c) ∈ ∆S belongs to ∆C if and only if c is a constant map taking the value C.

115
Define ∞-categories
E = MapN(∆S )op (N(∆S )op , N),

E0 = MapN(∆S )op (N(∆0S )op , N), E00 = MapN(∆S )op (N(∆C )op , N),
so that we have restriction maps
E → E0 → E00 .
We observe that ModT (C) can be identified with a full subcategory of E00 , and ADatU (C, D) can be identified
with a full subcategory of E. Let E00 ⊆ E0 be the full subcategory spanned by those functors f : N(∆0C,D )op →
N such that f is p-right Kan extension of f0 = f | N(∆C )op , and f0 ∈ ModT (C). Let E0 ⊆ E be the
full subcategory spanned by those functors g : N(∆S )op → N such that g is a p-left Kan extension of
g0 = g| N(∆0S )op , and g0 ∈ E00 . We will prove:

(i) Every f0 : N(∆C )op → N which belongs to ModT (C) admits a p-right Kan extension to N(∆0S )op .
(ii) Every g0 : N(∆0S )op → N whcih belongs to E00 admits a p-right Kan extension to N(∆S )op .
It will follow from Proposition T.[Link] that the restriction maps E0 → E00 → ModT (C) are trivial Kan
fibrations. We will then complete the proof by showing:

(iii) The full subcategory E0 ⊆ E coincides with ADatU (C, D).

Observe that, for every object ([n], c) ∈ ∆0S , the ∞-category N(∆op C )([n],c)/ has an initial object, given
by a map α : ([k], c0 ) → ([n], c) which induces a bijection [k] → {i ∈ [n] : c(i) = C}. Since ([n], c) ∈ ∆0S , the
map α satisfies α(k) = n. Combining Lemmas 3.3.7 and T.[Link], we deduce the following slightly stronger
version of (i):

(i0 ) Every functor f0 ∈ E00 admits a p-right Kan extension f : N(∆0S )op → N. Moreover, an arbitrary
functor f ∈ E0 is a p-right Kan extension of f | N(∆C )op if and only if for every map α : ([k], c0 ) → ([n], c)
in ∆0S which induces a bijection [k] → {i ∈ [n] : c(i) = C}, the morphism f (α) is locally p-coCartesian.

Combining (i0 ), Proposition T.[Link], and the definition of ModT (C), we deduce:

(i00 ) The restriction map E00 → ModT (C) is a trivial Kan fibration. Moreover, an arbitrary functor f ∈ E0
belongs to E00 if and only if, for every morphism α : ([k], c0 ) → ([n], c) in ∆S such that α(k) = n, the
morphism f (α) is locally p-coCartesian.

The proof of (ii) is a bit more subtle. Fix an functor g0 : N(∆0S )op → N which belongs to E00 . We wish to
prove that g0 admits a p-left Kan extension to N(∆S )op . According to Lemma T.[Link], it will suffice to
g0
show that, for every object ([n], c) ∈ ∆S , the induced diagram (N(∆0C,D )op )/([n],c) → N(∆0S )op → N admits
a p-colimit. If ([n], c) ∈ ∆0S , then there is nothing to prove. Assume instead that c(n) = D. We define a
functor ψ : ∆ → (∆0S )([n],c)/ by the formula

ψ([k]) = ([n + k + 1], c0 )

(
0 c(i) if 0 ≤ i ≤ n
c (i) =
C if n + 1 ≤ i ≤ n + k + 1.

We claim that the induced map N(∆)op → (N(∆0S )op )/([n],c) is cofinal. According to Theorem T.[Link], it
will suffice to show that for every object α : ([n], c) → ([m], c0 ) in (∆0S )([n],c)/ , the category

∆0 = ∆ ×(∆0S )([n],c)/ ((∆0S )([n],c)/ )/([m],c0 )

116
has weakly contractible nerve. Let J = {i ∈ [m] : (α(n) ≤ i) ∧ (c0 (i) = C)}. Then J contains m, and is
therefore nonempty. Unwinding the definitions, we see that ∆0 can be identified with ∆/J , which has a final
object.
Using Proposition T.[Link], we are reduced to proving that the diagram
g0 ◦ψ
N(∆ op ) /
_ r8 N
r
r
r p
r
 r 
N(∆op ).
h / N(∆S )op

admits an extension as indicated, which is a p-colimit diagram. Let h = h| N(∆)op . We observe that h
determines a natural transformation H : N(∆)op × ∆1 → N(∆S )op from h to the constant map taking the
value ([n], c). Using Lemma 3.3.7, we can lift H to a p-coCartesian natural transformation He : N(∆)op ×∆1 →
N from g0 ◦ ψ to a simplicial object V : N(∆) → N([n],c) . Using Proposition T.[Link], we are reduced to
op

proving that V can be extended to a p-colimit diagram.

We now observe that the fiber N([n],c) can be identiied with c(n) = D. We will prove that V is a G-split
simplicial object of D. In view of assumption (2), this will imply that V can be extended to a colimit diagram
V : N(∆op ). → N([n],c) ' D. We need to know a little bit more: namely, that U is a p-colimit diagram in
N. In view of Proposition T.[Link], it will suffice to show that for every morphism α : ([m], c0 ) → ([n], c) in
V α
∆S , the composition N(∆op ). → N([n],c) →! N([m],c0 ) is a colimit diagram. We observe that N([m],c0 ) can be
identified with the ∞-category c0 (m) ∈ {C, D}, and that α! is equivalent to some iterated composition of the
adjoint functors F and G. If α! is not an equivalence, then it is equivalent to a composition B ◦ G, for some
functor B : C → c0 (m). Invoking assumption (2), we deduce that G ◦ V is colimit diagram in C. Moreover,
since G ◦ V is split, the augmented simplicial object G ◦ V is likewise split. It follows that the image of G ◦ V
under any functor remains a split augmented simplicial object, and therefore a colimit diagram (Lemma
T.[Link]).
It remains to show that V is a G-split simplicial object of D. Our first step us to show that this assertion
is independent of the original object ([n], c) ∈ ∆S . To prove this, consider the map β : ([0], c0 ) → ([n], c),
where β(0) = n and c0 (0) = D. We define maps

ψ 0 : ∆ → (∆0S )([0],c0 )/ ,

H 0 : N(∆)op × ∆1 → N(∆S )op , e 0 : N(∆)op × ∆1 → N

H
as above, so that H e 0 is a p-coCartesian transformation from g0 ◦ ψ 0 to a simplicial object V 0 : N(∆) →
N([0],c0 ) ' D. We will show that V and V 0 are equivalent (as simplicial objects of D), so that V is G-split if
and only if V 0 is G-split.
Let us be a bit more precise. Let ([n], c) denote the constant functor N(∆)op → N(∆S )op taking
the value ([n], c), and define ([0], c0 ) likewise. The morphism β determines a natural transformation β :
([n], c) → ([0], c0 ) in the ∞-category of simplicial objects Fun(N(∆)op , N(∆S )op ). Let p : Fun(N(∆)op , N) →
Fun(N(∆)op , N(∆S )op ) be given by composition with p. Using Lemma 3.3.7, we can choose an p-coCartesian
transformation B : V → V 00 in Fun(N(∆)op , N). We will prove that V 0 and V 00 are equivalent (as simplicial
objects of N([0],c0 ) .
The functors ψ and ψ 0 determine simplicial objects N(∆)op → N(∆S )op , which we will denote by φ
and φ0 , respectively. The map β determines a natural transformation βe : φ → φ0 . We have a commutative
diagram
H /
φ ([n], c)

β
e β
 0

φ0
H / ([0], c),

117
which we can describe by a map ∆1 × ∆1 → Fun(N(∆)op , N(∆S )op ). Using Lemma 3.3.7, we deduce that p
restricts to a coCartesian fibration

Fun(N(∆)op , N) ×Fun(N(∆)op ,N(∆S )op ) (∆1 × ∆1 ) → ∆1 × ∆1 ,

so that we obtain an associated diagram of ∞-categories and functors

Fun(N(∆)op , N)φ
H!
/ Fun(N(∆)op , N([n],c) )

β
e! β!
 H!0

Fun(N(∆)op , N)φ0 / Fun(N(∆)op , N([0],c0 ) )

which commutes up to (canonical) homotopy. Since V is defined to be the image under H! of g0 ◦ ψ ∈

Fun(N(∆)op , N)φ and V 0 is defined to be the image under H!0 of g0 ◦ ψ 0 ∈ Fun(N(∆)op , N)φ0 , it will suffice to
prove that g0 ◦ψ 0 is equivalent to βe! (g0 ◦ψ). The desired equivalence is provided by the natural transformation
g0 ◦ βe : N(∆)op × ∆1 → N, which is p-coCartesian in view of Lemma 3.3.7, (i00 ), and our assumption that
g0 ∈ E00 .
We now return to the task of proving that V is G-split. Applying the above argument twice, we can
reduce to the case where n > 0 and c(n − 1) = C. Let c0 = c|[n − 1], and let γ : ([n − 1], c0 ) → ([n], c) be
the associated map in ∆S . Proceeding as above, we let ([n − 1, c0 ) denote the constant functor N(∆)op →
N(∆S )op taking the value ([n − 1], c0 ), γ : ([n], c) → ([n − 1], c0 ) the associated natural transformation in
Fun(N(∆)op , N(∆S )op ), and γ ! : Fun(N(∆)op , N([n],c) ) → Fun(N(∆)op , N([n−1],c0 ) the functor associated
to the locally coCartesian fibration p. Then γ ! can be identified with the functor Fun(N(∆)op , D) →
Fun(N(∆)op , C) given by composition with G. It will therefore suffice to show that γ ! (V ) is a split simplicial
object of N([n−1],c0 ) .
We have a commutative diagram

([n], c)
||= LLL
LLLγ
H ||
|| LLL
|| L%
|
φ
δ / ([n − 1], c0 )

described by a 2-simplex ∆2 → Fun(N(∆)op , N(∆S )op ). Applying Lemma 3.3.7, we deduce that p restricts
to a coCartesian fibration Fun(N(∆)op , N) ×Fun(N(∆)op ,N(∆S )op ) ∆2 → ∆2 , so that we have an equivalence
δ! ' γ ! ◦ H! of functors from Fun(N(∆)op , N)φ → Fun(N(∆)op , N([n−1],c0 ) ). It will therefore suffice to show
that δ! (g0 ◦ ψ) is split.
Define θ : ∆−∞ → (∆S )([n−1],c0 )/ as follows:

• The functor θ carries [k] to ([n + k], c00 ), where

(
0 c0 (i) if 0 ≤ i < n
c0 (i) =
C if n ≤ i ≤ n + k.

• Given a map α : {−∞} ∪ [k] → {−∞} ∪ [k 0 ] in ∆−∞ , the associated map θ(α) is given by the formula

i
 if i < n
θ(α)(i) = n − 1 if i ≥ n and α(i) = −∞

n + α(i − n) if i ≥ n and α(i) 6= −∞.


118
Let θ : N(∆−∞ )op → N(∆S )op be the map determined by θ, and let  be the induced natural transformation
from θ to the constant map N(∆−∞ )op → N(∆S )op taking the value ([n − 1], c0 ). Using Lemma 3.3.7, we
can choose a p-coCartesian transformation e  : (g0 ◦ θ) → W , where W is a map N(∆−∞ )op → N([n−1],c0 ) .
We now observe that W is an extension of δ! (g0 ◦ ψ), so that δ1 (g0 ◦ ψ) is split. This completes the proof of
(ii).
We now prove (iii). Suppose first that g ∈ E, and let g0 = g| N(∆0C,D )op . We wish to show that g ∈ E0
if and only if g ∈ ADatU (C, D). In view of (i00 ), either of these conditions implies that g0 ∈ E00 . We will
therefore assume that g0 ∈ E00 . By definition, g ∈ E0 if and only if g is a p-left Kan extension of g0 .
Unwinding the definitions, this condition holds if and only if g is a p-left Kan extension of g0 at ([n], c), for
every object ([n], c) ∈ ∆S . This condition is automatic if c(n) = C. Suppose instead that c(n) = D. Let
ψ : ∆ → (∆0C,D )([n],c)/ be defined as above. We observe that ψ extends to a map ψ : ∆+ → (∆S )([n],c)/ ,
which preserves initial objects. Since ψ induces a cofinal map N(∆)op → (N(∆0C,D )op , we conclude that g is
a p-left Kan extension of g0 at ([n], c) if and only if g ◦ ψ is a p-colimit diagram. Using Lemma 3.3.7, we can
construct a p-coCartesian transformation from g◦ψ to an augmented simplicial object V : N(∆op ). → N([n],c) .
In view of Proposition T.[Link], g is a p-left Kan extension of g0 at ([n], c) if and only if V is a p-colimit
diagram. The arguments given above show that this condition is satisfied if and only if V is a colimit diagram
in N([n],c) . Moreover, this condition is independent of the choice of ([n], c). In particular, we may suppose
that n = 1 and c is given by the formula
(
C if i = 0
c(i) =
D if i = 1.

Let c0 denote the restriction of c to [0] ⊆ [1]. Using Lemma 3.3.7, we can choose a p-coCartesian
transformation N(∆op ). × ∆1 → N from X to an augmented simplicial object X : N(∆op ). → N([0],c0 ) . If
we view V as an augmented simplicial object of D, then X is equivalent to the augmented simplicial object of
C obtained by composing V with the functor G. Since the underlying simplicial object of V is G-split, Remark
3.5.2 implies that V is a colimit diagram if and only X is a colimit diagram. Let W : N(∆−∞ )op → N([0],c0 )
be the map constructed in the argument above, and let W0 = W | N(∆op ). be the underlying augmented
simplicial object. Then g determines a natural transformation of augmented simplicial objects α : X → W0 ,
which induces an equivalence α[k] : X([k]) ' W0 ([k]) for all k ≥ 0. Lemma T.[Link] implies that W0 is a
colimit diagram in N([0],c0 ) . Consequently, X is a colimit diagram if and only if α[−1] : X([−1]) → W0 ([−1])
is an equivalence. In other words, g is a p-left Kan extension of g0 if and only if α[−1] is an equivalence.
Unwinding the definitions, we observe that this latter condition is equivalent to the requirement that g carry
the inclusion ([0], c0 ) → ([1], c) to a locally p-coCartesian morphism in N. We now observe that, in view (i00 )
and our assumption that g0 ∈ E00 , this last condition is equivalent to the requirement g ∈ ADatU (C, D).

3.6 Existence of Adjunction Data

Our goal in this section is to give the proof of Theorem 3.2.10. Fix a pair of ∞-categories C and D, and let
S = {C, D}. We wish to prove that every functor F : C → D which admits a right adjoint can be promoted
to an adjunction datum U . The proof is somewhat technical, and can be safely skipped by the reader; the
ideas introduced here will not be needed elsewhere.
The passage from F to U will proceed in stages. To describe these stages in more detail, we need to
introduce some terminology.
Definition 3.6.1. Let C and D be ∞-categories. A nonunital adjunction is a pair of functors F : C → D,
v α
G : D → C, and a pair of maps F ◦ G → U ← idD in the ∞-category Fun(D, D), where α is an equivalence.
We let Adj (C, D) denote the full subcategory of
nu

(Fun(C, D) × Fun(D, C)) ×End(D) Fun(Λ22 , End(D)) ×End(D) {idD }

119
spanned by the nonunital adjunctions. We will refer to Adjnu (C, D) as the ∞-category of nonunital adjunc-
tions. We let Adj(C, D) denote the full subcategory of Adjnu (C, D) spanned by those nonunital adjunctions
v ∼
F ◦ G → U ← idD

for which v is the counit of an adjunction between F and G.

Notation 3.6.2. Let the subcategory ∆nu

S ⊆ ∆S be defined as follows:

• Every object of ∆S belongs to ∆nu

S .

• A morphism α : ([m], c) → ([n], c0 ) of ∆S belongs to ∆nu

S if and only if, whenever α(i) = α(j) and
c(i) = c(j) = C, we have i = j.

Definition 3.6.3. Let q : End⊗ (C, D) → N(∆S )op denote the projection. A nonunital adjunction datum
⊗
is a functor F ∈ MapN(∆S )op (N(∆nuS ) , End (C, D)) which carries every C-convex morphism in ∆S to a
op nu

q-coCartesian morphism in End⊗ (C, D) (note that every C-convex morphism in ∆S is a morphism of ∆nu S ).
⊗
We let ADatnu (C, D) denote the full subcategory of MapN(∆S )op (N(∆nu
S ) , End (C, D)) spanned by the
op

nonunital adjunction diagrams.

S ⊆ ∆S induces a restriction functor ADat(C, D) → ADat (C, D). Similarly, evaluation

The inclusion ∆nu nu

on the diagram [D] ← [D, D] → [D, C, D] in ∆S determines a functor r : ADatnu (C, D) → Adjnu (C, D).
nu

Lemma 3.2.9 implies that the composition ADat(C, D) → ADatnu (C, D) → Adjnu (C, D) factors through the
full subcategory Adj(C, D). Consequently, we obtain a commutative diagram

ADat(C, D) / ADatnu (C, D)

 
 / Adjnu (C, D).
Adj(C, D)

We will break the proof of Theorem 3.2.10 into three steps:

Proposition 3.6.4. Let C and D be ∞-categories. Then the projection map Adjnu (C, D) → Fun(C, D)
induces a categorical equivalence Adj(C) → Fun0 (C, D), where Fun0 (C, D) is the subcategory of Fun(C, D)
whose objects are functors which admit right adjoints and whose morphisms are natural equivalences.
Proposition 3.6.5. Let C and D be ∞-categories. The restriction map r : ADatnu (C, D) → Adjnu (C, D) is
a trivial Kan fibration.
Proposition 3.6.6. Let C and D be ∞-categories. Then the restriction map ADat(C, D) → ADatnu (C, D)
induces an equivalence of ∞-categories

ADat(C, D) → ADatnu (C, D) ×Adjnu (C,D) Adj(C, D).

Let us grant these assertions for the moment.

Proof of Theorem 3.2.10. We observe that if F ∈ ADat(C, D), the induced object of Adjnu (C, D) belongs
to Adj(C, D) (Lemma 3.2.9). Consequently, the forgetful functor θ : ADat(C, D) → Fun(C, D) factors as a
composition of categorical equivalences
θ θ θ
ADat(C, D) →
0
ADatnu (C, D) ×Adjnu (C,D) Adj(C, D) →
1
Adj(C, D) →
2
Fun0 (C, D),

followed by the inclusion of Fun0 (C, D) into Fun(C, D). It follows that θ is a categorical equivalence. Since
θ is evidently a categorical fibration, it is a trivial Kan fibration.

120
 The proofs of Propositions 3.6.4, 3.6.5, and 3.6.6 are very different from one another. Proposition 3.6.4
is the easiest: it merely expresses the idea that when a functor F : C → D admits a right adjoint G, then G
is uniquely determined. Proposition 3.6.5 is more difficult. Roughly speaking, we need to show that every
nonunital adjunction F0 can be promoted to a nonunital adjunction datum F : N(∆nu S )
op
→ End⊗ (C, D).
nu
Our strategy is to choose an appropriate filtration of ∆S , and construct F from F0 by forming a series of
relative Kan extensions.
Proposition 3.6.6 involves rather different ideas. It asserts that a nonunital adjunction diagram F0 can be
extended to an adjunction diagram F if and only if the desired extension exists at the level of the homotopy
category; moreover, in this case, F is essentially unique. Our strategy is to reduce this assertion to an
analogous result for nonunital algebras, and then invoke the results of §2.8.
We begin with the proof of Proposition 3.6.4. The first step is to reduce the size of the ∞-category
Adjnu (C, D).
Notation 3.6.7. Let Adj− (C, D) denote the full subcategory of

(Fun(C, D) × Fun(D, C)) ×End(D) End(D)idD /

spanned by those transformations (F ◦ G) → idD which are counits for adjunctions between F and G. We
will regard Adj− (C, D) as a simplicial subset of Adj(C, D), whose vertices correspond to diagrams F ◦ G →
α
U ← idD for which α is a degenerate morphism in End(D).
Lemma 3.6.8. The inclusion i : Adj− (C, D) ⊆ Adj(C, D) is an equivalence of ∞-categories.
Proof. We have a homotopy Cartesian square

Adj− (C, D)
i / Adj(C, D)

 
∆0 / E,

where E is the ∞-category of initial objects of End(D)idD / . Since E is a contractible Kan complex, we
conclude that i is a categorical equivalence.
Lemma 3.6.9. The ∞-category Adj− (C, D) is a Kan complex.
Proof. We note that a morphism f in Adj− (C, D) can be identified with a pair of natural transformations
α : F → F 0 , β : G → G0 , and a commutative square

F ◦G
v / idD

α◦β =
 0 
F 0 ◦ G0
v / idD

in the ∞-category End(D). We wish to prove that α and β are equivalences. For this, we observe that α
induces a natural transformation of adjoint functors α0 : G0 → G, and β induces a natural transformation of
adjoint functors β 0 : F 0 → F . Using the commutativity of the above square, we deduce that α0 is a homotopy
inverse to β and β 0 is a homotopy inverse to α.
Lemma 3.6.10. The projection map

θ : Adjnu (C, D) → Fun(C, D)

induces a categorical equivalence θ0 : Adj− (C) → Fun0 (C, D), where Fun0 (C, D) is the subcategory of Fun(C, D)
whose objects are functors which admit right adjoints and whose morphisms are natural equivalences.

121
Proof. It is clear that the essential image of θ0 consists of functors which admit right adjoints, and Lemma
3.6.9 implies that the image of every morphism in Adj− (C, D) is an equivalence in Fun(C, D). Consequently,
θ0 factors through Fun0 (C, D) ⊆ Fun(C, D). It is easy to see that θ0 is a categorical fibration. Since
the domain and codomain of θ0 are Kan complexes (Lemma 3.6.9), we deduce that θ0 is a Kan fibration
(Propositions T.[Link], T.[Link], and Lemma T.[Link]). It will therefore suffice to show that every fiber
of θ0 is contractible. We now observe that, if F ∈ Fun0 (C, D), then the θ0−1 {F } can be identified with the
∞-category of initial objects of the fiber product Fun(D, C) ×End(D) End(D)idD / . Since this ∞-category is
nonempty, it is a contractible Kan complex (Proposition T.[Link]).
Proof of Proposition 3.6.4. Lemmas 3.6.10 and 3.6.8 imply that the restriction map Adj(C, D) → Fun(C, D)
r0
factors as a composition Adj(C, D) → Fun0 (C, D) ⊆ Fun(C, D). Since r0 is a categorical fibration, it will
suffice to show that r0 is a categorical equivalence. This follows immediately from Lemmas 3.6.10 and
3.6.8.

We now turn to the proof of Proposition 3.6.5. In what follows, we let q denote the projection map
End⊗ (C, D) → N(∆S )op .
Notation 3.6.11. We define full subcategories I0 ⊆ I1 ⊆ I2 ⊆ I3 ⊆ ∆nu
S as follows:

S belongs to I0 if and only if {i ∈ [n] : c(i) = C} is empty.

(J0 ) An object ([n], c) ∈ ∆nu

S belongs to I1 if and only if either ([n], c) ∈ I0 , or ([n], c) = [D, C, D].

(J1 ) An object ([n], c) ∈ ∆nu

S belongs to I2 if and only if {i ∈ [n] : c(i) = C} has at most one element.

(J2 ) An object ([n], c) ∈ ∆nu

S belongs to I3 if and only if {i ∈ [n] : c(i) = C} does not contain any pair of
(J3 ) An object ([n], c) ∈ ∆nu
consecutive integers {i, i + 1}.
⊗
For every full subcategory I ⊆ ∆nuS , we let ADatI (C, D) ⊆ MapN(∆S )op (N(I) , End (C, D)) denote the
nu op

full subcategory spanned by those functors F which carry every C-convex morphism in I to a q-coCartesian
morphism in End⊗ (C, D).
Proposition 3.6.5 is a consequence of the following sequence of lemmas:

I1 (C, D) → Adj (C, D) is a trivial Kan fibration.

Lemma 3.6.12. The restriction map r0 : ADatnu nu

I1 (C, D). Then:

Lemma 3.6.13. Let F0 ∈ ADatnu

(1) There exists a functor F ∈ MapN(∆S )op (N(I2 )op , End⊗ (C, D)) which is a q-left Kan extension of F0 .

(2) Let F ∈ MapN(∆S )op (N(I2 )op , End⊗ (C, D)) be an arbitrary extension of F0 . Then F is a q-left Kan
extension of F0 if and only if F ∈ ADatnuI2 (C, D).

I2 (C, D). Then:

Lemma 3.6.14. Let F0 ∈ ADatnu

(1) There exists a functor F ∈ MapN(∆S )op (N(I3 )op , End⊗ (C, D)) which is a q-right Kan extension of F0 .

(2) Let F ∈ MapN(∆S )op (N(I3 )op , End⊗ (C, D)) be an arbitrary extension of F0 . Then F is a q-right Kan
extension of F0 if and only if F ∈ ADatnuI3 (C, D).

I3 (C, D). Then:

Lemma 3.6.15. Let F0 ∈ ADatnu
⊗
S ) , End (C, D)) which is a q-left Kan extension of F0 .
(1) There exists a functor F ∈ MapN(∆S )op (N(∆nu op

⊗
(2) Let F ∈ MapN(∆S )op (N(∆nuS ) , End (C, D)) be an arbitrary extension of F0 . Then F is a q-left Kan
op

extension of F0 if and only if F ∈ ADatnu (C, D).

122
Proof of Proposition 3.6.5. The restriction map r : ADatnu (C, D) → Adjnu (C, D) factors as a composition
r
ADatnu (C, D) →
3
I3 (C, D)
ADatnu
r
I2 (C, D)
ADatnu
2
→
r
I1 (C, D)
ADatnu
1
→
r
0
→ Adjnu (C, D).

Lemma 3.6.12 implies that r0 is a trivial Kan fibration. Combining Lemmas 3.6.13, 3.6.14, and 3.6.15 with
Proposition T.[Link], we deduce that r1 , r2 , and r3 are trivial Kan fibrations. It follows that r = r0 ◦r1 ◦r2 ◦r3
is also a trivial Kan fibration.
Proof of Lemma 3.6.12. We have a coCartesian fibration p : N(I1 )op → ∆1 , where the inverse image of
{1} ⊆ ∆1 is N(I0 )op . Let K ⊆ N(I1 )op be the union of N(I0 ) and the p-coCartesian edge joining [D, C, D] to
[D, D]. Using Proposition T.[Link], we deduce that the inclusion K ⊆ N(I1 )op is a categorical equivalence.
Consequently, the restriction map

MapN(∆S )op (N(I1 )op , End⊗ (C, D)) → MapN(∆S )op (K, End⊗ (C, D))

is a trivial Kan fibration. We also observe that a functor F ∈ MapN(∆S )op (N(I1 )op , End⊗ (C, D)) belongs to
ADatnuI1 (C, D) if and only if F | N(I0 ) I0 (C, D). It will therefore suffice to show that the
belongs to ADatnu
op

restriction map

r00 : MapN(∆S )op (K, End⊗ (C, D)) ×MapN(∆ ⊗

op (N(I0 )op ,End (C,D)) I0 (C, D) → Adj (C, D)
ADatnu nu
S)

is a trivial Kan fibration. We observe that r00 is a pullback of the restriction map r000 : ADatnu I0 (C, D) → E,
where E is the ∞-category of initial objects in End(D)idD / and r000 is given by evaluation on the morphism
[D] → [D, D] in N(∆nu op
S ) .
We wish to show that r000 is a trivial Kan fibration. It is easy to see that r000 is a categorical fibration. It
will therefore suffice to show that r000 is a categorical equivalence. For this, it suffices to show that the source
and target of r000 are contractible Kan complexes. We observe that Proposition 1.4.3 allows us to identify
ADatnuI0 (C, D) with the ∞-category of initial objects in Alg(End(D)). The desired contractibility assertions
now follow immediately from Proposition T.[Link].
Proof of Lemma 3.6.13. We observe that for every object in ([n], c) ∈ I2 , the category I1 ×I2 (I2 )([n],c)/ has an
initial object. If ([n], c) ∈ I0 , this initial object is given by the identity map from ([n], c) to itself; otherwise,
it is given by the unique map ([n], c) → [D, C, D]. Since q is a coCartesian fibration, Lemma T.[Link]
immediately implies the following analogues of (1) and (2):
(10 ) For every functor F0 ∈ MapN(∆S )op (N(I1 )op , End⊗ (C, D)), there exists a functor

F ∈ MapN(∆S )op (N(I2 )op , End⊗ (C, D))

which is a q-left Kan extension of F0 .

(20 ) Let F ∈ MapN(∆S )op (N(I2 )op , End⊗ (C, D)). Then F is a q-left Kan extension of F0 = F | N(I1 )op if and
only if it satisfies the following condition:
(∗) For every object ([n], c) ∈ I2 which does not belong to I0 , F carries the canonical map α : ([n], c) →
[D, C, D] in I2 to a q-coCartesian morphism of End⊗ (C, D).
It is clear that (10 ) implies (1). To prove that (20 ) implies (2), we must show that a functor

F ∈ MapN(∆S )op (N(I2 )op , End⊗ (C, D))

123
belongs to ADatnu I2 (C, D) if and only if F satisfies (∗) and F | N(I1 ) belongs to ADatI1 (C, D). The “only if”
nu

direction is clear, since each of the morphisms α which appears in the statement of (∗) is C-convex.
Conversely, suppose that F0 = F | N(I1 ) belongs to ADatnu I1 (C, D) and that F satisfies (∗). Let α :
([m], c) → ([n], c0 ) be a C-convex morphism in I2 . We must show that F (α) is q-coCartesian. If ([n], c0 ) ∈ I0 ,
this follows from our assumption that F0 satisfies (∗). If ([n], c0 ) ∈/ I0 , then we have a commutative diagram

([n], c0 )
t: KKK
α tt
t KKKβ
tt KKK
t
tt K%
([m], c)
γ
/ [D, C, D].

According to Proposition T.[Link], it will suffice to show that F (β) and F (γ) are q-coCartesian. For β, this
/ I0 , then condition (∗) also guarantees that F (γ) is q-coCartesian. If
follows from condition (∗). If ([m], c) ∈
γ0 γ 00
([m], c) ∈ I0 , then γ factors as a composition ([m], c) → [D] → [D, C, D]. Then F (γ 0 ) is q-coCartesian (since
γ 0 is a C-convex morphism in I0 ) and F (γ 00 ) is q-coCartesian (since the ∞-category End⊗ (C, D)[D] is a Kan
complex). Applying Proposition T.[Link], we deduce that F (γ) is q-coCartesian, as desired.
Proof of Lemma 3.6.14. Fix an object ([n], c) ∈ I3 , and let J = I2 ×I3 (I3 )/([n],c) . Our first goal is to prove:
(a) Consider the commutative diagram

N(Jop
f
/ End⊗ (C, D)
_ )
q8
f q
q q
q
 q 
N(Jop )/ / N(∆S )op

where f denotes the composition

⊗ F
N(Jop ) → N(Iop
2 ) → End (C, D).
0

Then there exists a map f , as indicated, which is a q-limit diagram.

(b) An arbitrary extension f rendering the above diagram commutative is a q-limit diagram if and only if
the following condition is satisfied:
(?) For every convex morphism α : ([m], c0 ) → ([n], c) in J, the image of {α}/ ⊆ N(Jop )/ under f is a
q-coCartesian morphism of End⊗ (C, D).
Let J0 be the full subcategory of J spanned by those objects α : ([m], c0 ) → ([n], c) such that the induced
map [m] → [n] is injective. We can identify the category J0 with the partially ordered sets of nonempty
subsets S ⊆ [n] having the property that {j ∈ S : c(j) = C} has at most one element. Consider the following
further properties which S may or may not satisfy:

(P1 ) The set S contains {j ∈ [n] : c(j) = D}.

(P2 ) For some 0 < i < n such that c(i) = C, we have S = {i − 1, i + 1}.
(P3 ) For some 0 < i < n such that c(i) = C, we have S = {i − 1, i, i + 1}.

(P4 ) The set S consists of a single element.

(P5 ) For some 0 ≤ i < n, we have S = {i, i + 1}.

124
 For 1 ≤ i ≤ 5, let Ji denote the full subcategory of J0 spanned by those objects which satisfy (Pj ) for
some j ≥ i. Consider the following assertions:

(ai ) Consider the commutative diagram

N(Jop
fi
/ End⊗ (C, D)
i _ )
q8
fi q
q q
q
 q 
N(Jop / / N(∆S )op
i )

⊗ F
where fi denotes the composition N(Jop op
i ) ⊆ N(J ) → N(I2 ) → End (C, D). Then there exists a map
op 0

f i , as indicated, which is a q-limit diagram.

(bi ) An arbitrary extension f i rendering the above diagram commutative is a q-limit diagram if and only
if the following condition is satisfied:
(?0i ) For every morphism α : ([m], c0 ) → ([n], c) in J0 satisfying P5 , the image of {α}/ ⊆ N(Jop /
0 ) under
⊗
f i is a q-coCartesian morphism of End (C, D).

We note that, since F0 ∈ ADatnu I2 , if f is as in (b) and 0 ≤ i ≤ 5, then f satisfies (?) if and only if
op 0
f | N(Ji ) satisfies (?5 ).
The inclusions J1 ⊆ J, J2 ⊆ J3 , and J4 ⊆ J5 admit left adjoints, so the induced inclusions N(J1 ) ⊆ J,
N(J2 ) ⊆ N(J3 ), and N(J4 ) ⊆ N(J5 ) are cofinal. It follows that (a) ⇔ (a1 ), (a2 ) ⇔ (a3 ), and (a4 ) ⇔ (a5 ).
Using the assumption that F0 ∈ ADatnu I2 , we deduce that f1 is a q-right Kan extension of f2 and that f3 is
a q-right Kan extension of f4 . Applying Lemma T.[Link], we deduce that (a1 ) ⇔ (a2 ), and (a3 ) ⇔ (a4 ).
Composing these equivalences, we see that (a) is equivalent to (a5 ). The same argument shows that (b) is
equivalent to (b5 ). We now observe that (a5 ) and (b5 ) are obvious, since q is a coCartesian fibration which
exhibits End⊗ (C, D)([n],c) as equivalent to the product
Y
End⊗ (C, D)({i,i+1},c|{i,i+1}) .
0≤i<n

Assertion (1) now follows from (a) and Lemma T.[Link]. Moreover, (b) implies the following analogue
of (2):
(20 ) Let F ∈ MapN(∆S )op (N(I3 )op , End⊗ (C, D)) be an arbitrary extension of F0 . Then F is a q-right Kan
extension of F0 if and only if F satisfies the following condition:
(∗) For every convex morphism α : ([m], c) → ([n], c0 ) in I3 , where ([m], c) ∈ I2 , the image F (α) is a
q-coCartesian morphism in End⊗ (C, D).
To complete the proof, it will suffice to show that if F ∈ MapN(∆S )op (N(I3 )op , End⊗ (C, D)) is an extension of
F0 , then F satisfies (∗) if and only if F carries every C-convex morphism in I3 to a q-coCartesian morphism in
End⊗ (C, D). The “if” direction is obvious, since every convex morphism in I3 is C-convex. For the converse,
let us suppose that F satisfies (∗) and that α : ([m], c) → ([n], c0 ) is a C-convex morphism in I3 . We wish
to prove that F (α) is q-coCartesian. Using the product decomposition of End⊗ (C, D)([m],c) , we can reduce
to the case where m = 1, so that ([m], c) ∈ I2 . We now observe that α admits a unique factorization as a
α0 α00
composition ([m], c) → ([n0 ], c00 ) → ([n], c), where α0 preserves endpoints and α00 is convex. Assumption (∗)
guarantees that F (α00 ) is q-coCartesian. Moreover, using the assumption that α is C-convex, we deduce that
([n0 ], c00 ) ∈ I2 and the morphism α0 is C-convex. Since F0 ∈ ADatnu I2 (C, D), we conclude that F (α ) = F0 (α )
0 0

is q-coCartesian. Applying Proposition T.[Link], we deduce that F (α) is q-coCartesian, as desired.

The proof of Lemma 3.6.15 will require the following preliminary:

125
Lemma 3.6.16. Let p : X → S be a coCartesian fibration of ∞-categories, let K be a weakly contractible
simplicial set, and suppose given a diagram

K _
f0
/X
{=
f {
{ p
 { 
K.
g
/ S.

Suppose that f0 carries each edge of K to a p-coCartesian edge of X. Then:

(1) There exists a map f as indicated in the diagram, which is a p-colimit diagram.
(2) Let f be an arbitrary morphism which renders the above diagram commutative. The following conditions
are equivalent:
(i) The map f is a p-colimit diagram.
(ii) For every vertex k of K, the image of {k}. under f is a p-coCartesian morphism of X.
(iii) There exists a vertex k of K such that the image of {k}. under f is a p-coCartesian morphism
of X.
Proof. Using Proposition T.[Link], we can reduce to the case where the map g is constant at some object s
of S, so that f0 factors through the fiber Xs . Corollary T.[Link] guarantees the existence of an extension
f satisfying (ii). Moreover, Corollary T.[Link] also ensures that for every edge s → s0 in S, the image of f
under the associated functor Xs → Xs0 is a colimit diagram in Xs0 . Proposition T.[Link] implies that f is
a p-colimit diagram. This proves (1), as well as the implication (ii) → (i) of (2). The implication (i) → (ii)
follows from the uniqueness of relative colimit diagrams, and the equivalence (ii) ⇔ (iii) follows from the
connectedness of K.

S , and let J = I3 ×∆S (∆S )([m],c) . Let J0 denote the full

Proof of Lemma 3.6.15. Fix an object ([m], c) ∈ ∆nu
subcategory of J spanned by those morphisms α : ([m], c) → ([n], c0 ) with the following property:
(?) The map α preserves endpoints, and for 0 ≤ i < m, either α(i + 1) = α(i) + 1, or c(i) = c(i + 1) = C,
α(i + 1) > α(i) + 1, and c0 (j) = D for α(i) < j < α(i + 1).
The inclusion J0 ⊆ J admits a right adjoint, so that the inclusion N(J0 )op ⊆ N(J)op is cofinal. We observe
that J0 is equivalent to a product of copies of the category ∆, where the product is indexed by integers
i such that 0 ≤ i < i + 1 ≤ m and c(i) = c(i + 1) = C. In particular, the simplicial set N(J0 ) is weakly
contractible.
I3 (C, D), the composition
We observe that, since F0 ∈ ADatnu

F
N(J0 )op ⊆ N(J)op → N(I3 )op →0 End⊗ (C, D)op

carries each morphism in J0 to q-coCartesian edge of End⊗ (C, D). Combining Lemma 3.6.16 with Lemma
T.[Link], we deduce (1) and the following analogue of (2):
⊗
(20 ) Let F ∈ MapN(∆S )op (N(∆nu S ) , End (C, D)) be an arbitrary extension of F0 . Then F is a q-left Kan
op

extension of F0 if and only if F satisfies the following condition:

(∗) For every morphism α : ([m], c) → ([n], c0 ) in ∆nu
S which satisfies ?, the image F (α) is q-
coCartesian.
⊗
S ) , End (C, D)) is an extension
To complete the proof, it will suffice to show that if F ∈ MapN(∆S )op (N(∆nu op

of F0 , then F satisfies (∗) if and only if F carries every C-convex morphism of ∆nu S to a q-coCartesian
morphism in End⊗ (C, D). The “if” direction is obvious, since every morphism which satisfies (?) is C-convex.

126
 For the converse, let us suppose that F satisfies (∗) and that α : ([m], c) → ([n], c0 ) is a C-convex morphism
of ∆nu 0
S . We wish to prove that F (α) is q-coCartesian. First, choose a morphism β : ([n], c ) → ([p], c )
00

satisfying (?). Condition (∗) guarantees that F (β) is q-coCartesian. In view of Proposition T.[Link], it
will suffice to show that F (β ◦ α) is q-coCartesian. In other words, we may replace ([n], c0 ) by ([p], c00 ) and
thereby reduce to the case where ([n], c0 ) ∈ I3 . We now observe that α factors as a composition

α0 α00
([m], c) → ([n0 ], c00 ) → ([n], c),

where α0 satisfies (?) and α00 is C-convex. Condition (∗) implies that F (α0 ) is q-coCartesian, and our
assumption that F0 ∈ ADatnu I3 (C, D) guarantees that F (α ) is q-coCartesian. Invoking Proposition T.[Link],
00

we deduce that F (α) is q-coCartesian, as desired.

We now give the proof of Proposition 3.6.6. We begin with a rough outline. Every nonunital adjunction
datum F0 determines a nonunital monad on C: that is, a nonunital algebra T0 ∈ Algnu (End(C)). If F0
extends to an adjunction, then T extends to an algebra object in End(C). We would like to prove a converse
to this result: that is, the only obstruction to the existence of a adjunction datum extending F0 is the
existence of a monad extending T0 . Actually, this is not quite true: we will need to know not only that T
exists, but that various (nonunital) T0 -modules can be extended to (unital) T -modules. Nevertheless, the
arguments below will allow us to reduce to the problems regarding the existence and uniqueness of units
which were treated in §2.2 and §3.6.
We begin by introducing a bit of notation.

Notation 3.6.17. Let J denote the full subcategory of Fun([1], ∆S ) spanned by those morphisms α :
([m], c0 ) → ([n], c) which induce a bijection {i ∈ [m] : c0 (i) = D} → {j ∈ [n] : c(j) = D}. We define a
subcategory Jnu ⊆ J as follows:
• Every object of J belongs to Jnu .
• Given a pair of objects α, α0 ∈ J, a morphism

([m], c0 )
α / ([n], c)

β γ
 0 
([m0 ], c00 )
α / ([n0 ], c0 )

in J belongs to Jnu if and only if the morphism β induces an injection {i ∈ [m0 ] : c00 (i) = C} → {j ∈
[m1 ], c01 (j) = C}.
We will identify ∆S with the full subcategory of J spanned by constant functors [1] → ∆S . We observe
that the intersection Jnu ∩ ∆S can be identified with the full subcategory ∆nuS ⊆ ∆S . The inclusions ∆S ⊆ J
and ∆nuS ⊆ J nu
admit right adjoints ψ and ψ nu
, both given by the formula

(α : ([m], c0 ) → ([n], c)) 7→ id([m],c0 ) .

^ D) denote the full subcategory of MapN(∆ )op (N(J)op , End⊗ (C, D)) spanned by those
We let ADat(C, S

functors F : N(J)op → End⊗ (C, D) such that q ◦ F = N(ψ)op , and whenever γ : α → α0 is a morphism
in J such that ψ(γ) : ψ(α) → ψ(α0 ) is C-convex, the morphism F (γ) is a q-coCartesian. Similarly, we let
nu
^ (C, D) denote the full subcategory of MapN(∆ )op (N(Jnu )op , End⊗ (C, D)) spanned by those functors
ADat S

F0 : N(Jnu )op → End⊗ (C, D) such that q ◦ F0 = N(ψ nu )op , and whenever γ : α → α0 is a morphism in Jnu
such that ψ(γ) : ψ(α) → ψ(α0 ) is C-convex, the morphism F0 (γ) is a q-coCartesian.

127
Lemma 3.6.18. The restriction maps
nu
^ D) → ADat(C, D)
ADat(C, ^ (C, D) → ADatnu (C, D)
ADat

are trivial Kan fibrations.

Proof. We will give the proof in the unital case; the nonunital case uses exactly the same argument. According
to Proposition T.[Link], it will suffice to show:

(a) A functor F ∈ MapN(∆S )op (N(J)op , End⊗ (C, D)) belongs to ADat(C,
^ D) if and only if F0 = F | N(∆S )op
belongs to ADat(C, D), and F is a q-right Kan extension of F0 .
(b) Every functor F0 ∈ ADat(C, D) admits a q-right Kan extension

F ∈ MapN(∆S )op (N(J)op , End⊗ (C, D)).

We observe that a functor F ∈ MapN(∆S )op (N(J)op , End⊗ (C, D)) is a q-right Kan extension of F0 =
F | N(∆S )op at an object α ∈ J if and only if F carries the counit map vα : ψ(α) → α to an equivalence in
End⊗ (C, D). Assertion (b) now follows from Lemma T.[Link], and the “only if” direction of (a) is obvious
from the definitions. For the converse, let us suppose that F carries each vα to an equivalence in End⊗ (C, D),
and that F0 ∈ ADat(C, D). We wish to prove that F ∈ ADat(C, ^ D). Let γ : α → α0 be a morphism in J
such that ψ(γ) is C-convex. We wish to prove that F (γ) is q-coCartesian. Since F is a q-right Kan extension
of F0 , the morphisms F (γ) and F (ψ(γ)) are equivalent. Moreover, F (ψ(γ)) = F0 (ψ(γ)) is q-coCartesian in
virtue of our assumption that F0 ∈ ADat(C, D).
nu
^ D) and ADat
The advantage of the ∞-categories ADat(C, ^ (C, D) is that they are assembled out of
“local” information. To be more precise, we need to introduce a bit more notation.
Notation 3.6.19. Let ∆0S denote an isomorphic copy of the category ∆S (but with a new name, to avoid
confusion). Let π : J → ∆0S be defined by the formula

(α : ([m], c) → ([n], c0 )) 7→ ([n], c0 ).

We define a pair of simplicial sets M, Mnu equipped with maps M → Mnu → N(∆0S )op . These simplicial
sets are characterized by the property that, for every simplicial set K equipped with a map K → N(∆0S )op ,
there are canonical bijections

HomN(∆0S )op (K, M) ' HomN(∆S )op (K ×N(∆0S )op N(J)op , End⊗ (C, D))

HomN(∆0S )op (K, Mnu ) ' HomN(∆S )op (K ×N(∆0S )op N(Jnu )op , End⊗ (C, D)).
nu
^ (C, D) with a full subcategory of
It follows immediately from the definitions that we can identify ADat
the ∞-category MapN(∆0S )op (N(∆S ) , M) of sections of p . Similarly, we can identify ADat(C,
nu op nu ^ D) with
the full subcategory
nu
MapN(∆0S )op (N(∆0S )op , M) ×MapN(∆0 )op (N(∆nu
S )
^ (C, D)
op ,M) ADat
S

of MapN(∆0S )op (N(∆0S )op , M).

Proposition 3.6.6 is be a consequence of the following result:
Lemma 3.6.20. There exists a subcategory Mnu
0 ⊆M
nu
with the following properties:
(1) Let M0 = Mnu
0 ×Mnu M. Then the projection M0 → M0 is a trivial Kan fibration.
nu

128
 nu
^ (C, D) be such that the image of F in Adjnu (C, D) belongs to Adj(C, D). Then the
(2) Let F ∈ ADat
associated section N(∆0S )op → Mnu of pnu factors through Mnu
0 .

Let us assume Lemma 3.6.20 for the moment.

Proof of Proposition 3.6.6. In view of Lemma 3.6.18, it will suffice to show that the restriction map r :
nu
^ D) → ADat
ADat(C, ^ (C, D) induces an equivalence of ∞-categories
nu
^ D) → ADat
r0 : ADat(C, ^ (C, D) ×Adjnu (C,D) Adj(C, D).
nu
^ (C, D)×Adjnu (C,D) Adj(C, D), so that r0 is well-defined.
Lemma 3.2.9 guarantees that r factors through ADat
Now suppose that M0 and M0 are as in Lemma 3.6.20. Then r0 is a pullback of the map
nu

MapN(∆0S )op (N(∆0S ), M0 ) → MapN(∆0S )op (N(∆0S ), Mnu

0 ).

Since the projection M0 → Mnu

0 is a trivial Kan fibration, we conclude that r0 is a trivial Kan fibration.

0 ⊆ M
In order to produce the full subcategory Mnu nu
whose existence is asserted by Lemma 3.6.20, we
will need to understand the ∞-categories M and M0 a bit better. For this, we need a bit more notation.
Notation 3.6.21. Let ∆S,+ be the category obtained from ∆S by adjoining a new initial object; we will
think of this initial object as corresponding to a pair (∅, c), where c : ∅ → {C, D} denotes the empty word.
Fix an object σ = ([n], c) ∈ ∆S,+ . We define functors η L R
σ , η σ : ∆ → ∆S as follows:
0 00
ηL
σ ([m]) = ([m] ? [n], c ) ηR
σ ([m]) = ([n] ? [m], c ),

where ( (
0 C if 0 ≤ i ≤ m 00 c(i) if 0 ≤ i ≤ n
c (i) = c (i) =
c(j) if i = n + 1 + j > n C if i > n.
We will let Fun⊗ (σ, C) denote the fiber product N(∆)op ×N(∆S )op End⊗ (C, D), where N(∆)op maps to
N(∆S )op via η L ⊗
σ , and Fun (C, σ) the fiber product N(∆)
op
×N(∆S )op End⊗ (C, D) where N(∆)op maps to
op R
N(∆S ) via η σ .
We observe that there is a commutative diagram

Fun⊗ (C, σ) / End⊗ (C) o Fun⊗ (σ, C)

MMM qq
MMM qqq
MMM qq
M&  xqqq
N(∆)op .

This diagram exhibits Fun(σ, C) = Fun⊗ (σ, C)[0] as left-tensored over the monoidal ∞-category End(C)
and Fun(C, σ) = Fun⊗ (C, σ)[0] as right-tensored over the monoidal ∞-category End(C). Note also that if
σ = (∅, c), then the horizontal arrows in the above diagram are isomorphisms.
Notation 3.6.22. Fix an object ([n], c) ∈ ∆0S , and let Σ = {i ∈ [n] : c(i) = C}.QWe observe that the fiber
π −1 {([n], c)} can be identified with a full subcategory D of the product ∆Σ+ = i∈T ∆+ ; this subcategory
either coincides with ∆Σ + (if Σ 6
= [n]) or with the subcategory obtained by deleting the initial object of ∆S+
(if Σ = [n]).
ψ E
Let E ([n],c) : D → Set∆ denote the composition D ' π −1 {([n], c)} ⊆ J → ∆S → Set∆ , where the functor
([n],c)
E is defined as in Notation 3.2.2. Let E+ + → Set∆ be given by the formula
: ∆Σ
(
([n],c) E ([n],c) (X) if X ∈ D
E+ (X) =
∆0 otherwise.

129
 Let ∆nu
+ ⊆ ∆+ be the subcategory consisting of all objects of ∆+ and injective maps between them. For
every subset Σ0 ⊆ Σ, we set
Y
T (Σ0 ) = ( N(∆nu op
+ ) ) × j∈Σ N(∆+ ) .
Q op
0
i∈Σ−Σ0

V Σ0 ([n], c) = MapN(∆Σ+ )op (T (Σ0 ), NE ([n],c) ((∆Σ op

+ ) )).
+

Let V ([n], c) = V ([n], c) and V ([n], c) = V ([n], c). We observe that V + and V − can be viewed as
+ Σ − ∅

functors from (∆0S )op to the category of simplicial sets, and that we have canonical isomorphisms

M ' NV + ((∆0S )op ) Mnu ' NV − ((∆0S )op ).

Moreover, the restriction map M → Mnu is induced by a natural transformation V + → V − . More generally
we have canonical maps V Σ0 ([n], c) → V Σ1 ([n], c) for every Σ1 ⊆ Σ, given by restriction of functors.
Notation 3.6.23. Let ([n], c) and Σ be as in Notation 3.6.22. For every morphism α : ([m], c0 ) → ([n], c)
in J, we set
[m]+ = {j ∈ [m] : α(j) > i} [m]− = {j ∈ [m] : α(j) < i}
σαL = ([m]+ , c0 |[m]+ } ∈ ∆S,+ σαR = ([m]− , c0 |[m]− } ∈ ∆S,+ .
Let π : (∆nu Σ
+ ) → (∆+ )
nu Σ−{i}
be the restriction map, and fix X ∈ (∆nu + )
Σ−{i}
. We observe that, if
X ∈ π {X} corresponds to an object α ∈ J([n],c) , then σα , σα ∈ ∆S,+ are independent of the choice of X.
−1 L R
L
We will denote these words by σX and σXR
, respectively. Restriction to π −1 {X} determines functors

li,X : V − ([n], c) → MapN(∆)op (N(∆nu )op , Fun⊗ (σX

L
, C))

ri,X : V − ([n], c) → MapN(∆)op (N(∆nu )op , Fun⊗ (C, σX

R
)).
We will say that an object v ∈ V − ([n], c) is i-good if, for each X ∈ (∆nu + )
Σ−{i}
, the functors li,X and
⊗ L ⊗
ri,X carry v to quasi-unital module objects of Fun (σX , C) and Fun (C, σX ), respectively In this case, both
R

quasi-unital modules have the same underlying quasi-unital algebra Ai,X : N(∆nu )op → End⊗ (C). We will
say that a morphism v → v 0 in V − ([n], c) is i-good if v and v 0 are i-good and the induced map of nonunital
algebras Ai,X → A0i,X is quasi-unital, for every X ∈ (∆nu + )
Σ−{i}
.
−
Let us say that an object v ∈ V ([n], c) is good if it is i-good for each i ∈ Σ, and we will say that a
morphism v → v 0 in V − ([n], c) is good if it is i-good for each i ∈ Σ. More generally, for each Σ0 ⊆ Σ, we will
say that an object or morphism in V Σ0 ([n], c) is good if its image in V − ([n], c) is good. The collection of good
objects and morphsims in V Σ0 ([n], c) determines a subcategory V0Σ0 ([n], c) ⊆ V Σ0 ([n], c). In particular, we
have subcategories (generally not full)

V0+ ([n], c) ⊆ V + ([n], c) V0− ([n], c) ⊆ V − ([n], c).

It is not difficult to see that this construction determines a pair of functors V0+ , V0− : (∆0S )op → Set∆ . We
now define
M0 ' NV + ((∆0S )op ), Mnu 0 op
0 ' NV − ((∆S ) ).
0 0

We are now almost ready to give the proof of Lemma 3.6.20. We require only one more preliminary:
Lemma 3.6.24. Let A ⊆ B be a right anodyne inclusion of simplicial sets, K an arbitrary simplicial set,
and p : X → Y a categorical fibration. Then the map
a
Fun(B . × K, X) → Fun(B . × K, Y ) ×Fun((B×K) ‘A×K (A. ×K),Y ) Fun((B × K) (A. × K), X)
A×K

is a trivial Kan fibration.

130
Proof. Replacing p by the induced map ` X K → Y K , we can reduce to the case where K = ∅. It now
suffices to show that the inclusion i : B A A. → B . is a trivial cofibration (with respect to the Joyal
model structure). This is equivalent to the assertion that i has the left lifting property with respect to
every categorical fibration. In fact, i has the left lifting property with respect to every inner fibration (see
Proposition T.[Link]).
Remark 3.6.25. According to Proposition T.[Link], an inclusion of simplicial sets is right anodyne if
and only if it is cofinal. In particular, Lemma 3.6.24 applies in the special case where B = N(∆)op and
A = N(∆nu )op (Lemma T.[Link]).
Proof of Lemma 3.6.20. We evidently have a Cartesian diagram

M0 /M

f
 
Mnu / Mnu .
0

It remains only to show that the subcategory Mnu 0 ⊆M

nu
satisfies conditions (1) and (2).
We first prove (1). We wish to show that the projection f : M0 → Mnu 0 is a trivial Kan fibration. Since
f is a categorical fibration (Lemma T.[Link]), it will suffice to show that f is a categorical equivalence.
According to Proposition T.[Link], it will suffice to show that for every object ([n], c) ∈ ∆0S , the induced
map f([n],c) : V0+ ([n], c) → V0− ([n], c) is a categorical equivalence.
Let Σ = {i ∈ [n] : c(i) = C}. We will prove that, for each Σ1 ⊆ Σ0 ⊆ Σ, the restriction map
V0Σ0 ([n], c) → V0Σ1 ([n], c) is a trivial Kan fibration. Arguing inductively, it suffices to treat the case where
Σ0 = Σ1 ∪ {i}. Let T = ∆Σ + ×(∆+ )
1 nu Σ−Σ0
, and set

Y = MapN(∆Σ+ )op (N(T )op × N(∆)op , NE ([n],c) ((∆S+ )op ))

+

Y nu = MapN(∆Σ+ )op (N(T )op × N(∆nu )op , NE ([n],c) ((∆Σ op

+ ) )).
+

Let Y0nu be a subcategory of Y nu which contains the image of the restriction map V0S1 ([n], c) → Y nu , and
set Y0 = Y ×Y nu Y0 . We have a commutative diagram of simplicial sets

V0Σ0 ([n], c)
g
/ V Σ1 ([n], c) ×Y nu Y / Y0 /Y
QQQ 0
QQQ
QQQ g 00 g0
QQQ
(   
V0Σ0 ([n], c) / Y0nu / Y nu .

Using Lemma 3.6.24 (and Remark 3.6.25), we deduce that g is a trivial Kan fibration. Since each square in
the diagram is a pullback, it will suffice to show that g 0 is also a trivial Kan fibration, provided that Y0nu is
appropriately chosen.
We now analyze the map g 0 . For each object t of T , consider the composition
([n],c)
E+
Et : ∆op ' (∆{i} )op × {t} ⊆ (∆S+ )op → Set∆ .

Set
H(t) = MapN(∆)op (N(∆)op , NEt (∆op )), H nu (t) = MapN(∆)op (N(∆nu )op , NEt (∆op )).
We observe that, as in Notation 3.6.22, the object t ∈ T determines a pair of words σtL , σtR ∈ ∆C,D,+ and
functors
lt : H nu (t) → MapN(∆)op (N(∆nu )op , Fun⊗ (σtL , C)

131
 rt : H nu (t) → MapN(∆)op (N(∆nu )op , Fun⊗ (C, σtR ).
Let H0nu (t) denote the intersection of lt−1 Modqu (Fun(σtL , C)) with rt−1 Modqu (Fun(C, σtR )) (as subcategories
of H nu (t), and let H0 (t) = H(t) ×H nu (t) H0nu (t). This construction determines functors

H, H nu , H0 , H0nu : T op → Set∆ .

Set Z = NH (T op ), and define Z0 , Z nu , and Z0nu similarly. We have canonical isomorphisms

Y ' MapN(T )op (N(T )op , Z) Y nu ' MapN(T )op (N(T )op , Z nu )

Set Y0 = MapN(T )op (N(T )op , Z0 ) and Y0nu = MapN(T )op (N(T )op , Z0nu ). It is clear that we have a Cartesian
diagram
Y0 /Y

g0
 
Y0nu / Y nu ,

and that the restriction map V0S1 ([n], c) → Y nu factors through Y0nu . It remains only to show that g 0 is a
trivial fibration. For this, it will suffice to show that the map h : Z0 → Z0nu is a trivial Kan fibration. The map
h is a categorical fibration (Lemma T.[Link]), so it will suffice to show that h is a categorical equivalence.
Invoking Proposition T.[Link], we are reduced to proving that for each t ∈ T , the map ht : H0 (t) → H0nu (t)
is an equivalence of ∞-categories. We observe that ht is induced by a map between the pullback squares

H0 (t) / Mod(Fun(σ L , C)) H0nu (t) / Modqu (Fun(σ L , C))

t t

   
Mod(Fun(C, σtR )) / Alg(End(C)) Modqu (Fun(C, σtR )) / Algqu (End(C)).

The desired result now follows from Theorem 2.8.1 and (two applications of) Corollary 2.8.5. This completes
the proof of (1).
nu v
We now prove (2). Let s ∈ ADat^ (C, D), and let F ◦ G → U ' idD denote the associated object in
Adj (C, D). Suppose that v is the counit of an adjunction between F and G. We wish to prove that the
nu

associated section of the projection M → N(∆0S )op factors through M0 . We observe that s determines a
nonunital algebra T ∈ Algnu (End(C)), whose underlying object of End(C) can be identified with G ◦ F .
Unwinding the definitions, we see that the desired result is equivalent to the following assertions:
(i) The nonunital algebra T ∈ Algnu (End(C)) admits a quasi-unit u.

(ii) The left actions of T on G and G ◦ F are quasi-unital.

(iii) The right actions of T on F and F ◦ G are quasi-unital.
A quasi-unit u for T can be identified with a morphism u : idC → G ◦ F . Moreover, it is easy to see that u
satisfies (i), (ii), and (iii) if and only if the compositions
u v
F ' F ◦ idC → F ◦ (G ◦ F ) = (F ◦ G) ◦ F → idD ◦F = F
u v
G ' idC ◦G → (G ◦ F ) ◦ G = G ◦ (F ◦ G) → G ◦ idD = G
are homotopic to the identity. In other words, (2) is equivalent to the assertion that there exists a map
u : idC → G ◦ F which is a compatible unit for the adjunction determined by the counit v. This follows
immediately from our assumption that the image of s lies in Adj(C, D) ⊆ Adjnu (C, D).

132
4 The Monoidal Structure on Stable Homotopy Theory
Let Sp denote the ∞-category of spectra (as defined in §S.9). The homotopy category hSp can be identified
with the classical stable homotopy category. Given a pair of spectra X, Y ∈ hSp, one can define new spectrum
called the smash product of X and Y . The smash product operation determines a monoidal structure on
hSp. In this section, we will show that this monoidal structure is determined by a monoidal structure which
exists on the ∞-category Sp itself. There are at least three ways to see this.

(S1) Choose a simplicial model category A equipped with a compatible monoidal structure, whose under-
lying ∞-category is equivalent to Sp. For example, we can take A to be the category of symmetric
spectra (see [28]). According to Proposition 1.6.5, the underlying ∞-category N(Ao ) ' Sp is endowed
with a monoidal structure. The advantage of this perspective is that it permits us to easily compare
the algebras and modules considered in this paper with more classical approaches to the theory of
structured ring spectra. For example, Theorem 1.6.17 implies that Alg(Sp) is (equivalent to) the ∞-
category underlying the model category of algebras in symmetric spectra (that is, strictly associative
monoids in A); see Example 1.6.19.
The main disadvantage of this approach is that it seems to require auxiliary data (namely, a strictly
associative model for the smash product functor), which could be supplied in many different ways.
From a conceptual point of view, the existence of such a model ought to be irrelevant: the very
purpose of higher category theory is to provide a formalism which allows us to avoid assumptions like
strict associativity.
(S2) Let FunL (Sp, Sp) denote the full subcategory of Fun(Sp, Sp) spanned by those functors from Sp to
Sp which preserve small colimits. Corollary S.15.6 asserts that evaluation on the sphere spectrum
yields an equivalence of ∞-categories FunL (Sp, Sp) → Sp. On the other hand, since FunL (Sp, Sp) is
stable under composition in Fun(Sp, Sp), the composition monoidal structure on Fun(Sp, Sp) induces
a monoidal structure on FunL (Sp, Sp). This definition has the virtue of being very concrete (the smash
product operation is simply given by composition of functors), and it allows us to identify the algebra
objects of FunL (Sp, Sp): they are precisely the colimit-preserving monads on the ∞-category Sp (for an
application of this last observation, see Theorem 4.4.9). The disadvantage of this definition is that it
is very “associative” in nature, and therefore does not generalize easily to show that Sp is a symmetric
monoidal ∞-category (as we will see in [43]).
(S3) Let LPr be the ∞-category whose objects are presentable ∞-categories and whose morphisms are
LPr,σ
colimit-preserving functors (see §T.5.5.3), and let Cat
d
∞ be the full subcategory of LPr spanned by
LPr,σ
those presentable ∞-categories which are stable. As we will explain below, Cat d
∞ admits a monoidal
LPr,σ
structure. Moreover, the algebra objects of Cat∞ d can be identified with monoidal ∞-categories
C which are stable, presentable, and have the property that the bifunctor ⊗ : C × C → C is colimit
LPr,σ
preserving separately in each variable. We will show that Sp is the unit object of Cat
d
∞ (with respect
to its tensor structure). It follows from Proposition 1.4.3 that Sp can be endowed with the structure
LPr,σ
of an algebra object of Cat
d
∞ , and in fact is initial among such algebra objects. This establishes not
only the existence of a monoidal structure on Sp, but also a universal property which can be used to
prove uniqueness (Corollary 4.2.6).

We will follow approach (S3). Our first step is to construct a monoidal structure on the ∞-category LPr
of presentable ∞-categories. This construction will be carried out in §4.1. Roughly speaking, given a pair
of presentable ∞-categories C and D, the tensor product C ⊗ D is universal among presentable ∞-categories
which receive a bifunctor C × D → C ⊗ D which preserves (small) colimits separately in each variable. In §4.2,
we will see that this monoidal structure induces (via the constructions of §1.3) another monoidal structure
on the ∞-category of stable presentable ∞-categories, which we can use to carry out the reasoning outlined
in the above discussion.

133
 We will define an A∞ -ring to be an algebra object of the ∞-category Sp. In §4.3, we will study some of
the basic formal properties of the ∞-category of A∞ -rings. For example, we will show that an A∞ -ring A
satisfying πi A ' 0 for all i 6= 0 is essentially the same thing as an associative ring, in the sense of classical
algebra (Proposition 4.3.13).
If A is an A∞ -ring, then we have an associated ∞-category ModL A = ModA (Sp) of left A-module spectra,
or simply left A-modules. In §4.4, we will study the ∞-category ModL A in some detail. In particular, we will
show that ModL A is a stable ∞-category, so that the homotopy category hModLA is triangulated (Proposition
4.4.3). In the case where A corresponds to an ordinary associative ring, hModLA can be identified with the
usual derived category of that associative ring (see Proposition 4.4.7 and Remark 4.4.8).
There is also an (entirely dual) theory of right modules over A. Given a right A-module M and every left
A-module N , one can construct a relative tensor product M ⊗A N ∈ Sp. We will describe this construction (in
a very general setting) in §4.5. In §4.6, we will use the theory of relative tensor products to introduce a theory
of flat modules. Roughly speaking, a left A-module N is flat if the tensor product functor M 7→ M ⊗A N
has good exactness properties (at least when A is connective). Our main result is an analogue of Lazard’s
theorem: any flat A-module can be obtained as a filtered colimit of free A-modules (Theorem 4.6.19). To
prove this, we will introduce a spectral sequence which can be used (in favorable cases) to compute the
homotopy groups πn (M ⊗A N ) in terms of the homotopy groups of M , A, and N .
A theorem of Schwede and Shipley asserts that a stable ∞-category C can be realized as the ∞-category
of left modules over an A∞ -ring A if and only if C is presentable and compactly generated (we will give
a proof of this result in §4.4, using the ∞-categorical Barr-Beck theorem). In particular, the ∞-category
ModL L
A is compactly generated. The compact objects of ModA are called perfect A-modules. We will study
the class of perfect A-modules in §4.7, together with the somewhat larger class of almost perfect A-modules
which arise frequently in applications.

4.1 Tensor Products of Presentable ∞-Categories

Our goal in this section is to show that the ∞-category LPr of presentable ∞-categories admits a monoidal
structure. This monoidal structure can be described informally as follows: given a pair of presentable ∞-
categories C and D, the tensor product C ⊗ D is the recipient of a universal bifunctor C × D → C ⊗ D which
is “bilinear”; that is, which preserves colimits separately in each variable.
We begin by introducing a bit of notation.

Notation 4.1.1. Let Catd∞ denote the ∞-category of (not necessarily small) ∞-categories. Then Cat
d∞
⊗
admits finite products. Consequently, there exists a monoidal ∞-category Cat
d endowed with a Cartesian
∞
⊗ ⊗
structure Cat → Cat∞ which induces an equivalence (Cat )[1] ' Cat∞ .
d
∞
d d d
∞
⊗
We can construct an explicit model for Cat
d as follows. Let A be the category of (not necessarily small)
∞
marked simplicial sets, as in Example 1.6.20. Then A is endowed with the structure of a monoidal model
⊗
category, with monoidal structure given by Cartesian product. We now let Cat
d = N(A⊗,o ), as defined in
∞
the statement of Proposition 1.6.5. More concretely:
⊗
(i) The objects of Cat
d are finite sequences [X1 , . . . , Xn ], where each Xi is an ∞-category.
∞
⊗
(ii) Given a pair of objects [X1 , . . . , Xn ], [Y1 , . . . , Ym ] ∈ Cat
d , a morphism from [X1 , . . . , Xn ] to [Y1 , . . . , Ym ]
∞
consists of an order-preserving map f : [m] → [n] and a collection of functors ηi : Xf (i−1)+1 × . . . ×
Xf (i) → Yi .
Pr,⊗ ⊗
We define a subcategory Cat
d
∞ ⊆ Cat
d as follows:
∞

⊗ Pr,⊗
(iii) An object [X1 , . . . , Xn ] ∈ Cat
d belongs to Cat
∞
d
∞ if and only if each Xi is a presentable ∞-category.

134
 Pr,⊗
(iv) Let [X1 , . . . , Xn ], [Y1 , . . . , Ym ] ∈ Cat
d
∞ , and let F : [X1 , . . . , Xn ] → [Y1 , . . . , Ym ] be a morphism in
⊗ Pr,⊗
Cat∞ covering a map f : [m] → [n] in ∆. Then F belongs to Cat
d d
∞ if and only if each of the induced
functors ηi : Xf (i−1)+1 × . . . × Xf (i) → Yi is colimit-preserving in each variable.
Pr,⊗
Our next goal is to show that Cat
d
∞ is a monoidal ∞-category. First, we need a lemma.
Notation 4.1.2. Let C and D be ∞-categories which admit small limits. Then we let FunR (C, D) denote
the full subcategory of Fun(C, D) spanned by those functors which preserve small limits. Similarly, if C and
D admit small colimits, we let FunL (C, D) denote the full subcategory of Fun(C, D) which preserve small
colimits.
Warning 4.1.3. We used the same notation in §T.5.2.6 for a slightly different purpose: there, FunL (C, D)
denoted the ∞-category of functors from C to D which were left adjoints, and FunR (C, D) the ∞-category of
functors which were right adjoints. In what follows below, we will work with presentable ∞-categories, so that
the two notations are almost consistent with one another in view of the adjoint functor theorem(Corollary
T.[Link]).
Lemma 4.1.4. Let C and D be presentable ∞-categories. Then FunR (Cop , D) is a presentable ∞-category.
Proof. Using Theorem T.[Link] and the results of §T.5.5.4, we can choose a small ∞-category C0 , a small
collection S of morphisms in P(C0 ), and an equivalence C ' S −1 P(C0 ). Then
op
FunR (P(C0 )op , D) ' FunL (P(C0 ), Dop )op ' Fun(C0 , Dop )op ' Fun(C0 , D)

is presentable by Proposition T.[Link], where the second equivalence is given by composition with the
Yoneda embedding (Theorem T.[Link]). For each morphism α ∈ S, let E(α) denote the full subcategory
of FunR (P(C0 )op , D) spanned by T
those functors which carry α to an equivalence in D. Then FunR (Cop , D)
is equivalent to the intersection α∈S E(α). In view of Lemma T.[Link], it will suffice to show that each
E(α) is a localization of FunR (P(C0 )op , D). We now observe that E(α) is given by a pullback diagram

E(α) / FunR (P(C00 )op , D)

  
E / Fun(∆1 , D),

where E denotes the full subcategory of Fun(∆1 , D) spanned by the equivalences. According to Lemma
T.[Link], it will suffice to show that E is an accessible localization of Fun(∆1 , D), which is clear.
Lemma 4.1.5. Let C1 , . . . , Cn be a finite collection of presentable ∞-categories. Then there exists a pre-
sentable ∞-category C1 ⊗ . . .⊗Cn and a functor f : C1 × . . .×Cn → C1 ⊗ . . .⊗Cn with the following properties:
(1) The functor f preserves colimits separately in each variable.
(2) For every presentable ∞-category D, composition with f induces an equivalence from FunL (C1 ⊗ . . . ⊗
Cn , D) onto the full subcategory of Fun(C1 × . . . × Cn , D) spanned by those functors which preserve
colimits separately in each variable.
Proof. The proof goes by induction on n. If n = 0, we take C1 ⊗ . . . ⊗ Cn = S, choose the functor f to classify
a final object of S, and apply Theorem T.[Link]. If n = 1 there is nothing to prove. Assume therefore that
n > 1.
Let D be an arbitrary presentable ∞-category, and let Fun0 (C1 × . . . × Cn , D) be the full subcategory
of Fun(C1 × . . . × Cn , D) which preserve colimits separately in each variable. Then we have a canonical
isomorphism Fun0 (C1 × . . . Cn , D) ' Fun0 (C1 × . . . Cn−1 , FunL (Cn , D)). Here the ∞-category FunL (Cn , D) is

135
presentable (Proposition T.[Link]), so that the inductive hypothesis allows us to identify this ∞-category
with FunL (C1 ⊗ . . . ⊗ Cn−1 , FunL (Cn , D)). If n > 2, then we can conclude by identifing the last ∞-category
with FunL ((C1 ⊗ . . . ⊗ Cn−1 ) ⊗ Cn , D).
Suppose instead that n = 2. Using Corollary T.[Link] and Proposition T.[Link], we can identify
FunL (C2 , D) with the full subcategory of FunR (D, C2 )op spanned by those functors which are accessible.
Consequently, we get a fully faithful embedding
Fun0 (C1 × C2 , D) → FunL (C1 , FunR (D, C2 )op )
' FunL (C1 , FunL (Dop , Cop
2 ))
' FunL (Dop , FunL (C1 , Cop
2 ))
' FunR (D, FunR (Cop
1 , C2 ))
op

whose essential image consists of the collection of accessible functors from D to FunR (Cop 1 , C2 ). We now apply
Lemma 4.1.4 to conclude that FunR (Cop 1 , C2 ), so that (using Corollary T.[Link] and Proposition T.[Link]
again) Fun0 (C1 × C2 , D) can be identified with FunL (FunR (Cop 1 , C2 ), D). We now conclude by defining C1 ⊗ C2
to be FunR (Cop
1 , C2 ).

Remark 4.1.6. Let C1 , . . . , Cn be a finite sequence of presentable ∞-categories. It follows from the proof of
Lemma 4.1.5 that, for n > 0, the tensor product can be canonically identified with the iterated functor ∞-
category FunR (Cop R op R op
1 , Fun (C2 , . . . Fun (Cn−1 , Cn ) . . .)). Combining this observation with Theorem T.[Link],
we conclude that the bifunctor ⊗ : LPr × LPr → LPr preserves colimits separately in each variable (remember
that colimits in LPr can also be computed as limits in PrR , which are computed by forming limits in Cat∞
by Theorem T.[Link]). Alternatively, one can observe that LPr is actually a closed monoidal category, with
internal mapping objects given by FunL (C, D) (see Proposition T.[Link]).
Pr,⊗
Proposition 4.1.7. The natural map q : Cat d
∞ → N(∆)op determines a monoidal structure on the ∞-
Pr,⊗ Pr,⊗ ⊗
category LPr ' (Cat
d
∞ )[1] . Moreover, the inclusion functor Cat∞
d ⊆ Cat
d is lax monoidal.
∞

Proof. We will show that q is a coCartesian fibration; the remaining condition that q induces equivalences
Pr,⊗
(Cat
d n
∞ )[n] ' (LPr) will then follow by inspection. According to Proposition T.[Link], it will suffice to
verify the following conditions:
Pr,⊗
(1) Let α : [m] → [n] be a morphism in ∆, and let X[n] be an object of Cat
d
∞ lying over [n]. Then there
Pr,⊗
exists a locally q-coCartesian morphism X[n] → X[m] in Cat
d covering α.
∞

(2) Let α : [m] → [n] and β : [n] → [p] be morphisms in ∆, covered by locally q-coCartesian morphisms
X[p] → X[n] → X[m]
Pr,⊗
in Cat
d
∞ . Then the composite morphism X[p] → X[m] is locally q-coCartesian.

Using the product structure on the fibers of q, we can reduce to the case where m = 1 and the morphisms
α : [m] → [n] and β : [n] → [p] preserve the endpoints of intervals. Unwinding the definitions, we obtain the
following reformulation of conditions (1) and (2):
(10 ) Let C1 , . . . , Cn be a finite collection of presentable ∞-categories. There exists a presentable ∞-category
C1 ⊗ . . . ⊗ Cn and a functor f : C1 × . . . × Cn → C1 ⊗ . . . ⊗ Cn , which preserves colimits separately in
each variable and which is universal in the following sense:
(∗) Let D be a presentable ∞-category, let Fun0 (C1 × . . . × Cn , D) ⊆ Fun(C1 × . . . × Cn , D) be the full
subcategory spanned by those functors which preserve colimits separately in each variable. Then
the functor FunL (C1 ⊗ . . . ⊗ Cn , D) → Fun0 (C1 × . . . × Cn , D) given by composition with f induces
a homotopy equivalence between maximal Kan complexes contained in FunL (C1 ⊗ . . . ⊗ Cn , D)
and Fun0 (C1 × . . . × Cn , D).

136
 (20 ) Given an endpoint-preserving map α : [n] → [p] and a collection of presentable ∞-categories C1 , . . . , Cp .
Then the canonical map
C1 ⊗ . . . ⊗ Cp → (Cα(0)+1 ⊗ . . . ⊗ Cα(1) ) ⊗ . . . ⊗ (Cα(n−1)+1 ⊗ . . . ⊗ Cα(n) )
is an equivalence of (presentable) ∞-categories.
Assertion (10 ) follows immediately from Lemma 4.1.5, and (20 ) follows from Remark 4.1.6.

Remark 4.1.8. In view of Remark 1.2.15, we can identify Alg(Cat d∞ ) with the ∞-category of monoidal ∞-
Pr,⊗ ⊗
categories. The (lax monoidal) inclusion Cat
d
∞ ⊆ Cat
d allows us to identify Alg(LPr) with a subcategory
∞
of Alg(Cat∞ ). Unwinding the definitions, we see that a monoidal ∞-category C belongs to CAlg(LPr) if and
d
only if C is presentable, and the tensor product operation ⊗ : C × C → C preserves (small) colimits separately
in each variable.
Example 4.1.9. Let X and Y be ∞-topoi. Then X ⊗ Y is an ∞-topos, and can be identified with the
(Cartesian) product of X and Y in the ∞-category of ∞-topoi. For a proof of a slightly weaker assertion, we
refer the reader to Theorem T.[Link]. The general statement can be proved using the same argument.
Example 4.1.10. The proof of Proposition 4.1.7 shows that the ∞-category S is the unit object of LPr.
In particular, for every presentable ∞-category C we have canonical equivalences C ' C ⊗ S ' FunR (Cop , S).
The essential surjectivity of the composition is a restatement of the representability criterion of Proposition
T.[Link].
Example 4.1.11. Recall that, for every ∞-category C, the ∞-category C∗ of pointed objects of C is defined to
be the full subcategory of Fun(∆1 , C) spanned by those functors F : ∆1 → C for which F (0) is a final object
of C. The canonical isomorphism of simplicial sets FunR (Cop , D∗ ) ' FunR (Cop , D)∗ induces an equivalence
C ⊗ D∗ ' (C ⊗ D)∗ for every pair of presentable ∞-categories C, D ∈ LPr. In particular, we have a canonical
equivalence C ⊗ S∗ ' C∗ .
Example 4.1.12. Let Sp denote the ∞-category of spectra. Then Sp can be identified with a homotopy
limit of the tower
Ω Ω
. . . → S∗ → S∗ .
Consequently, for every presentable ∞-category C, we have equivalences
C ⊗ S∞ ' FunR (Cop , Sp) ' holim{FunR (Cop , S∗ )} ' holim{C∗ } ' Stab(C),
where Stab(C) denotes the stabilization of C defined in §S.8.

4.2 The Smash Product Monoidal Structure

In this section, we will apply the results of §4.1 to construct a monoidal structure on the ∞-category Sp of
spectra. Our first step is to construct a monoidal structure on the ∞-category of stable presentable monoidal
∞-categories.
σ,⊗ Pr,⊗
Notation 4.2.1. Let Catd
∞ ⊆ Cat∞
d denote the full subcategory spanned by the objects [X1 , . . . , Xn ]
where each Xi is a stable presentable ∞-category.
LPr,σ
Remark 4.2.2. Let Cat
d
∞ denote the full subcategory of LPr spanned by the stable ∞-categories. In
LPr,σ LPr,σ
view of Corollary S.15.5, the inclusion Cat
d
∞ ⊆ LPr admits a left adjoint L : LPr → Cat
d
∞ , given by
the formula
C 7→ Stab(C).
In view of Example 4.1.12, we can identify L with the functor C 7→ C ⊗ S∞ . It follows easily that L is
compatible with the monoidal structure on LPr. (We could equally well make the same argument with
“stable” replaced by “pointed”, and Example 4.1.11 in place of Example 4.1.12).

137
 σ,⊗
Proposition 4.2.3. (1) The projection Cat
d
∞ → N(∆)op determines a monoidal structure on the ∞-
LPr,σ σ,⊗
category Cat
d
∞ ' (Cat )[1] .
d
∞

σ,⊗ Pr,⊗
(2) The inclusion functor i : Cat
d
∞ ⊆ Cat∞
d is a lax monoidal functor.
LPr,σ Pr,⊗ σ,⊗
(3) The localization functor Stab : LPr → Cat
d
∞ extends to a monoidal functor from Cat
d
∞ to Cat
d .
∞
LPr,σ
(4) The unit object of Cat
d
∞ is the ∞-category Sp of spectra.
Proof. Assertions (1) through (3) follow from Proposition 1.3.9 and Remark 4.2.2. Assertion (4) follows from
(3), Example 4.1.10, and the observation that Sp ' Stab(S).
LPr,σ
Remark 4.2.4. Applying the reasoning of Remark 4.1.8, we deduce that Alg(Cat d
∞ ) can be identified
LPr,σ
with a full subcategory of Alg(Cat∞ ). A monoidal ∞-category C belongs to this full subcategory if and
d
only if C is stable, presentable, and the bifunctor ⊗ : C × C → C preserves colimits separately in each variable.
σ,⊗ Pr,⊗
Remark 4.2.5. The inclusion functor i : Cat d
∞ ⊆ Cat∞
d is almost a monoidal functor. Unwinding the
definitions, we see that i is monoidal if and only if, for every finite collection [C1 , . . . , Cn ] of presentable
stable ∞-categories, the tensor product C1 ⊗ . . . ⊗ Cn is again stable. This is true provided that n > 0; in
fact, Remark 4.1.6 implies that C1 ⊗ . . . ⊗ Cn is stable provided that Cj is stable for some j ∈ {1, . . . , n}.
However, in the degenerate case n = 0, the tensor product is equivalent to S, which is not stable. In other
words, i fails to be a monoidal functor only because it does not preserve the unit object.
Mon
Corollary 4.2.6. Let Cat
d
∞ denote the ∞-category of (not necessarily small) monoidal ∞-categories. Let
σ,Mon
Cat∞
d denote the subcategory whose objects are required to be stable presentable monoidal ∞-categories
such that the bifunctor ⊗ preserves colimits in separately in each variable, and whose morphisms are given
by colimit-preserving monoidal functors. Then:
σ,Mon
(1) The ∞-category Cat
d
∞ has an initial object C⊗ .
(2) The underlying ∞-category C⊗
[1] is equivalent to the ∞-category of spectra.

(3) Let D⊗ be an arbitrary monoidal ∞-category. Suppose that:

(i) The underlying ∞-category D = D⊗
[1] is stable and presentable.
(ii) The functor Sp → D determined by the unit object of D (see Corollary S.15.6) is an equivalence
of ∞-categories.
(iii) The bifunctor ⊗ : D × D → D preserves small colimits in each variable.
Then there exists an equivalence of monoidal ∞-categories C⊗ → D⊗ . Moreover, the collection of such
equivalences is parametrized by a contractible Kan complex.
σ,Mon LPr,σ
Proof. In view of Remark 4.2.4, we can identify Cat
d
∞ with the ∞-category of algebra objects of Cat
d
∞ .
σ,Mon ⊗
Proposition 1.4.3 implies that Cat∞
d has an initial object C . This proves (1). Moreover, Proposition
σ,⊗
1.4.3 also asserts that the underlying ∞-category C = C⊗ is equivalent to the unit object of (Cat
[1]
d )[1] ,
∞
which is the ∞-category of spectra (Proposition 4.2.3); this proves (2).
Suppose that D⊗ is a monoidal ∞-category satisfying (i), (ii), and (iii). Since C⊗ is an initial object of
σ,Mon
Cat
d
∞ , there exists a monoidal functor f : C⊗ → D⊗ , unique up to a contractible space of choices. We
claim that f is an equivalence of monoidal ∞-categories. According to Remark 1.1.13, it will suffice to show
that f induces an equivalence f[1] : C → D = D⊗ [1] of ordinary underlying categories. Corollary S.15.6 implies

138
that f[1] is determined, up to equivalence, by the image of the sphere spectrum S ∈ C. Since S is the unit
object of C, f[1] carries S to the unit object of D. Condition (ii) now implies that f[1] is an equivalence, as
desired.
It follows from Corollary 4.2.6 that the ∞-category Sp admits an essentially unique monoidal structure,
which may be characterized by the following properties:

(i) The bifunctor ⊗ : Sp × Sp → Sp preserves small colimits separately in each variable.

(ii) The unit object of Sp is the sphere spectrum S.

We will refer to this monoidal structure on Sp as the smash product monoidal structure. From the
uniqueness, we conclude that the smash product monoidal structure on Sp is equivalent to any monoidal
structure obtained via the constructions described in (S1) or (S2) above.

Proposition 4.2.7. The forgetful functor ModSp (LPr) → LPr is fully faithful, and induces an equivalence
LPr,σ
ModSp (LPr) → Cat
d
∞ .
In other words, every presentable stable ∞-category is tensored over Sp in an essentially unique way.
LPr,σ LPr,σ
Proof. Corollaries 2.4.4 and 4.2.6 imply that the forgetful functor ModSp (Cat
d
∞ ) → Cat
d is an equiv-
∞
LPr,σ
alence of ∞-categories. To complete the proof, we must show that the inclusion ModSp (Cat∞ ) ⊆d
ModSp (LPr) is also an equivalence of ∞-categories. In other words, we must show that if a presentable
∞-category C is tensored over Sp (in a colimit-preserving fashion), then C is stable.
We first show that C is pointed. Let ⊗ : Sp × C → C be the action of Sp on C. Let S ∈ Sp denote the
sphere spectrum, let 0 ∈ Sp be a zero object, and let α : S → 0 be a morphism. Let 1C ∈ C be a final object.
Then we obtain an induced map 1C ' S ⊗ 1C → 0 ⊗ 1C . Since the functor • ⊗ C preserves colimits, the tensor
product 0 ⊗ 1C is an initial object of C. We now apply Remark S.2.2 to conclude that C is pointed.
We have a pushout diagram
S /0

 
0 / S[1].

We therefore obtain an associated pushout diagram

S⊗• / 0⊗•

 
0⊗• / S[1] ⊗ •

in the ∞-category Fun(C, C). It follows that tensor product with S[1] can be identified with the suspension
functor Σ : C → C. Consequently, the suspension Σ : C → C is an equivalence: a homotopy inverse is given
by tensor product with S[−1] ∈ Sp. Corollary S.8.28 implies that C is stable, as desired.

Remark 4.2.8. Let C be a small stable ∞-category. Then the ∞-category Ind(C) is a presentable stable
∞-category, so we may regard Ind(C) as tensored over Sp in an essentially unique way such that the tensor
product functor
⊗ : Sp × Ind(C) → Ind(C)
preserves colimits separately in each variable. Let C0 ⊆ Ind(C) denote the essential image of the Yoneda
embedding j : C → Ind(C), so that j induces an equivalence of C with C0 . Let Sfin
∞ denote the full subcategory

139
of Sp spanned by the finite spectra (see §S.9). We observe that Sfin
∞ is generated under finite colimits by the
collection of n-spheres {S[n]}n∈Z . Moreover, the operation

S[n] ⊗ • : Ind(C) → Ind(C)

can be identified with the shift functor C 7→ C[n]. Since C0 is a stable subcategory of Ind(C) (Proposition
S.4.5), we conclude that the tensor product ⊗ induces a functor
0 0
Sfin
∞ ⊗C → C .

We observe that the full subcategory Sfin

∞ ⊆ Sp is closed under tensor products and contains the unit object
S ∈ Sp, and therefore inherits a monoidal structure (Proposition 1.3.1). Using the same argument, one
can show that C0 inherits the structure of an ∞-category tensored over Sfin 0
∞ . Identifying C with C via the
Yoneda embedding j, we conclude that every (small) stable ∞-category C can be regarded as tensored over
the ∞-category of finite spectra, in such a way that the tensor product

⊗ : Sfin
∞ ×C → C

preserves finite colimits in each variable. With a bit more effort, one can prove an analogue of Proposition
4.2.7 in this context.
Remark 4.2.9. Proposition 4.2.7 and Remark 4.2.8 have analogues in the setting of pointed ∞-categories:
(1) The ∞-category S∗ of pointed spaces admits an (essentially unique) monoidal structure such that the
tensor product ⊗ : S∗ × S∗ → S∗ preserves colimits separately in each variable.
(2) The forgetful functor ModS∗ (LPr) → LPr is fully faithful, and its essential image consists of pointed
presentable ∞-categories.
(3) Let C be a pointed ∞-category which admits finite colimits, and let Sfin
∗ ⊆ S∗ denote the full subcategory
∗ is a monoidal subcategory of S∗ , and
of S∗ spanned by the finite pointed spaces (see §S.9). Then Sfin
C admits the structure of an ∞-category tensored over Sfin
∗ in such a way that the tensor product

⊗ : Sfin
∗ ×C → C

preserves finite colimits in each variable.

We conclude with the following counterpart to Corollary 4.2.6:
Proposition 4.2.10. Let C⊗ → N(∆)op be a monoidal ∞-category which satisfies conditions (i) through
(iii) of Corollary 4.2.6 (so that C⊗ is equivalent, as a monoidal ∞-category, to the smash product monoidal
structure on Sp). Let q : M⊗ → C⊗ be an ∞-category left-tensored over C⊗ . Then q is equivalent (as an
object of Cat
\ Mod
∞ ) to the canonical action of C⊗ on itself, if and only if the following conditions are satisfied:
(1) The underlying ∞-category M = M⊗
[0] is equivalent to the ∞-category Sp.

(2) The tensor product functor ⊗ : C × M → M is colimit-preserving in each variable.

Proof. The necessity of (1) and (2) is clear. For the converse, we invoke the equivalence Cat
\ Mod
∞ ' Mod(Cat
d∞ )
⊗ ⊗
of Corollary 2.6.6. In virtue of assumption (2), we can identify both q : M → C and the canonical action
of C⊗ on itself with objects of ModC⊗ (LPr). Combining Corollary 4.2.6 with Proposition 4.2.7, we deduce
LPr,σ
that the forgetful functor ModC⊗ (LPr) → Cat d
∞ is an equivalence. It therefore suffices to prove that the
underlying ∞-categories C and M are equivalent, which follows from (1).
Warning 4.2.11. Given a pair of objects X, Y ∈ hSp, the smash product of X and Y is usually denoted
by X ∧ Y . We will depart from this convention by writing instead X ⊗ Y for the smash product.

140
4.3 Associative Ring Spectra
In this section, we will introduce the theory of A∞ -ring spectra, or, as we will call them, A∞ -rings. Roughly
speaking, an A∞ -ring is to an ordinary associative ring as a spectrum is to an abelian group, or as a homotopy
type is to a set. In particular, if we restrict our attention to discrete A∞ -rings, then we recover classical ring
theory (Proposition 4.3.13).
A large portion of classical (noncommutative) algebra can be generalized to the setting of A∞ -rings; we
will see some examples in the next few sections. For the time being, we will concern ourselves only with the
definition and basic formal properties.
Definition 4.3.1. An A∞ -ring is an algebra object of the ∞-category of Sp (endowed with its smash product
monoidal structure). We let A∞ denote the ∞-category Alg(Sp) of A∞ -rings.
Let R be an A∞ -ring. We will generally not distinguish notationally between R and its underlying
spectrum. In particular, for each n ∈ Z, we let πn R denote the nth homotopy group of the underlying
spectrum. We observe that πn R can be identified with the set π0 MapSp (S[n], R), where S denotes the
sphere spectrum. Since S is the identity for the smash product, there is a canonical equivalence S ⊗ S ' S;
using the fact that ⊗ is exact in each variable, we deduce the existence of equivalences S[n]⊗S[m] ' S[n+m]
for all n, m ∈ Z. The composition
MapSp (S[n], R) × MapSp (S[m], R) → MapSp (S[n] ⊗ S[m], R ⊗ R) → MapSp (S[n + m], R)

L of abelian groups πn R ⊗ πm R → πn+m R. It is not difficult to see that these maps

determines a map
endow π∗ R = n πn R with the structure of a graded associative ring, which depends functorially on R. In
particular, π0 R is an ordinary associative ring, and each πn R has the structure of a π0 R-bimodule.
Let Ω∞ : Sp → S be the 0th space functor. If R is an A∞ -ring, we will refer to X = Ω∞ R as the
underlying space of R. The underlying space X is equipped with an addition X × X → X (determined
by the fact that it is the 0th space of a spectrum) and a multiplication X × X → X (determined by the
map R ⊗ R → R); these maps endow X with the structure of a ring object in the homotopy category H of
spaces. However, the structure of an A∞ -ring is much richer: not only do the ring axioms on X hold up to
homotopy, they hold up to coherent homotopy.
The functor Ω∞ : A∞ → S is not conservative: a map of A∞ -rings f : A → B which induces a homotopy
equivalence of underlying spaces need not be an equivalence in A∞ . We observe that f is an equivalence
of A∞ -rings if and only if it is an equivalence of spectra; that is, if and only if πn (f ) : πn A → πn B is an
isomorphism of abelian groups for all n ∈ Z. However, Ω∞ (f ) is a homotopy equivalence of spaces provided
only that πn (f ) is an isomorphism for n ≥ 0; this is generally a weaker condition.
Example 4.3.2. Let C be a stable ∞-category, and let X ∈ C be an object. Then it is possible to extract
from C an A∞ -ring spectrum EndC (X) with the property that πn EndC (X) ' Ext−n C (X, X) for all n ∈ Z,
and the ring structure on π∗ EndC (X) is given by composition in the triangulated category hC. We will
describe the argument in the case where C is presentable. According to Proposition 4.2.7, the ∞-category
C is naturally left-tensored over Sp. Proposition 2.1.12 implies that C is also enriched over Sp, so that
there exists a morphism object MorC (X, X). Proposition 2.7.3 implies that EndC (X) = MorC (X, X) can be
lifted to a final object of the monoidal ∞-category Sp[X], and therefore inherits the structure of an algebra
object. The identification of the homotopy groups of EndC (X) follows from the homotopy equivalence
MapSp (S[n], EndC (X)) ' MapC (S[n] ⊗ X, X).
Remark 4.3.3. Combining the uniqueness assertion of Corollary 4.2.6 with Theorem 1.6.17, we conclude
that A∞ is equivalent to the underlying ∞-category of strictly associative monoids in any sufficiently nice
monoidal model category of spectra (see Example 1.6.19). With minor modifications, the same argument
can be applied to the model of spectra described in [17] (though Theorem 1.6.17 does not quite apply in its
present form).
Remark 4.3.4. Combining the uniqueness assertion of Corollary 4.2.6 with the construction (S2) of §4, we
conclude that A∞ can be identified with the ∞-category of colimit-preserving monads on Sp.

141
Lemma 4.3.5. The t-structure on the ∞-category Sp determined by the class of connective objects is com-
patible with the smash product monoidal structure (in the sense of Definition 1.3.10). In other words, the
full subcategory Spconn ⊆ Sp spanned by the connective objects is closed under smash products and contains
the unit object. Consequently, the monoidal structure on Sp determines a monoidal structure on Spconn .
Proof. It follows from the results of §S.16 that Spconn is the smallest full subcategory of Sp which contains
the sphere spectrum S ∈ Sp and is stable under colimits and extensions. Let C be the full subcategory of
Sp spanned by those spectra X such that, for all Y ∈ Spconn , X ⊗ Y is connective. We wish to prove that
Spconn ⊆ C. Since the smash product preserves colimits separately in each variable, we conclude that C is
closed under colimits and extensions in Sp. It will therefore suffice to prove that S ∈ C. This is clear, since
S is the unit object of Sp.
We will say that an A∞ -ring R is connective if its underlying spectrum is connective. We let Aconn ∞
denote the full subcategory of A∞ spanned by the connective objects. Equivalently, we may view Aconn ∞ as
the ∞-category Alg(Spconn ) of algebra objects in connective spectra.
When restricted to connective A∞ -rings, the functor Ω∞ detects equivalences: if f : A → B is a
morphism in Aconn
∞ such that Ω∞ (f ) is an equivalence, then f is an equivalence. We observe that the functor
Ω : A∞ → S is a composition of a pair of functors Alg(Spconn ) → Spconn → S, both of which preserve
∞ conn

geometric realizations (Corollary 1.5.11 and S.9.11) and admit left adjoints. It follows from Theorem 3.5.1
that Aconn
∞ can be identified with the ∞-category of modules over a suitable monad on S. In other words, we
can view connective A∞ -rings as spaces equipped with some additional structures. Roughly speaking, these
additional structures consist of an addition and multiplication which satisfy the ring axioms, up to coherent
homotopy.
Definition 4.3.6. Let R be an A∞ -ring. A connective cover of R is a morphism φ : R0 → R of A∞ -rings
with the following properties:
(1) The A∞ -ring R0 is connective.
(2) For every connective A∞ -ring R00 , composition with φ induces a homotopy equivalence

MapA∞ (R00 , R0 ) → MapA∞ (R00 , R).

Remark 4.3.7. In the situation of Definition 4.3.6, we will generally abuse terminology and simply refer to
R0 as a connective cover of R, in the case where the map φ is implicitly understood.
Proposition 4.3.8. (1) Every A∞ -ring R admits a connective cover.
(2) An arbitrary map φ : R0 → R of A∞ -rings is a connective cover of R if and only if R0 is connective,
and the induced map πn R0 → πn R is an isomorphism for n ≥ 0.
(3) The inclusion Aconn
∞ ⊆ A∞ admits a right adjoint G, which carries each A∞ -ring R to a connective
cover R0 of R.
Proof. Combine Lemma 4.3.5, Proposition 1.3.1 and Remark 1.3.3.
Recall that an object X of an ∞-category C is said to be n-truncated if the mapping spaces MapC (Y, X)
are n-truncated, for every Y ∈ C (see §T.5.5.6). Proposition 1.5.15 implies that the ∞-categories A∞ and
Aconn
∞ are both presentable, so that we have a good theory of truncation functors.
Proposition 4.3.9. Let R be a connective A∞ -ring and let n be a nonnegative integer. The following
conditions are equivalent:
(1) As an object of Aconn
∞ , R is n-truncated.

(2) As an object of Spconn , R is n-truncated.

142
 (3) The space Ω∞ (R) is n-truncated.

(4) For all m > n, the homotopy group πm R is trivial.

Proof. The equivalence (3) ⇔ (4) is easy (Remark T.[Link]), and the equivalence (2) ⇔ (3) was explained
in Warning S.6.9. The implication (1) ⇒ (2) follows from Proposition T.[Link], since the forgetful functor
Aconn
∞ → Spconn preserves small limits (Corollary 1.5.3).
We now prove that (2) ⇒ (1). Assume that R is n-truncated as a spectrum. Let T : (Aconn ∞ )
op
→ S
be the functor represented by R. Let C ⊆ A∞ be the full subcategory of A∞ spanned by those objects
conn conn

B such that T (B) is n-truncated. We wish to prove that C = Aconn ∞ . Since T preserves limits (Proposition
T.[Link]) and the class of n-truncated spaces is stable under limits (Proposition T.[Link]), we conclude that
C is stable under small colimits in Aconn
∞ . Let F be a left adjoint to the forgetful functor A∞
conn
→ Spconn
(Theorem 1.4.2). Proposition 3.5.5 implies that C is generated under colimits by the essential image of F .
Consequently, it will suffice to show that F (M ) ∈ C, for every M ∈ Spconn . Equivalently, we must show that
the space MapAconn
∞
(F (M ), R) ' MapSpconn (M, R) is n-truncated, which follows from (2).
Remark 4.3.10. An A∞ -ring R is n-truncated as an object of A∞ if and only if it is equivalent to zero
(since the ∞-category Sp has no nontrivial n-truncated objects).
Alg
Let τ≤n : Spconn → Spconn be the truncation functor on connective spectra, and let τ≤n : Aconn
∞ → Aconn
∞
conn conn
be the truncation functor on connective A∞ -rings. Since the forgetful functor θ : A∞ → Sp preserves
Alg
n-truncated objects, there is a canonical natural transformation α : τ≤n ◦ θ → θ ◦ τ≤n . Our next goal is to
show that α is an equivalence.
Proposition 4.3.11. Let Spconn be the ∞-category of connective spectra, endowed with the smash product
monoidal structure, and let (Spconn )≤n be the ∞-category of n-truncated objects of Spconn . Then:
(1) The localization functor τ≤n : Spconn → Spconn is compatible with the smash product monoidal structure,
in the sense of Definition 1.3.4.
(2) The smash product monoidal structure on Sp≥0 induces a monoidal structure on (Spconn )≤n and an
Alg conn
identification Alg((Spconn )≤n ) ' τ≤n A∞ .

Proof. Assertion (1) is a special case of Proposition 1.3.12. Assertion (2) follows from (1) and Proposition
4.3.9.
In other words, if R is a connective A∞ -ring, then the ring structure on R determines a ring structure
on τ≤n R for all n ≥ 0.

Definition 4.3.12. We will say that an A∞ -ring is discrete if it is connective and 0-truncated. We let Adisc
∞
denote the full subcategory of A∞ spanned by the discrete objects.
Since the mapping spaces in Adisc disc
∞ are 0-truncated, it follows that A∞ is equivalent to the nerve of an
ordinary category. We conclude this section by identifying the relevant category.
Proposition 4.3.13. The functor R 7→ π0 R determines an equivalence from Adisc
∞ to the (nerve of the)
category of associative rings.
Proof. According to Proposition 4.3.11, we can identify Adisc
∞ with the ∞-category of algebra objects of the
heart S♥
∞ , which inherits a monoidal structure from Sp in view of Proposition 1.3.12 and Lemma 4.3.5.
Proposition S.9.2 allows us to identify S♥
∞ with (the nerve of) the category of abelian groups. Moreover,
the induced monoidal structure on S♥ ∞ has π0 S ' Z as unit object, and the tensor product functor ⊗
preserves colimits separately in each variable. It follows that this monoidal structure coincides (up to
canonical equivalence) with the usual monoidal structure on S♥∞ , given by tensor products of abelian groups.
Consequently, we may identify Alg(S♥ ∞ ) with the (nerve of the) category of associative rings.

143
4.4 Modules over A∞ -Rings
Let R be an A∞ -ring. In this section, we will describe the associated theory of R-module spectra, or simply R-
modules. This can be regarded as a generalization of homological algebra: if R is an ordinary ring (regarded
as a discrete A∞ -ring via Proposition 4.3.13), then the homotopy category of R-module spectra coincides
with the classical derived category of R (Proposition 4.4.7); in particular, the theory of R-module spectra is
a generalization of the usual theory of R-modules.
For any A∞ -ring R, the ∞-category ModR of R-modules is stable (Proposition 4.4.3), so its homotopy
category hModR is triangulated. According to a result of Schwede and Shipley (Theorem 4.4.9), a large
variety of stable ∞-categories have the form ModR , for appropriately chosen R. In particular, hModR is
generally not equivalent to the derived category of an abelian category.
Definition 4.4.1. Let R be an A∞ -ring. We let ModR denote the ∞-category ModR (Sp); we will refer to
ModR as the ∞-category of (left) R-modules.
We will generally not distinguish between an R-module M and the underlying spectrum. In particular,
the homotopy groups πn M are defined to be the homotopy groups of the underlying spectrum. The action
map R ⊗ M → M induces bilinear maps πn R × πm M → πn+m M , which endow π∗ M with the structure of
a graded left module over π∗ R. We will say that M is connective if its underlying spectrum is connective;
that is, if πn M ' 0 for n < 0.
Remark 4.4.2. Roughly speaking, if we think of R as a space equipped with the structure of an associative
ring up to coherent homotopy, then an R-module can be thought of as another space which has an addition
and an action of R, up to coherent homotopy in the same sense. This intuition is really only appropriate
in the case where R and M are connective, since the homotopy groups in negative degree have no ready
interpretation in terms of underlying spaces.
Our first goal is to prove that the ∞-category of modules over an A∞ -ring is stable. This is a consequence
of the following more general assertion:
Proposition 4.4.3. Let C be an ∞-category equipped with a monoidal structure and a left action on an ∞-
category M. Assume that M is a stable ∞-category, and let R ∈ Alg(C) be such that the functor M 7→ R ⊗ M
is exact. Then ModR (M) is a stable ∞-category. Moreover, if N is an arbitrary stable ∞-category, then
a functor N → ModR (M) is exact if and only if the composite functor N → ModR (M) → M is exact. In
particular, the forgetful functor ModR (M) → M is exact.
Proof. This is a special case of Proposition 2.3.6.
If R is a connective A∞ -ring, the homotopy groups of an R-module M can be interpreted in terms of an
appropriate t-structure on ModR .
Notation 4.4.4. If R is an A∞ -ring, we let Mod≥0
R be the full subcategory of ModR spanned by those
R-modules M for which πn M ' 0 for n < 0, and Mod≤0
R the full subcategory of ModR spanned by those
R-modules M for which πn M ' 0 for n > 0.
Notation 4.4.5. Let R be an A∞ -ring, and let M and N be left R-modules. We let ExtiR (M, N ) denote
the abelian group π0 MapModR (M, N [i]).
Proposition 4.4.6. Let R be a connective A∞ -ring. Then:
(1) The full subcategory Mod≥0
R ⊆ ModR is the smallest full subcategory which contains R (regarded as an
R-module in the natural way; see Example 2.1.7) and is stable under small colimits.

(2) The subcategories Mod≥0 ≤0

R , ModR determine an accessible t-structure on ModR (see §S.16).

(3) The t-structure described in (2) is both left and right complete, and the functor π0 determines an
equivalence of the heart Mod♥
R with the (nerve of the) ordinary category of π0 R-modules.

144
 (4) The subcategories Mod≥0 ≤0
R , ModR ⊆ ModR are stable under small products and small filtered colimits.

Proof. According to Proposition S.16.1, there exists an accessible t-structure (Mod0R , Mod00R ) with the fol-
lowing properties:
(a) An object M ∈ ModR belongs to Mod00R if and only if ExtiR (R, M ) ' 0 for i < 0.
(b) The ∞-category Mod0R is the smallest full subcategory of ModR which contains the object R and is
stable under extensions and small colimits.
Corollary 4.5.14 implies that R (regarded as an object of ModR ) corepresents the composition ModR →
Ω∞ ≤0
Sp → S . It follows that Mod00R = ModR . Because the forgetful functor ModR → Sp preserves small
colimits (Corollary 2.3.7), we conclude that Mod≥0R is stable under extensions and small colimits. Since R
≥0 0 ≥0
is connective, R ∈ ModR , so that ModR ⊆ ModR . Let C be the smallest full subcategory of ModR which
contains R and is stable under small colimits, so that C ⊆ Mod0R . We will complete the proof of (1) and (2)
by showing that C = Mod≥0 R .
Let M ∈ Mod≥0 R . We will construct a diagram

M (0) → M (1) → M (2) → . . .

in (ModR )/M with the following properties:

(i) Let i ≥ 0, and let K(i) be a kernel of the map M (i) → M . Then πj K(i) ' 0 for j < i.

(ii) The R-module M (0) is a coproduct of copies of R.

(iii) For i ≥ 0, there is a pushout diagram

F [i] /0

 
M (i) / M (i + 1),

where F is a coproduct of copies of R.

We begin by choosing M (0) to be any coproduct of copies of R equipped with a map M (0) → M which
induces a surjection π0 M (0) → π0 M ; for example, we can take M (0) to be a coproduct of copies of R indexed
by π0 M . Let us now suppose that the map f : M (i) → M has been constructed, with K(i) = ker(f ) such
that πj K(i) ' 0 for j < i. We now choose F to be a coproduct of copies of R and a map g : F [i] → K(i)
which induces a surjection π0 F → πi K(i). Let h denote the composite map F [i] → K(i) → M (i), and let
M (i + 1) = coker(h). The canonical nullhomotopy of K(i) → M (i) → M induces a factorization

f0
M (i) → M (i + 1) → M

of f . We observe that there is a canonical equivalence ker(f 0 ) ' coker(g), so that πj ker(f 0 ) ' 0 for j ≤ i.
Let M (∞) be the colimit of the sequence {M (i)}, and let K be the kernel of the canonical map M (∞) →
M . Then K can be identified with a colimit of the sequence {K(i)}i≥0 . Since the formation of homotopy
groups is preserves filtered colimits, we conclude that πj K ' colim πj K(i) ' 0. Thus M (∞) ' M , so that
M ∈ C as desired.
Assertion (4) follows from the corresponding result for Sp, since the forgetful functor ModR → Sp
preserves all limits and colimits (Corollaries 2.3.5 and 2.3.7). Since ModR → Sp is a conservative functor,
an R-module M is zero if and only if πn M is zero for all n ∈ Z. It follows from Proposition S.7.3 that ModR
is both right and left complete.

145
 Let F be the functor from Mod≥0 R to the (nerve of the) ordinary category of left π0 R-modules, given by
M 7→ π0 M . It is easy to see that F preserves colimits, and that the restriction of F to Mod♥ R is an exact
functor. We wish to prove that F0 = F | Mod♥ R is an equivalence. We first show that the restriction of F0
≥0
is fully faithful. Fix N ∈ Mod♥ R , and let D be the full subcategory of ModR spanned by those objects
M for which the map π0 MapModR (M, N ) → Hom(F (τ≤0 M ), F (N )) is bijective, where the right hand side
indicates the group of π0 R-module homomorphims. It is easy to see that D is stable under colimits and
contains R. The first part of the proof shows that D = Mod≥0 R . In particular, F0 is fully faithful.
It remains to show that F0 is essentially surjective. Since F0 is fully faithful and exact, the essential
image of F0 is closed under the formation of cokernels. It will therefore suffice to show that every free left
π0 R-module belongs to the essential image of F0 . Since F0 preserves coproducts, it will suffice to show that
π0 R itself belongs to the essential image of F0 . We now conclude by observing that F0 (τ≤0 R) ' π0 R.
Let R be a connective A∞ -ring, let A be the abelian category of modules over the (ordinary) ring π0 R.
Then A has enough projective objects, so we can consider the derived ∞-category D− (A) as described in
§S.13. Part (3) of Proposition 4.4.6 determines an equivalence N(A) ' Mod♥ R . Applying Corollary S.14.12,
we deduce the existence of an (essentially unique) right t-exact functor θ : D− (A) → ModR .
Proposition 4.4.7. Let R be a connective A∞ -ring, and let θ : D− (A) → ModR be as described above. The
following conditions are equivalent:
(1) The A∞ -ring R is discrete.
(2) The functor θ is fully faithful, and induces an equivalence of D− (A) with the ∞-category of right
bounded objects of ModR .
Proof. Let P ∈ A be projective object corresponding to the free R-module on one generator. Then, for
M ∈ D− (A), we have a canonical isomorphism Ext0D− (A) (P, M ) ' π0 M. If (2) is satisfied, then we deduce
the existence of a canonical isomorphisms

Ext0R (θ(P ), M ) ' π0 M ' Ext0R (R, M )

for M ∈ Mod≥0 R . Thus θ(P ) and R are isomorphic in the homotopy category hModR . Since θ(P ) is discrete,
we conclude that R is discrete, which proves (1).
For the converse, let us suppose that R is discrete. Let us regard (the nerve of) A as a full subcategory
of both D− (A) and ModR . For M, N ∈ A, let ExtiA (M, N ) denote the abelian group π0 MapD− (A) (M, N )
(in other words, ExtiA (M, N ) is the classical Yoneda Ext-group computed in the abelian category A). We
claim that canonical map ExtiA (M, N ) → ExtiR (M, N ) is an isomorphism. For i < 0, both sides vanish. The
proof in general goes by induction on i, the case i = 0 being trivial. For i > 0, we choose an exact sequence

0→K→P →M →0

in A, where P is a free π0 R-module. We have a commutative diagram of abelian groups with exact rows

Exti−1 / Exti−1 (K, N ) / Exti (M, N ) / Exti (P, N )

A (P, N ) A A A

ψ1 ψ2 ψ3
   
Exti−1 / Exti−1 (K, N ) / Exti (M, N ) / Exti (P, N ).
R (P, N ) R R R

We wish to show that ψ3 is an isomorphism. Since ψ1 and ψ2 are bijective by the inductive hypothesis,
it will suffice to show that ExtiA (P, N ) ' 0 ' ExtiR (P, N ). The first equivalence follows from the fact that
P is a projective object of A. For the second, we observe that as an object of ModR , P coincides with a
coproduct of copies of R (in virtue of assumption (1)). Consequently, ExtiR (P, N ) can be identified with a
product of copies of π−i N , which vanishes since i > 0 and N ∈ Mod≥0 R .

146
 Now suppose that M ∈ A, and consider the full subcategory C ⊆ D− (A) spanned by those objects N for
which the canonical map ExtiD− (A) (M, N ) → ExtiR (θ(M ), θ(N )) is an isomorphism for all i ∈ Z. Applying
the five lemma to the relevant long exact sequences, we conclude that C is stable under extensions in D− (A).
The above argument shows that C contains the heart of D− (A); it therefore contains the full subcategory
Db (A) of bounded object of D− (A).
Now let C0 ⊆ D− (A) spanned by those objects M having the property that for every N ∈ Db (A), the
canonical map ExtiD− (A) (M, N ) → ExtiR (θ(M ), θ(N )) is an isomorphism for i ∈ Z. Repeating the above
argument, we conclude that Db (A) ⊆ C0 . In particular, the restriction θ| Db (A) is fully faithful.
We claim that the essential image of θ| Db (A) consists of precisely the bounded objects of ModR . Let
M ∈ ModR be a bounded object. We wish to prove that M belongs to the essential image of θ. Without
loss of generality, we may suppose that M ∈ Mod≥0 R . Since M is bounded, we have also M ∈ ModR for
≤n

some n ≥ 0. We now work by induction on n. If n = 0, then M belongs to the heart of ModR and the result
is obvious. If n > 0, then we have a distinguished triangle

τ≥n M → M → τ≤n−1 M → τ≥n M [1].

Since θ is exact and fully faithful, it will suffice to show that τ≥n M [−n] and τ≤n−1 M belong to the essential
image of θ, which follows from the inductive hypothesis.
The preceding argument shows that θ induces an equivalence Db (A) → ModbR between the full subcate-
gories of bounded objects. We now conclude by observing that both D− (A) and ModR are left-complete.
Remark 4.4.8. Let R and A be as in Proposition 4.4.7. The abelian category A also has enough injective
objects. The proof of Proposition 4.4.7 can be repeated, without essential change, to obtain an identification
of D+ (A) with the ∞-category of left bounded objects of ModR . In other words, we may view ModR as a
candidate for an unbounded derived ∞-category of R-modules. Using a completeness argument, it is not
difficult to show that this coincides with the usual unbounded derived category of R-modules; we will not
pursue this point further.
We conclude this section by addressing the following question: given an ∞-category C, under what
conditions does there exist an A∞ -ring R and an equivalence C → ModR ? Of course, there might be several
candidates for R. For example, if R is a discrete ring, then the module categories (in the ordinary or derived
sense) of R and M2 (R) are equivalent, where M2 (R) denotes the ring of 2 × 2 matrices with coefficients in
R. To eliminate this ambiguity, we should specify an object C ∈ C which is the hypothetical preimage of
R ∈ ModR under the functor θ. In this case, we have the following theorem of Schwede and Shipley:
Theorem 4.4.9. [Schwede-Shipley [61]] Let C be a stable ∞-category. Then C is equivalent to ModR , for
some A∞ -ring R, if and only if there exists an object C ∈ C satisfying the following conditions:

(1) The ∞-category C is presentable.

(2) The object C ∈ C is compact.
(3) The object C generates C in the following sense: if D ∈ C has the property that ExtnC (C, D) for all
n ∈ Z, then D ' 0.

Proof. Suppose first that C ' ModR . Then (1) follows from Corollary 2.3.8. To prove (2) and (3), we take
C to be R itself, regarded as an R-module as explained in Example 2.1.7. Corollary 4.5.14 implies that R
Ω∞
corepresents the composite functor ModR → Sp → S, which preserves filtered colimits in virtue of Corollary
2.3.7. This proves (2). If ExtiModR (R, D) ' π−i D vanishes for all i ∈ Z, so that D ' 0. This proves (3).
Conversely, suppose that C satisfies conditions (1), (2), and (3) for an appropriately chosen object C ∈ C.
Applying Corollary S.15.6, we deduce that there is a colimit-preserving functor F : Sp → C, determined
up to equivalence by the requirement that F (S) ' C, where S denotes the sphere spectrum. Applying the
adjoint functor theorem (Corollary T.[Link]), we deduce that F admits a right adjoint G. We now apply

147
Theoream 3.2.10 to extend the functor F to a strong adjunction diagram s ∈ ADat(Sp, C). By restrction, s
determines a monad T ∈ Alg(End(Sp)).
Let End0 (Sp) be the full subcategory of End(Sp) spanned by those functors which preserve colimits.
Then End0 (Sp) is closed under composition. We let End⊗ 0 (Sp) denote the corresponding full (monoidal)
subcategory of End⊗ (Sp), and let
⊗ ⊗
End0 (Sp) = End (Sp) ×End⊗ (Sp) End⊗
0 (Sp).

G Ω∞−n
The composition gn : C → Sp ∗→ S∗ → S can be identified with the composition of a shift functor from C
to itself (an equivalence of ∞-categories) and the functor corepresented by C ∈ C. Since C is assumed to be
compact, we conclude that gn preserves filtered colimits. Since the forgetful functor S∗ ' S∗/ → S detects
filtered colimits, we conclude that Ω∞−n ◦ G preserves filtered colimits. Since Sp is defined as the homotopy
inverse limit of a tower
Ω Ω
. . . → S∗ → S∗
of continuous functors, we conclude that G itself preserves filtered colimits. Since G is exact, it preserves all
colimits. It follows that T can be identified with an algebra object of End0 (Sp).
Invoking Remark 3.3.4 and Corollary 3.3.6, we obtain a functor θ : C → ModT (Sp). Assumption (3)
implies that G is conservative. Since G preserves geometric realizations, Corollary 3.5.4 implies that θ is an
equivalence. Proposition 4.2.10 implies that the tensored ∞-category
⊗
End0 (Sp) → End⊗
0 (Sp) → N(∆)
op

is equivalent (in CatMod

∞ ) to the standard left action of Sp on itself, where Sp is endowed with the smash
product monoidal structure. It follows that we can identify T with an A∞ -ring R, and that there is an
equivalence ModT (Sp) ' ModR (Sp). Composing this equivalene with θ, we obtain the desired result.
Remark 4.4.10. Let C be a presentable stable ∞-category containing an object C. The A∞ -ring R ap-
pearing in the proof of Theorem 4.4.9 can be identified with the A∞ -ring EndC (C) constructed in Example
4.3.2.

4.5 Balanced Pairings and the Bar Construction

Let A be an associative ring. If M is a right A-module and N is a left R-module spectrum, then we can
define an abelian group M ⊗A N . The functor (M, N ) 7→ M ⊗A N preserves colimits in separately in each
variable. Moreover, if F is a colimit-preserving functor F from left (right) A-modules to abelian groups, then
F (A) inherits a right (left) A-module structure, and there is a canonical isomorphism F (N ) ' F (A) ⊗A N
(F (M ) ' M ⊗A F (A)). Our goal in this section is to obtain an analogous picture, where we allow A to be
an A∞ -ring and M and N to be module spectra over A.
We will begin with a much more general situation: a monoidal ∞-category C, an ∞-category M which
is right-tensored over C, an ∞-category N which is left-tensored over C, and a pairing M × N → D which is
balanced in a suitable sense. Given this data, we will define a relative tensor product functor
ModA (M) × ModA (N) → D
for every algebra object A ∈ Alg(C) (see Definition 4.5.1). We will proceed to establish some of the basic
properties of the relative tensor product; In particular, we will prove a “base change” or “push-pull” formula,
which asserts that if f : A → B is a map of algebras, M is a (right) B-module and N is a (left) A-module,
then there is a canonical equivalence M ⊗A N ' M ⊗B (B ⊗A N ), where B ⊗A N is the B-module induced
from N (see Proposition 4.5.13 for a more precise formulation).
The principal example of interest to us are when M or N (or both) coincide with the original monoidal
∞-category C. In this case, there are canonical examples of balanced pairings
M × C → M, C × N → N;

148
see Definition 4.5.5 (in §4.6, we will specialize further to the case where C is the ∞-category of spectra, with
the smash product monoidal structure).
To begin, let us suppose that C is a monoidal ∞-category, and that A is an algebra object of C. We will
let ModLA (C) denote the ∞-category of left A-modules in C, as defined in §2.1. As explained in Remark 2.1.5,
we have an entirely dual theory of right modules; we let ModR A (C) denote the ∞-category of right modules
over A. We wish to define a functor ⊗A : ModR A (C) × Mod L
A (C) → C.
Fix objects M ∈ ModR A (C), N ∈ Mod L
A (C). We will abuse notation by identifying M and N with the
underlying object of the ∞-category C. Using the monoidal structure on C, we can form the tensor product
M ⊗ N . This should be regarded as a “first approximation” to M ⊗A N . We will have a natural map
α : M ⊗ N → M ⊗A N . We do not expect α to be an equivalence; rather, M ⊗A N should be obtained from
M ⊗ N by forcing the two natural actions of A to coincide. To be more precise, consider the diagram

// M ⊗ N .
f
M ⊗A⊗N g

where f is given by the right action of A on M , and g is given by the left action of A on N . We should
expect α ◦ f ' α ◦ g. In other words, we expect that α should factor through the coequalizer of the maps f
and g. In classical category theory, we can define M ⊗A N to be this coequalizer. However, in the setting of
higher category theory, we are not yet done. Namely, we should not expect an equality α ◦ f = α ◦ g; rather,
we expect a homotopy h : α ◦ f → α ◦ g. Consider now the diagram

f0
M ⊗A⊗A⊗N // M ⊗ N .
g0

where, as before, f 0 is obtained using the action of A on M (twice) and g 0 is obtained using the action of A
on N (twice). The homotopy h can be used to obtain two different homotopies between α ◦ f 0 and α ◦ g 0 :
first by applying h to each copy of A individually, and second by applying h together with the observation
that f 0 and g 0 factor through the map M ⊗ A ⊗ A ⊗ N → M ⊗ A → N determined by the algebra structure
on A. In order to obtain the correct relative tensor product M ⊗A N , we need to guarantee that these two
homotopies coincide, up to a higher homotopy. Of course, this leads to still higher obstructions which need
to be taken under consideration.
To efficiently organize all of the relevant data, it is useful to consider a simplicial object BarA (M, N )• ,
which may be informally described as follows:
(1) For each n ≥ 0, the object BarA (M, N )n ∈ C is given by the tensor product M ⊗ A ⊗ . . . ⊗ A ⊗ N ,
where n factors of A appear.
(2) If i = 0, the face map di : BarA (M, N )n → BarA (M, N )n−1 is given by the right action of A on M . If
i = n, di is given by the left action of A on N . If 0 < i < n, then di is given by the algebra structure
on A, applied to the i and (i + 1)st factors.
(3) For 0 ≤ i ≤ n, the degeneracy map si : BarA (M, N )n → BarA (M, N )n+1 is given by the composition

M ⊗ A ⊗ ... ⊗ A ⊗ N ' M ⊗ A ⊗ ... ⊗ 1 ⊗ ... ⊗ A ⊗ N → M ⊗ A ⊗ ... ⊗ A ⊗ ... ⊗ A ⊗ N

where the second map is given by the unit map of the (i + 1)st factor of A.
We can then define M ⊗A N to be the geometric realization of the simplicial object BarA (M, N )• . The
functor (M, N ) 7→ | BarA (M, N )• | is called the (two-sided) bar construction. We now recast the preceding
discussion in more formal terms.
Definition 4.5.1. Let p : C⊗ → N(∆)op be a monoidal ∞-category. Let q : M⊗ → C⊗ be an ∞-category
which is right-tensored over C⊗ , and let q 0 : N⊗ → C⊗ be an ∞-category which is left-tensored over C. A
balanced pairing is a functor F : M⊗ ×C⊗ N⊗ → D with the following property: let (α, β) be a morphism in

149
M⊗ ×C⊗ N⊗ such that α is a (p ◦ q)-coCartesian morphism in M⊗ and β is a (p ◦ q 0 )-coCartesian morphism
in N⊗ . Then F (α, β) is an equivalence in D.
Under these conditions, we let Bar• denote the composition
F◦
ModR (M) ×Alg(C) ModL (N) ⊆ Fun(N(∆)op , M⊗ ×C⊗ N⊗ ) → Fun(N(∆)op , D).

We will refer to Bar• as the two-sided bar construction. If A ∈ Alg(C), M ∈ ModR L

A (M) and N ∈ ModA (N),
we will denote the image of (M, N ) under Bar• by BarA (M, N )• .
Suppose now that D admits geometric realizations for simplicial objects. We then define the relative
tensor product functor to be the composition
Bar ||
ModR (M) ×Alg(C) ModL (N) →• Fun(N(∆)op , D) → D,

where the second arrow is a geometric realization functor (that is, a left adjoint to the diagonal embedding
D → Fun(N(∆)op , D)).

Remark 4.5.2. In the situation of Definition 4.5.1, we can identify objects of ModR (M) ×Alg(C) ModL (N)
with triples (M, A, N ), where A is an algebra object of C, M is a right A-module, and N is a left A-module.
In this case, we will denote the image of (M, A, N ) under the relative tensor product functor by M ⊗A N .
q q0
Remark 4.5.3. Let M⊗ → C⊗ ← N⊗ be as in Definition 4.5.1, and let F : M⊗ ×C⊗ N⊗ → D be a
balanced pairing. Let M = M⊗ ⊗ ⊗ ⊗
[0] , N = N[0] . Then the inclusion M[0] ×C⊗ N[0] ⊆ M × N is an equivalence
[0]
of ∞-categories, so that F the restriction of F to the inverse image of [0] ∈ ∆ determines a functor
F0 : M × N → D, well-defined up to equivalence. We will denote this functor by F0 (M, N ) = hM, N i.
Likewise, the restriction of F to the inverse image of [1] ∈ ∆ determines a functor F1 : M × C × N → D
Moreover, since F carries coCartesian morphisms to equivalences in D, we have natural equivalences

F0 (M ⊗ C, N ) ' F1 (M, C, N ) ' F0 (M, C ⊗ N ),

which we can view as a compatibility condition on h, i:

γ : hM ⊗ C, N i ' hM, C ⊗ N i.

The restriction of F to the inverse image of the remainder of N(∆)op can be viewed as expressing the
compatibility of γ with the associative tensor product on the ∞-category C.
Let C be a monoidal ∞-category, let M be an ∞-category which is right-tensored over C. Our next goal
is to show that the prescription hM, Ci = M ⊗ C can be extended to a balanced pairing M⊗ ×C⊗ C⊗,L → M.
Proposition 4.5.4. Let C be a monoidal ∞-category with unit object 1C , let M be an ∞-category which is
right-tensored over C, let D be an arbitrary ∞-category, and let

Fun0 (M⊗ ×C⊗ C⊗,L , D) ⊆ Fun(M⊗ ×C⊗ C⊗,L , D)

denote the full subcategory spanned by the balanced pairings. Then the functor

F 7→ h1C , •i

induces an equivalence of ∞-categories Fun0 (M⊗ ×C⊗ C⊗,L , D) → Fun(M, D).

Proof. We will use the theory of marked simplicial sets described in §T.3.1. However, we make a slight
departure from the notation employed there: given a coCartesian fibration of simplicial sets p : X → S,
we let X \ denote the marked simplicial set (X, E), where E is the set of p-coCartesian edges of X. This
notation is potentially ambiguous, since it depends not only on X but also on the map p. In practice, we

150
will have either S ' N(∆)op or S ' ∆0 (in the latter case, X is an ∞-category and E is the collection of all
equivalences in X).
Let us introduce a bit of notation. Let q : M⊗ → C⊗ → N(∆)op exhibit M = M⊗ [0] as right-tensored over
C = C⊗ ⊗
[1] . Let N denote the fiber product M ×C⊗ C
⊗,L
. Let α : ∆1 × N(∆)op → N(∆)op be defined as in
Example 2.1.3. We define a simplicial set M⊗,L equipped with a map M⊗,L → N(∆)op via the formula

HomN(∆)op (K, M⊗,L ) = Hom0N(∆)op (∆1 × K, M⊗ )

α
where ∆1 × K maps to N(∆)op via the composition ∆1 × K → ∆1 × N(∆)op → N(∆)op , and

Hom0N(∆)op (∆1 × K, M⊗ ) ⊆ HomN(∆)op (∆1 × K, M⊗ )

denotes the subset consisting of those maps f : ∆1 × K → M⊗ which carry each edge ∆1 × {k} to a
q-coCartesian edge of M⊗ .
Let N(∆0 )op denote the ∞-category N(∆)op , regarded as an object of (Set∆ )/ N(∆)op via the map α0 :
α|{0}×N(∆)op , and let M⊗ ⊗ 0 op
L denote the fiber product M ×N(∆)op N(∆ ) . Restriction determines a functor
M⊗,L → M⊗ ⊗
L . Regard ML as a simplicial set over N(∆)
op
via the composition M⊗ L → N(∆ )
0 op
' N(∆)op .
⊗ ⊗,L 0
Let N denote the fiber product M ×C⊗ C . We observe that Fun (N, D) and Fun(M, D) can be
identified with Map[ (N\ , D\ ) and Map[ (M\ , D\ ), respectively. In view of Proposition T.[Link], it will suffice
to show that the map M ' {1C } × M ⊆ C × M ' N[0] ⊆ N induces a weak of marked simplicial sets
M\ ⊆ N\ . Corollary T.[Link] implies that the restriction maps N ← M⊗,L → M⊗ L . are trivial Kan fibrations.
Consequently, it will suffice to prove that the composition ψ : M ' {1C } × M ⊆ C × M ' (M⊗ L )[0] ⊆ ML
⊗

induces a weak equivalence M\ → (M⊗ \

L) .
Let F : N(∆) → Cat∞ classify the coCartesian fibration q. According to Proposition T.[Link], the
op
F
marked simplicial set (M⊗ \
L ) can be identified with the colimit of diagram N(∆ )
0 op
→ N(∆)op → Cat∞ .
Lemma T.[Link] implies that this colimit is equivalent to F ([0]) ' M. We now observe that the composition
of this equivalence with the map ψ is given by the functor M 7→ 1C ⊗ M , which is equivalent to idM and
therefore an equivalence.

Definition 4.5.5. Let C be a monoidal ∞-category with unit object 1C , let M be an ∞-category which is
right-tensored over C. A canonical balanced pairing is a balanced pairing M⊗ ×C⊗ C⊗,L → M whose image
under the equivalence described in Proposition 4.5.4 is equivalent to the identity functor idM ∈ Fun(M, M).
In other words, a canonical balanced pairing is a balanced pairing for which the induced functor h1C , •i :
M → M is equivalent to the identity.

Remark 4.5.6. In the situation of Proposition 4.5.4, composition with a canonical balanced pairing

M⊗ ×C⊗ C⊗,L → M

induces a homotopy inverse to the equivalence Fun0 (M⊗ ×C⊗ C⊗,L , D) → Fun(M, D).

Remark 4.5.7. If M is instead right-tensored over C, then we have a dual notion of a canonical balanced
pairing C⊗,R ×C⊗ M⊗ → M. We note that this leads to two different definitions of a canonical balanced
pairing C⊗,R ×C⊗ C⊗,L → C . However, these two notions coincide, since the functors

C 7→ h1C , Ci, C 7→ hC, 1C i

are equivalent to one another.

Proposition 4.5.4 is really a special case of the following more general result (applied to a canonical
pairing Cat×,R ×,L
∞ ×Cat∞ Cat∞ → Cat∞ ):

151
Proposition 4.5.8. Let C be a monoidal ∞-category, M an ∞-category right-tensored over C, A an algebra
object of C, D an ∞-category which admits geometric realizations, F : M⊗ ×C⊗ C⊗,L → D a balanced pairing,
and
⊗A : ModRA (M) × ModA (C) → D
L

the relative tensor product functor. Then the canonical map

φ : hM, 1C i → hM, Ai ' BarA (M, A)1 → | BarA (M, A)• | ' M ⊗A A

determines an equivalence between • ⊗A A and h•, 1C i (regarded as functors from ModA (M) to M).
Proof. Let N, M⊗,L , and M⊗ 0
L be defined as in the proof of Proposition 4.5.4, and let F denote the compo-
s F0
sition M⊗
L →M
⊗,L
→ N → D . Define subcategories ∆00 ⊆ ∆0 ⊆ ∆ as follows:
(i) Every object of ∆ is an object of ∆0 ; an object [n] ∈ ∆ belongs to ∆00 if and only if n ≥ 0.
(ii) A morphism α : [m] → [n] of ∆ belongs to ∆0 if and only if α(m) = n. If m, n > 0, then α belongs to
∆00 if and only if α−1 {n} = {m}.
We have an equivalence of categories ∆ ' ∆00 , given by [n] 7→ [n] ? [0].
Observe that N(∆0 )op contains [0] as a final object (in fact, as a zero object). Let M⊗ ∞ denote the fiber
product N(∆0 )op ×N(∆)op M⊗ , and let q∞ : M⊗ ∞ → N(∆ 0 op
) be the projection. Lemma T.[Link] and
Proposition T.[Link] imply that the inclusion (M⊗ L )\
⊆ (M ⊗ \
∞ ) is a weak equivalence of marked simplicial
0 ⊗ 0
sets. It follows that F extends to a map F : M∞ → D, where F carries q∞ -coCartesian morphisms to
0

equivalences in D.
The simplicial object BarA (M, A)• can be identified with the composition
M F0
N(∆)op ' N(∆00 )op → M⊗
L → D.

0
M F
It follows that BarA (M, A)• extends to a functor U : N(∆∞ )op → M⊗
∞ → D . Lemma T.[Link] implies that
U determines an equivalence
0
M ⊗A A ' | BarA (M, A)• | ' U ([0]) ' F (M ([0])).

We note that there is a q∞ -coCartesian edge joining (M ([0]), 1C ) ∈ M × C ' M⊗ ⊗

[1] to M ([0]) in M∞ . Since
0
F carries q∞ -coCartesian edges to equivalences in D, we conclude that the map φ is an equivalence, as
desired.
q q0
Corollary 4.5.9. Let M⊗ → C⊗ ← N⊗ be as in Definition 4.5.1, let F : M⊗ ×C⊗ N⊗ → D be a balanced
pairing. Suppose that D admits geometric realizations. Let A ∈ Alg(C), M ∈ ModA (M), N0 ∈ N, and let
N ∈ ModA (N) be a free A-module generated by N0 . Then the composite map

hM, N0 i → hM, N i = BarA (M, N )• → | BarA (M, N )• | ' M ⊗A N

is an equivalence in D.
Proof. Using Corollary 2.6.7, we can choose a map of ∞-categories F : C⊗,L → N⊗ of ∞-categories left-
tensored over C⊗ , such that F carries the unit object of C to N0 ∈ N. In view of Proposition 2.4.2, we may
identify N with the image under F of A, regarded as a left A-module over itself as in Example 2.1.7. We
may therefore reduce to the case where N⊗ = C⊗,L , and N0 = 1C ∈ C. The desired result now follows from
Proposition 4.5.8.
We now establish some formal properties enjoyed by the two-sided bar construction and the relative
tensor product.

152
 q q0
Proposition 4.5.10. Let M⊗ → C⊗ ← N⊗ be as in Definition 4.5.1, let F : M⊗ ×C⊗ N⊗ → D be a balanced
pairing. Let K be a simplicial set and let A ∈ Alg(C). Suppose that:
(i) The ∞-categories D and N = N⊗
[0] admit K-indexed colimits.

(ii) The ∞-category D admits geometric realizations for simplicial objects.

(iii) For each M ∈ M, the pairing hM, •i : N → D preserves K-indexed colimits.

A (M), the relative tensor product functor M ⊗A • : ModA (N) → D preserves

Then for each M ∈ ModR L

K-indexed colimits.
Proof. In view of Lemma T.[Link], it will suffice to show that for every n ≥ 0, the functor

N 7→ BarA (M, N )•

preserves K-indexed colimits. We now observe that there is canonical equivalence BarA (M, N )• ' hM ⊗
A⊗n , N i of functors N → D. The desired result now follows from (iii).
q q0
Let M⊗ → C⊗ ← N⊗ be as in Definition 4.5.1, and let F : M⊗ ×C⊗ N⊗ → D be a balanced pairing. Let
A be an algebra object of C, and let

A (M) → M,
θ : ModR A (N) → N
θ0 : ModL

denote the forgetful functors. We observe that, for every M ∈ ModR L

A (M) and N ∈ ModA (N), we have a
canonical map
hθ(M ), θ0 (N )i ' BarA (M, N )• → | BarA (M, N )• | ' M ⊗A N.
q q0
Proposition 4.5.11. Let M⊗ → C⊗ ← N⊗ be as in Definition 4.5.1, let F : M⊗ ×C⊗ N⊗ → D be a
balanced pairing. Suppose that D admits geometric realizations for simplicial objects, and let A be an initial
object of Alg(C) (see Proposition 1.4.3). Then for every M ∈ ModR L
A (M), N ∈ ModA (N), the natural map
φ : hθ(M ), θ0 (N )i → M ⊗A N is an equivalence in D.
Proof. The relative tensor product M ⊗A N is defined as the geometric realization of the simplicial object
BarA (M, N )• : N(∆)op → D. Since N(∆)op is weakly contractible, Corollary T.[Link] implies that φ is an
equivalence provided that BarA (M, N )• carries each morphism in ∆ to an equivalence in D. Unwinding the
definitions, we see that each of these morphisms can be identified with the map

hθ(M ), A⊗m ⊗ θ(N )i → hθ(N ), A⊗n ⊗ θ(N )i

determined by a map of linearly ordered sets [m] → [n]. The desired result now follows easily from the fact
that A is a unit object of C (Proposition 1.4.3).
Lemma 4.5.12. Let C be an ∞-category equipped with a monoidal structure, and let M be an ∞-category
which is left-tensored over C. Suppose that M admits geometric realizations, and that for each C ∈ C the
functor C ⊗• preserves geometric realizations. Let f : A → A0 be a morphism in Alg(C). Then the associated
functor θ : ModA0 (M) → ModA (M) (see Corollary 2.3.3) admits a left adjoint.
Proof. Apply Corollary 3.5.7 to the homotopy commutative diagram

ModA0 (M) / ModA (M)

JJ
JJ uu
JJ
JJ uuuuu
JJ
$ uz uu
M,

together with Proposition 2.4.2 and Corollary 2.3.7.

153
Proposition 4.5.13 (Push-Pull Formula). Let

q q0
M⊗ → C⊗ ← N⊗

be as in Definition 4.5.1, let F : M⊗ ×C⊗ N⊗ → D be a balanced pairing. Suppose that:

(i) The ∞-categories D and N admit geometric realizations.
(ii) For every C ∈ C, the functor C ⊗ • : N → N preserves geometric realizations.

(iii) For every M ∈ M, the pairing hM, •i : N → D preserves geometric realizations.

r r0
Let ModR (M) → Alg(C) ← ModL (N) be the induced functors, and let

α = (αR , α0 , αL ) : (M, A, N ) → (M 0 , A0 , N 0 )

be a morphism in ModR (M) ×Alg(C) ModL (N). Suppose that αR is an r-Cartesian morphism of ModR (M)
and αL is an r0 -coCartesian morphism of ModL (N). Then α induces an equivalence M ⊗A N → M 0 ⊗A0 N 0
in D.
Proof. Let us first regard αR : M → M 0 and α0 : A → A0 as fixed. Lemma 4.5.12 implies that r0 is a
coCartesian fibration, so that for each N ∈ ModA (M) there exists an (essentially unique) r0 -coCartesian
morphism αL : N → α!0 N lifting α0 . Let Mod0A (N) ⊆ ModA (N) denote the full subcategory spanned by
those A-modules N for which the induced map M ⊗A N → M 0 ⊗A0 (α!0 N ) is an equivalence. Proposition
4.5.10 implies that Mod0A (N) is closed under geometric realizations in ModA (N) (which admits geometric
realizations in view of Corollary 2.3.7). Proposition 3.5.5 implies that ModA (N) is generated, under geometric
realizations, by the essential image of the functor β! : ModA0 (N) → ModA (N), where A0 denotes the initial
object of Alg(C). Consequently, we may assume that N ' β! N0 . Let M0 denote the image of M under the
forgetful functor ModA (M) → ModA0 (M). We have a commutative triangle

M7 ⊗A NN
ppp NNN
pp
ψ0
p NNψN
ppp NNN
pp N&
ψ 00
M0 ⊗A0 N0 / M 0 ⊗A0 N 0

in D. Consequently, to prove that ψ is an equivalence, it will suffice to prove that ψ 0 and ψ 00 are equivalences.
In other words, we may reduce to the case where A is an initial object of Alg(C). The desired result now
follows immediately from Proposition 4.5.11 and Corollary 4.5.9.
Corollary 4.5.14. Let C, M, and f : A → A0 be as in Lemma 4.5.12, and let F : ModA (M) → ModA0 (M)
be the left adjoint to the forgetful functor ModA0 (M) → ModA (M), and let θ : ModA0 (M) → M be the
F θ
forgetful functor. Then the composition ModA (M) → ModA0 (M) → M can be identified with the relative
tensor product M 7→ A0 ⊗A M determined by a canonical balanced pairing C⊗,R ×C⊗ M⊗ → M.

Proof. According to Proposition 4.5.8, the forgetful functor θ can be identified with the relative tensor
product A0 ⊗A0 •. Proposition 4.5.13 implies that A0 ⊗A0 F (M ) is (canonically) equivalent to A0 ⊗A M . Here
we identify A0 with its image in ModR R R
A (C) under the forgetful functor ModA0 (C) → ModA (C).

Remark 4.5.15. Let C, M, and f : A → A0 satisfy the hypotheses of Lemma 4.5.12. We will sometimes
abuse notation by indicating the associated functor ModL L 0
A (M) → ModA0 (M) by M 7→ A ⊗A M . This
notation is partially justified by Corollary 4.5.14.

154
4.6 Flat Modules
Let A be a connective A∞ -ring. In this section, we will show that there is a good theory of flat and projective
A-modules, which reduces to the classical theory in the case where A is discrete.
Definition 4.6.1. Let A be an A∞ -ring. We will say that a left A-module M is free if is equivalent to a
coproduct of copies of A (where we view A as a left module over itself, as in Example 2.1.7). We will say
that a free module M is finitely generated if it can be written as a finite coproduct of copies of A.
Suppose that A is connective. We will say that a map f : M → N of connective (left) A-modules is
surjective if it induces a surjection of π0 A-modules π0 M → π0 N .
Warning 4.6.2. The terminologies introduced in Definitions 4.6.1 and 2.4.1 are not quite compatible with
one another. Let M be a (left) module over an A∞ -ring A. Then M is free (in the sense of Definition 4.6.1)
if and only if it is freely generated (in the sense of Definition 2.4.1) by a coproduct of copies of the sphere
spectrum. If M0 is a general spectrum, then the left A-module A ⊗ M0 freely generated by M0 is typically
not free in the sense of Definition 4.6.1.
Remark 4.6.3. Using the long exact sequence of homotopy groups associated to an exact triangle, we
conclude that a map f : M → N of connective modules over a connective A∞ -ring is surjective if and only
if ker(f ) is also connective.
Definition 4.6.4. Let A be a connective A∞ -ring. We will say that a left A-module P is projective if it
a projective object of the ∞-category (ModL
A )≥0 of connective left A-modules, in the sense of Definition
T.[Link].
Remark 4.6.5. The terminology of Definition 4.6.4 is potentially ambiguous: a projective left A-module is
typically not projective as an object of ModL
A . However, there is little risk of confusion, since the ∞-category
ModL A has no nonzero projective objects.

Proposition 4.6.6. Let A be a connective A∞ -ring, and let P be a connective left A-module. The following
conditions are equivalent:

(1) The left A-module P is projective.

(2) There exists a free A-module M such that P is a retract of M .
Proof. Suppose first that P is projective. Choose a map of left A-modules p : M → P , where M is free and
the induced map π0 M → π0 P is surjective (for example, we can take M to be a direct sum of copies of A
indexed by the set π0 P ). Invoking Proposition S.6.16, we deduce that p admits a section (up to homotopy), so
that P is a retract of M . This proves (2). To prove the converse, we observe that the collection of projective
left A-modules is stable under retracts. It will therefore suffice to show that every free left A-module is
projective. This follows immediately from the characterization given in Proposition S.6.16.
Remark 4.6.7. It follows from the proof of Proposition 4.6.6 that, if π0 P is a finitely generated left module
over π0 A, then we can choose M to be a finitely generated free A-module.

Corollary 4.6.8. Let A be a connective A∞ -ring. Then the ∞-category (ModL A )≥0 is projectively generated
(Definition T.[Link]). Moreover, the following conditions on a connective left A-module P are equivalent:
(1) The A-module P is projective, and π0 P is finitely generated as a π0 A-module.
(2) The A-module P is a compact projective object of (ModL
A )≥0 .

(3) There exists a finitely generated free A-module M such that P is a retract of M .

155
Proof. The first assertion and the equivalence (2) ⇔ (3) follow by applying Corollary 3.5.8 to the composition
Ω∞
(ModL
A )≥0 → Sp
conn
→ S,

and invoking Example T.[Link]. The equivalence (1) ⇔ (3) follows from Remark 4.6.7.
Recall that, if M is a left module over an ordinary ring A, we say that M is flat if the functor N 7→ N ⊗A M
is exact.
Definition 4.6.9. Let M be a left module over an A∞ -ring A. We will say that M is flat if the following
conditions are satisfied:
(1) The homotopy group π0 M is flat as a left module over π0 A, in the classical sense.
(2) For each n ∈ Z, the natural map πn A ⊗π0 A π0 M → πn M is an isomorphism of abelian groups.
Remark 4.6.10. Let A be a connective A∞ -ring. Then every flat left A-module is also connective.
Remark 4.6.11. Let A be a discrete A∞ -ring. A left A-module M is flat if and only if M is discrete, and
π0 M is flat over π0 A (in the sense of classical ring theory). In other words, Definition 4.6.9 is compatible
with the classical definition of flatness, if we identify discrete A∞ -rings and modules with the underlying
classical objects.
In order to use Definition 4.6.9 effectively, we need to understand the relationship between the relative
tensor products constructed in §4.5 and the tensor products in classical noncommutative algebra.
Notation 4.6.12. Let A be an ordinary associative ring, let M be a right A-module and N a left A-module.
Homological algebra associates to the triple (M, A, N ) a collection of groups TorA p (M, N ), given by the left
derived functors of the (classical) tensor product ⊗A (see [75]). We observe that if A is a graded ring and the
modules M and N are compatibly graded, then TorA p (M, N ) inherits a grading (as can be seen by performing
homological algebra in the setting of graded A-modules). In this case, we will denote the qth graded piece of
TorA A
p (M, N ) by Torp (M, N )q .

Proposition 4.6.13. Let A be an A∞ -ring, let M be a right A-module, and let N be a right A-module.
We regard π∗ M and π∗ N as graded modules over the graded ring π∗ A. There exists a convergent spectral
sequence with E2 -page
E2p,q = Torpπ∗ A (π∗ M, π∗ N )q ⇒ πp+q (M ⊗A N ).
Proof. We will say that a left A-module P is quasi-free if P can be obtained as a direct sum of A-modules
of the form A[n], where n ∈ Z. We will say that a map of left A-modules P → N is quasi-surjective if the
induced map π∗ P → π∗ N is a surjection of graded π∗ A-modules. We observe that for every left A-module
N , there exists a quasi-surjection P → N , where P is quasi-free. For example, we can take P to be a direct
sum ⊕η A[n], where η ranges over πn N .
We now construct a filtered object

. . . → Q(−1) → Q(0) → Q(1) → . . .

in the ∞-category (ModL A )/N . We begin by setting Q(i) = 0 for i < 0. Suppose that i ≥ 0, and that the
map f (i − 1) : Q(i − 1) → N has already been constructed. Choose a quasi-surjection P (i) → coker f (i − 1),
where P (i) is quasi-free, and form a pullback square

Q(i) / P (i)

f (i)
 
N / coker f (i − 1)

156
in the ∞-category (ModL L
A )Q(i−1)/ (and therefore also in ModA ). We have a map of distinguished triangles

Q(i − 1) / Q(i) / P (i) / P (i)[1]

f (i)
f (i−1)   
Q(i − 1) /N / coker f (i − 1) / Q(i − 1)[1].

We have a quasi-surjective map δ(0) : P (0) ' Q(0) → N . For each i > 0, let δ(i) denote the composite
map
P (i + 1) → Q(i)[1] → P (i)[1].
We claim that the induced sequence of left π∗ A-modules

. . . → π∗+1 P (1) → π∗ P (0) → π∗ N → 0

is exact. The exactness at π∗ N is obvious. We will prove the exactness at π∗+i P (i) for i > 0; the proof for
i = 0 requires only simple changes of notation and is left to the reader. We first observe that the composition
δ(i) ◦ δ(i + 1)[−1] factors through Q(i) → P (i) → Q(i − 1)[1], which is nullhomotopic. Now suppose that
η ∈ πn P (i) lies in the kernel of the map πn P (i) → πn−1 P (i − 1). Let η0 ∈ πn−1 Q(i − 1) denote the image
of η. Using the exactness of the sequence

πn−1 Q(i − 2) → πn−1 Q(i − 1) → πn−1 P (i),

we conclude that η0 is the image of some η1 ∈ πn−1 Q(i − 2), such that the image of η1 in πn−1 N is zero.
Using the exactness of the sequence

πn coker f (i − 2) → πn−1 Q(i − 2) → πn−1 N

and the assumption that the map P (i − 1) → coker f (i − 2) is quasi-surjective, we conclude that η1 is itself
the image of some η2 ∈ πn P (i − 1). Since the composition P (i − 1) → Q(i − 2)[1] → Q(i − 1)[1] is null, we
conclude that η0 = 0. The exactness of the sequence

πn Q(i − 1) → πn Q(i) → πn P (i) → πn−1 Q(i − 1)

shows that η is the image of an element ηe ∈ πn Q(i). Moreover, we are free to modify ηe by adding the image
of any element in πn Q(i − 1). Since the map πn Q(i − 1) → πn N is surjective, we may assume without loss
of generality that the image of ηe in πn N is zero. Using the exactness of the sequence

πn+1 coker f (i) → πn Q(i) → πn N

and the assumption that the map P (i + 1) → coker f (i) is quasi-surjective, we conclude that ηe is the image
of an element of πn+1 P (i + 1), as desired.
Let Q(∞) be a colimit of the sequence

. . . Q(−1) → Q(0) → Q(1) → . . .

in the ∞-category of left A-modules. Applying Proposition S.11.13 to the above sequence (regarded as
a filtered object of the ∞-category of spectra), we conclude that there is a spectral sequence {Erp,q , dr }
which converges to π∗ Q(∞). The exactness statement established above shows that this spectral sequence
degenerates at r = 2, with (
p,q 0 if p 6= 0
E2 =
πq N if p = 0.
It follows that the natural map Q(∞) → N is an equivalence, so that we can identify N with a colimit of
the sequence Q.

157
 Proposition 4.5.10 implies that M ⊗A N can be identified with a colimit of the filtered spectrum

. . . → M ⊗A Q(0) → M ⊗A Q(1) → . . .

Applying Proposition S.11.13 again, we deduce the existence of another spectral sequence

E1p,q = πp+q (M ⊗A P (p)) ⇒ πp+q (M ⊗A N ).

To complete the proof, it suffices to observe that since each P (p) is quasi-free, the natural map π∗ M ⊗π∗ A
π∗ P (p) → π∗ (M ⊗A P (p)) is an isomorphism. It follows that the E2p,q term of the above spectral sequence
can be identified with the homologies of the complex

. . . → π∗ M ⊗π∗ A π∗+1 P (1) → π∗ M ⊗π∗ A π∗ P (0),

which coincide with the Tor-groups Tor∗π∗ A (π∗ M, π∗ N ).

Remark 4.6.14. In the situation of Proposition 4.6.13, the E1 -page of the spectral sequence depends on
the choice of the filtered object
. . . Q(−1) → Q(0) → Q(1) → . . .
However, one can show that the E2 -page is independent of this choice, and is functorially determined by the
triple (M, A, N ).
Corollary 4.6.15. Let A be an A∞ -ring, M a right A-module, and N a left A-module. Suppose that A, M ,
and N are discrete. Then there exists a canonical isomorphism

πn (M ⊗A N ) ' Tornπ0 A (π0 M, π0 N ).

Corollary 4.6.16. Let A be an A∞ -ring, M a right A-module, and N a left A-module. Suppose that N is
flat. For each n ∈ Z, the canonical homomorphism

θ : πn M ⊗π0 A π0 N → πn (M ⊗A N )

is an isomorphisms of abelian groups.

Proof. If N is flat, then Torpπ∗ A (π∗ M, π∗ N ) vanishes for p > 0, and is isomorphic to π∗ M ⊗π0 A π0 N for
p = 0. It follows that the spectral sequence of Proposition 4.6.13 degenerates at the E2 -page and yields the
desired result.
Corollary 4.6.17. Let A be a connective A∞ -ring, M a connective right A-module, and N a connective left
A-module. Then:
(1) The relative tensor product M ⊗A N is connective.

(2) There is a canonical isomorphism π0 (M ⊗A N ) ' π0 M ⊗π0 A π0 N in the category of abelian groups.
Proof. This follows from the spectral sequence of Proposition 4.6.13, since E2p,q vanishes for p < 0 or q < 0,
while E20,0 ' π0 M ⊗π0 A π0 N .
Our next goal is to prove an analogue of Lazard’s theorem, which characterizes the class of flat modules
over a connective A∞ -ring.
Lemma 4.6.18. Let A be an A∞ -ring.
(1) The A∞ -ring A is flat, when viewed as a left module over itself.
(2) The collection of flat A-modules is stable under coproducts, retracts, and filtered colimits.

158
 (3) Every free A-module is flat. If A is connective, then every projective A-module is flat.

Proof. Assertions (1) and (2) are obvious, and (3) follows from Proposition S.6.16.
Theorem 4.6.19 (Lazard’s Theorem). Let A be a connective A∞ -ring, and let N be a connective left
A-module. The following conditions are equivalent:
(1) The left A-module N can be obtained as a filtered colimit of finitely generated free modules.

(2) The left A-module N can be obtained as a filtered colimit of projective left A-modules.
(3) The left A-module N is flat.
(4) The functor M 7→ M ⊗R N is left t-exact; in other words, it carries (ModR
A )≤0 into (Sp)≤0 .

(5) If M is a discrete right A-module, then M ⊗R N is discrete.

Proof. The implications (1) ⇒ (2) ⇒ (3) are obvious. The implication (3) ⇒ (4) follows from Corollary
4.6.16, and (4) ⇒ (5) follows from Corollary 4.6.17.
We next show that (5) ⇒ (3). Suppose that (5) is satisfied. The functor M 7→ M ⊗R N is exact, and
carries the heart of ModR A into itself. It therefore induces an exact functor from the abelian category of right
π0 A-modules to itself. According to Corollary 4.6.17, this functor is given by (classical) tensor product with
the left π0 A-module π0 N . From the exactness we conclude that π0 N is a flat π0 A-module.
We now prove that the natural map φ : πn A ⊗π0 A π0 N → πn N is an isomorphism for all n ∈ Z. For
n < 0, this follows from the assumption that both N and A are connective. If n = 0 there is nothing to
prove. We may therefore assume that n > 0, and we work by induction on n. Let M be the discrete right
A-module corresponding to π0 A. The inductive hypothesis implies that Torπp ∗ A (π∗ M, π∗ N )q vanishes unless
p = q = 0 or q ≥ n. These Tor-groups can be identified with the E2 -terms of the spectral sequence of
Proposition 4.6.13, which computes the homotopy groups of the discrete spectrum M ⊗A N . Consequently,
we have E∞ 0,0
= E20,0 ' π0 N . A simple calculation shows that E20,n ' coker(φ), and that if φ is surjective
1,n
then E2 ' ker(φ). To complete the proof, it suffices to prove that E2i,n ' ∗ for 0 ≤ i ≤ 1. To see this, we
observe that the vanishing of the groups E2i−r,n+r−1 and E2i+r,n−r+1 for r ≥ 2 implies that E2i,n ' E∞ i,n
, and
the latter is a subquotient of πi+n (M ⊗A N ), which vanishes in view of assumption (5).
To complete the proof, it will suffice to show that (3) implies (1). Let C be the full subcategory of ModA
spanned by a set of representatives for all free, finitely generated A-modules. Then C is small, and consists of
compact projective objects of (ModA )≥0 , which generate (ModA )≥0 under colimits. It follows (see §T.5.5.8)
that the inclusion C ⊆ ModA induces an equivalence PΣ (C) → (ModA )≥0 . Applying Lemma T.[Link], we
conclude that the identity functor from (ModA )≥0 is a left Kan extension of its restriction to C. It follows
that for every connective left A-module N , the canonical diagram C/N = C ×ModA (ModA )/N → ModA has
N as a colimit. To complete the proof, it will suffice to show that C/N is filtered provided that N is flat.
According to Proposition T.[Link], it will suffice to verify the following conditions:
(i) For every finite collection of objects {Xi } of C/N , there exists an object X ∈ C/N together with
morphisms Xi → X.

(ii) For every pair X, Y ∈ C/N , every nonnegative integer n ≥ 0, and every map S n → MapC/N (X, Y )
in the homotopy category H, there exists a morphism Y → Z in C/N such that the induced map
S n → MapC (X, Z) is nullhomotopic.

Assertion (i) follows immediately from the stability of C under finite coproducts. We now prove (ii).
Suppose given a pair of maps f : X → N and g : Y → N , respectively. We have a homotopy fiber sequence

MapC/N (f, g) → MapC (X, Y ) → MapModA (X, N ).

159
Since ModA is stable, MapC/N (f, g) is a torsor over MapModA (X, ker(g)). It follows that any map S n →
MapModA (X, ker(g)) determines a homotopy class η ∈ Ext−n ModA (X, ker(g)). We wish to prove that there
exists a commutative diagram
Y @ /Z
@@ g ~~
@@ ~~
@@
@ ~~~~ h
N
such that the image of η in Ext−n ModA (X, ker(h)) vanishes. Arguing iteratively, we can reduce to the case
where X ' A, so that η can be identified with an element of πn ker(g).
Let η 0 ∈ πn Y denote the image of η. Our first step is to choose a diagram as above with the property
that the image of η 0 in πn Z vanishes. We observe that η 0 lies in the kernel of the natural map πn (g) :
πn Y → πn N ' Torπ0 0 A (πn A, π0 N ). The classical version of Lazard’s theorem (see [39]) implies that π0 N is
isomorphic to a filtered colimit of free left π0 A-modules. It follows that there exists a commutative diagram

{= CCC
P
k {{{ CCh
{{ CC
{{ C!
π0 Y
π0 g
/ π0 N

of left π0 A-modules, where P is a finitely generated free module, and the image of η 0 in πn A ⊗π0 A P vanishes.
Using the freeness of P , we can realize h as π0 h, where h : Z → N is a morphism of left A-modules, and Z
is a finitely generated free module with π0 Z ' P . Similarly, we can realize k as π0 k, where k : Y → Z is a
morphism of left A-modules. Using the freeness of Y again, we conclude that the diagram

? Z @@
k ~~~ @@h
~~ @@
~~ @
Y
g
/N

commutes in the homotopy category hModA , and can therefore be lifted to a commutative triangle in ModA .
By construction, the image of η 0 in πn Z vanishes. Replacing Y by Z, we may reduce to the case where
η 0 = 0.
We now invoke the exactness of the sequence sequence

πn+1 N → πn ker(g) → πn Y

to conclude that η is the image of a class η 00 ∈ πn+1 N ' πn+1 A ⊗π0 A π0 N . Invoking Lazard’s theorem once
more, we deduce the existence of a commutative diagram of left π0 A-modules

= Q CC
||| CC
|| CC
| CC
|| !
π0 Y
π0 g
/ π0 N

where Q is a finitely generated free module, and η 00 is the image of some element of πn+1 A ⊗π0 A N . Arguing
as before, we may assume that the preceding diagram is induced by a commutative triangle of left A-modules

? Z AA
~~~ AA
AA
~~ AA
~~
Y
g
/ N.

160
Replacing Y by Z, we may assume that η 00 lies in the image of the map πn+1 Y → πn+1 N . The exactness of
the sequence
πn+1 Y → πn+1 N → πn ker(g)
now implies that η = 0, as desired. This completes the proof of the implication (3) ⇒ (1).
We now study the behavior of flatness under base change.
Proposition 4.6.20. Let f : A → B be a map of A∞ -rings, let G : ModB → ModA be the forgetful functor,
and let F : ModA → ModB be a left adjoint to G (given by M 7→ B ⊗A M , in view of Corollary 4.5.14).
Then:
(1) The functor F carries free (projective, flat) A-modules to free (projective, flat) B-modules.
(2) Suppose that B is free (projective, flat) as a left A-module (that is, G(B) is free, projective, or flat).
Then G carries free (projective, flat) B-modules to free (projective, flat) A-modules.
(3) Suppose that, for every n ≥ 0, the map f induces an isomorphism πn A → πn B. Then F induces an
equivalence of categories Mod[A → Mod[B ; here Mod[A denotes the full subcategory of ModA spanned by
the flat left A-modules, and Mod[B is defined likewise.
Proof. Assertions (1) and (2) are obvious. To prove (3), we first choose A0 to be a connective cover of A (see
Proposition 4.3.8). We have a homotopy commutative triangle of ∞-categories

Mod[A
v; HH
vv HH
vv HH
vv HH
vv H$
[
ModA0 / Mod[B .

It therefore suffices to prove the analogous assertion for the morphisms A0 → A and A0 → B. In other words,
we may reduce to the case where A is connective.
Since A is connective, the ∞-category ModA admits a t-structure. Let F 0 denote the composite functor
F
(ModA )≥0 ⊆ ModA → ModB .

Then F 0 has a right adjoint, given by the composition τ≥0 ◦G. Assertion (1) implies that F 0 preserves flatness,
and a simple calculation of homotopy groups shows that G0 preserves flatness as well. Consequently, F 0 and
G0 induce adjoint functors
F 00 /
Mod[A o 00 Mod[B .
G

It now suffices to show that the unit and counit of the adjunction are equivalences. In other words, we must
show:
(i) For every flat left A-module M , the unit map M → τ≥0 B ⊗A M is an equivalence. For this, it suffices
to show that πi M ' πi τ≥0 B ⊗A M is an isomorphism for i ∈ Z. If i < 0, then both groups vanish,
so there is nothing to prove. If i ≥ 0, then we must show that πi M ' πi (B ⊗A M ), which follows
immediately from Corollary 4.6.16 and the assumption that πi A ' πi A.
(ii) For every flat left B-module N , the counit map B ⊗A τ≥0 G(N ) → N is an equivalence. In other words,
we must show that for each j ∈ Z, the map πj (B ⊗A τ≥0 G(N )) → πj N is an isomorphism of abelian
groups. Since G(N ) is flat over A, Corollary 4.6.16 implies that the left side is given by πj B ⊗π0 A π0 N .
The desired result now follows immediately from our assumption that N is flat.

161
 In general, if M is a flat left A-module, then “global” properties of M as an A-module are often controlled
by “local” properties of π0 M , viewed as a module over the ordinary ring π0 A. Our next pair of results
illustrates this principle.
Lemma 4.6.21. Let A be an A∞ -ring, and let f : M → N be a map of flat left A-modules. Then f is an
equivalence if and only if it induces an isomorphism π0 M → π0 N .
Proof. This follows immediately from the definition of flatness.
Proposition 4.6.22. Let A be a connective A∞ -ring. A flat left A-module M is projective if and only if
π0 M is a projective module over π0 A.
`
Proof. Suppose first that π0 M is freely generated by elements {ηi }i∈I . Let P = i∈I A, and let f : P → M
be a map represented by {ηi }i∈I . By construction, f induces an isomorphism π0 P → π0 M . Lemma 4.6.21
implies that f is an equivalence, so that M is free.
In the general case, there exists a free π0 A-module F0 and a direct sum decomposition F0 ' N0 ⊕ π0 M .
Replacing F0 by ⊕n≥0 F0 if necessary, we may assume that N0 is itself free. The projection map F0 → π0 M
is induced by a map g : F → M of left A-modules, where F is free. Then g induces a surjection π0 F ' F0 →
π0 M . Using the flatness of M , we conclude that the maps
π i F ' π i A ⊗π 0 A π 0 F → π i A ⊗π 0 A π 0 M ' π i M
are also surjective. Let N be a kernel of g, so that we have a commutative diagram

0 / πi A ⊗π0 A π0 N / π i A ⊗π 0 A π 0 F / π i A ⊗π 0 A π 0 M /0

φ0 φ φ00
  
0 / πi N / πi F / πi M / 0.

Using the flatness of F and M , we deduce that the upper row is short exact, and that the maps φ and
φ00 are isomorphisms. The snake lemma implies that φ is an isomorphism; moreover, π0 N is isomorphic to
the kernel of a surjection between flat π0 A-modules, and is therefore itself flat. It follows that N is a flat
A-module. Since π0 N is free, the first part of the proof shows that N is itself free.
Let p : N → F denote the natural map. Since π0 M is projective, the inclusion π0 N ⊆ π0 F is split. Since
F is free, we can lift this splitting to a morphism q : F → N . Then q ◦ p : N → N induces the identity map
from π0 N to itself. Since N is free, we conclude that there is a homotopy q ◦ p ' idN . It follows that N is a
direct summand of F in the homotopy category hC. Consequently, M ' coker(p) can be identified with the
complementary summand, and is therefore projective.

4.7 Finiteness Properties of Rings and Modules

In this section, we will discuss some finiteness conditions on the ∞-categories of modules over an A∞ -ring
R. We begin by introducing the definition of a perfect R-module. Roughly speaking, an R-module M is
perfect if it can be obtained by gluing together finitely many (possibly shifted) copies of R, or is a retract of
such an R-module. Alternatively, we can describe the class of perfect R-modules as the compact objects of
the ∞-category ModR (Proposition 4.7.2).
In general, the condition that an R-module be perfect is very strong. For example, if R is a discrete
commutative ring and M is a finitely generated (discrete) module over R, then M need not be perfect
when viewed as an object of the stable ∞-category ModR . In general, this is true if and only if R is a
regular Noetherian ring. To remedy the situation, we will introduce the weaker notion of an almost perfect
R-module. This notion has a much closer relationship with finiteness conditions in the classical theory of
rings and modules. In particular, we will show that a left R-module M is almost perfect if and only if each
πi M is a finitely presented left module over π0 R, and πi M ' 0 for i  0 (Proposition 4.7.19), at least when
R itself satisfies a suitable finiteness condition (that is, when R is left coherent in the sense of Definition
4.7.15).

162
Definition 4.7.1. Let A be an A∞ -ring. We let ModL,perf
A denote the smallest stable subcategory of ModL
A
which contains A (regarded as a left module over itself) and is closed under retracts. Similarly, we let
ModR,perf
A denote the smallest stable subcategory of ModR A which contains A and is closed under retracts.
We will say that a left (right) A-module M is perfect if it belongs to ModL,perf
A (ModR,perf
A ).
Proposition 4.7.2. Let A be an A∞ -ring. Then:
(1) The ∞-category ModL
A is compactly generated.

(2) An object of ModL

A is compact if and only if it is perfect.

Proof. The compact objects of ModA form a stable subcategory which is closed under the formation of
Ω∞
retracts. Moreover, Proposition 2.4.2 implies A ∈ ModL A corepresents the composition ModA → Sp → S,
where both maps preserve filtered colimits. It follows that A is compact as an A-module. This proves that
every perfect object of ModL A is compact.
According to Proposition T.[Link], the inclusion f : ModL,perfA ⊆ ModL A induces a fully faithful functor
L,perf L
F : Ind(ModA ) → ModA . To complete the proof, it will suffice to show that F is essentially surjective.
Since f is right exact, F preserves all colimits (Proposition T.[Link]), so the essential image of F is stable
under colimits. If F is not essentially surjective, then Proposition S.16.1 implies that there exists a nonzero
N ∈ ModL A such that MapModL A
(N 0 , N ) ' ∗ for all N 0 belonging to the essential image of F . In particular,
taking N 0 = A[n], we conclude that πn N ' ∗. It follows that N is a zero object of ModL
A , contrary to our
assumption.
Proposition 4.7.3. Let A be an A∞ -ring. The relative tensor product functor

⊗A : ModR L
A × ModA → Sp

induces fully faithful embeddings

θ : ModR L
A → Fun(ModA , Sp) θ0 : ModL R
A → Fun(ModA , Sp).

A functor f : ModL 00 R 0 0
A → Sp (f : ModA → Sp) belongs to the essential image of θ (θ ) if and only if f (f )
preserves small colimits.

Proof. Let C be the full subcategory of Fun(ModL A , Sp) spanned by those functors which preserve colimits.
Proposition T.[Link] implies that C is presentable, and Proposition 4.5.10 implies that θ factors through C.
We will show that θ induces an equivalence G : ModR A → C; the analogous assertion for left modules follows
by the same argument.
L,perf
Proposition 4.7.2 implies that ModL A is equivalent to the ∞-category Ind(ModA ). It follows from
Propositions T.[Link], T.[Link] and S.5.1 that C is equivalent to the ∞-category FunEx (ModL,perf A , Sp) of
exact functors from ModL,perf
A to spectra. In particular, for every perfect left A-module N , evaluation on N
induces a functor C → Sp that preserves all small limits and colimits.
Let G0 : C → Sp be given by evaluation on A, regarded as a (perfect) left module over itself. If α : f → f 0
is a natural transformation of functors from ModL,perf
A to Sp, then the full subcategory of hModL,perf
A spanned
by objects C such that α(C) : f (C) → f 0 (C) is an equivalence is a triangulated subcategory of hModL,perf A
which is stable under retracts. If follows that G0 is conservative: if G0 (α) is an equivalence, then α is an
equivalence. Since G0 preserves small limits, we deduce also that G0 detects small limits: if p : K / → C is
such that G0 ◦ p is a limit diagram in Sp, then p is a limit diagram in C. Similarly, G0 detects small colimits.
The composite functor G0 ◦ G : ModR A → Sp can be identified with the forgetful functor, in view of
Proposition 2.4.2. It follows that G0 ◦ G preserves all limits and colimits (Corollaries 2.3.5 and 2.3.7). Since
G0 detects small limits and colimits, we deduce that G preserves small limits and colimits. Corollary T.[Link]
implies that G and G0 admit left adjoints, which we will denote by F and F 0 .

163
 Choose unit and counit transformations

u : id → G ◦ F, v : F ◦ G → id .

We wish to prove that u and v are equivalences. Since G0 ◦ G detects equivalences, the functor G detects
equivalences, so that v is an equivalence if and only if G(v) : G ◦ F ◦ G → G is an equivalence. Since G(v)
has a section determined by u, it will suffice to prove that u is an equivalence.
Let C0 be the full subcategory of C spanned by those objects f ∈ C for which the map u(f ) : f → (G◦F )(f )
is an equivalence. Since G and F preserve small colimits, we deduce that C0 is stable under shifts and colimits
in C. Proposition 3.5.5 implies that C is generated, under geometric realizations, by the essential image of
F 0 . Let Sp0 ⊆ Sp be the inverse image of C0 under F 0 . Since F 0 is a colimit-preserving functor, we deduce
that Sp0 ⊆ Sp is closed under shifts and colimits. Since Sp is generated under colimits by the objects S[n],
where n ∈ Z and S ∈ Sp denotes the sphere spectrum, it will suffice to show that S ∈ Sp0 .
Proposition 2.4.2 allows us to identify (F ◦ F 0 )(S) with S ⊗ A ' A, regarded as a left module over itself.
Proposition 4.5.8 implies that (G ◦ F )(F 0 (S)) can be identified with the forgetful functor f0 : ModL A → Sp.
We are reduced to proving that the unit map S → A ' f0 (A) induces an equivalence F 0 (S) ' f0 in C.
Applying Proposition S.10.12, we can identify C with the full subcategory D ⊆ Fun(ModL,perf A , S) =
L,perf,op
P(ModA ) spanned by those functors which preserve finite limits. Under this equivalence, f0 corresponds
Ω∞
to the composition ModL,perf
A A → Sp → S, while F (S) corresponds to the image of ∗ ∈ S under the
⊆ ModL 0

Σ∞ F0
composition S → Sp → C ' D which is the left adjoint to the the functor D → S given by evaluation at
A ∈ ModL,perf
A . To complete the proof, it will suffice to show that the unit 1 ∈ π0 A exhibits the composite
Ω∞
functor ModL A → Sp → S as corepresented by A ∈ ModA . In other words, we must show that for every
L
L
M ∈ ModA , the canonical map MapModLA (A, M ) → Ω∞ M is a homotopy equivalence. Using Proposition
2.4.2, we can reduce to the case where A is the unit object of Alg(Sp). In view of Corollary 2.4.4, we are
reduced to proving that if M ∈ Sp, then the canonical map MapSp (S, M ) → Ω∞ M is an equivalence, which
is clear (since S ' Σ∞ (∗) by definition).
Remark 4.7.4. Proposition 4.7.3 admits the following generalization. Suppose that A and B are A∞ -rings.
Then the ∞-category of colimit preserving functors from ModL L
A to ModB is equivalent to an ∞-category of
A-B-bimodules. A precise formulation (and proof) of this statement require some elaboration on the ideas
presented in this section. For a proof in the language of model categories, we refer the reader to [61].
Proposition 4.7.5. Let A be an A∞ -ring, and let M be a left A-module. The following conditions are
equivalent:

(1) The left A-module M is perfect.

(2) The left A-module M is a compact object of ModL
A.

M ∨ ⊗A • Ω∞
(3) There exists a right A-module M ∨ such that the composition ModL
A → Sp → S is equivalent to
the functor corepresented by M .

Moreover, if these conditions are satisfied, then the object M ∨ ∈ ModR

A is also perfect.

Proof. The equivalence (1) ⇔ (2) is a consequence of Proposition 4.7.2. Let C denote the full subcategory
of Fun(ModL A , Sp) spanned by those functors which are continuous and exact. Proposition 4.7.3 yields an
equivalence of ∞-categories ModR A → C. According to Propositions S.10.12 and S.15.4, composition with Ω
∞
0
induces a fully faithful embedding C → Fun(ModL A , S), whose essential image C consists precisely of those
functors which are continuous and left exact. The functor co-represented by M is automatically left-exact,
and is continuous if and only if M is compact. This proves that (2) ⇔ (3).
Let j : (ModL → Fun(ModL A , S) denote the dual Yoneda embedding, so that j restricts to a map
op
A)
L,perf op 0 0
0
j : (ModA ) → C . Composing j 0 with a homotopy inverse to the equivalence ModR A → C → C , we

164
obtain a “dualization” map (ModL,perf
A )op → ModR A , which we will indicate by M 7→ M . Let D ⊆ ModA
∨ L,perf
∨
be the full subcategory spanned by those objects M such that M is perfect. We wish to show that
D = ModL,perf
A . The functor M 7→ M ∨ is exact, and ModR,perf
A is a stable subcategory of ModR A which is
closed under retracts. It follows that D is a stable subcategory of ModL A which is closed under retracts. It
now suffices to observe that A∨ ' A, so that A ∈ D.
Corollary 4.7.6. Let A be a connective A∞ -ring, and let M be a perfect left A-module. Then:

(1) The homotopy groups πm M vanish for m  0.

(2) Suppose that πm M ' ∗ for all m < k. Then πk M is a finitely presented module over π0 A.
Proof. We have an equivalence M ' lim(τ≥−n M ). Since M is compact (Proposition 4.7.2), we conclude
−→
that M is a retract of some τ≥−n M , so that πm M ' ∗ for m < −n. This proves (1). To prove (2), we
observe that the inclusion of the heart of ModA into (ModA )≥0 preserves filtered colimits, so the right adjoint
τ≤0 : (ModA )≥0 → ModA ♥ preserves compact objects. It now suffices to observe that the compact objects
in the ordinary category of π0 A-modules are precisely the finitely presented modules.
In the situation of Proposition 4.7.5, we will refer to M ∨ as a dual of M . We next prove that this notion
of duality determines an antiequivalence between the ∞-categories ModL,perf A and ModR,perf
A .
Lemma 4.7.7. Let C and D be ∞-categories, and let F : C × D → S be a bifunctor. The following conditions
are equivalent:
(1) The induced map f : C → Fun(D, S) = P(Dop ) is fully faithful, and the essential image of f coincides
with the essential image of the Yoneda embedding Dop → P(Dop ).
(2) The induced map f 0 : D → Fun(C, S) = P(Cop ) is fully faithful, and the essential image of f coincides
with the essential image of the Yoneda embedding Cop → P(Cop ).
Proof. Let
GC : C × Cop → S, GD : Dop × D → S
be the maps used in the definition of the Yoneda embeddings (see §T.5.1.3). Then (1) is equivalent to the
F
existence of an equivalence α : Dop → C such that the composition Dop × D → C × D → S is homotopic to
β op F
GD . If β is a homotopy inverse to α, then the composition C × Cop → C × D → S is homotopic to GC , which
proves (2). The converse follows by the same argument.

We will say that a functor F : C × D → S is a perfect pairing if it satisfies the hypotheses of Lemma 4.7.7.
In this case, F determines an equivalence between C and Dop , well-defined up to homotopy. The proof of
Proposition 4.7.5 yields the following:
Proposition 4.7.8. Let A be an A∞ -ring. Then the bifunctor
⊗ Ω∞
ModR,perf
A × ModL,perf
A →A Sp → S

is a perfect pairing.

Remark 4.7.9. Let A be an A∞ -ring. It follows from Proposition 4.7.8 that the ∞-category of right A-
modules is formally determined by the ∞-category of left A-modules. Namely, ModR A is equivalent to the
∞-category of Ind-objects of (ModL,perf
A )op
, where ModL,perf
A is the ∞-category of compact objects of ModLA.

165
 Let A be an A∞ -ring, and let M be a left A-module. Roughly speaking, M is perfect if it can be
obtained from the zero module through a finite process of attaching copies of A and forming retracts. This
is a very strong condition which is often violated in practice. For example, suppose that A is a commutative
Noetherian ring (viewed as a discrete A∞ -ring), and that M is a discrete A-module, such that π0 M is finitely
generated over A in the sense of classical commutative algebra. In this case, we can choose a resolution

. . . → P2 → P1 → P0 → π 0 M

where each Pi is a free A-module of finite rank. However, we cannot usually guarantee that Pn ' 0 for
n  0; in general this is possible only when A is regular ([16]). The module M is not generally perfect as
an A-module spectrum (see Example 4.7.27 below). Nevertheless, the fact that we can choose each Pi to be
of finite rank imposes a strong finiteness condition on M , which we now formulate.

Definition 4.7.10. Let C be a compactly generated ∞-category. We will say that an object C ∈ C is
compact to order n if τ≤n (C) is a compact object of τ≤n C (see §T.5.5.6 for an explanation of this notation).
We will say that C ∈ C is almost compact if it is compact to order n for all n ≥ 0.
Lemma 4.7.11. Let C be a compactly generated ∞-category, let C be an object of C, and let m ≤ n be
integers. If C is compact to order n, then C is compact to order m. If C is compact, then it is compact to
order n for all integers n.
Proof. Corollary T.[Link] implies that the ∞-categories τ≤m C and τ≤n C are compactly generated and stable
under filtered colimits in C, so that τ≤m C is stable under filtered colimits in τ≤n C. The desired conclusion
now follows from Proposition T.[Link].
Definition 4.7.12. Let A be a connective A∞ -ring. We will say that a left A-module M is almost perfect
if there exists an integer k such that for all n ≥ 0, M ∈ (ModA )≥k and is compact to order n as an object
of (ModA )≥k . We let ModaperfA denote the full subcategory of ModA spanned by the almost perfect left
A-modules.
Proposition 4.7.13. Let A be a connective A∞ -ring. Then:

(1) The full subcategory Modaperf

A ⊆ ModL
A is closed under translations and finite colimits, and is therefore
a stable subcategory of ModA .
(2) The full subcategory Modaperf
A ⊆ ModA is closed under the formation of retracts.
(3) Every perfect left A-module is almost perfect.

(4) The full subcategory (Modaperf

A )≥0 ⊆ ModA is closed under the formation of geometric realizations of
simplicial objects.
(5) Let M be a left A-module which is connective and almost perfect. Then M can be obtained as the
geometric realization of a simplicial left A-module P• such that each Pn is a free A-module of finite
rank.
Proof. Assertions (1) and (2) are obvious, and (3) follows from Proposition 4.7.2. To prove (4), it suffices to
show that the collection of compact objects of τ≤n (ModA )≥0 is closed under geometric realizations, which
follows from Lemma S.14.8.
We now prove (5). In view of Theorem S.12.8 and Remark S.12.10, it will suffice to show that M can be
obtained as the colimit of a sequence
f1 f2
D(0) → D(1) → D(2) → . . .

166
where each coker(fn )[−n] is a free A-module of finite rank; here we agree by convention that f0 denotes the
zero map 0 → D(0). The construction goes by induction. Suppose that the diagram
g
D(0) → . . . → D(n) → M

has already been constructed, and that N = ker(g) is n-connective. Part (1) implies that N is almost perfect,
so that the bottom homotopy group πn N is a compact object in the category of left π0 A-modules. It follows
that there exists a map β : Q[n] → N , where Q is a free left A-module of finite rank, and β induces a
surjection π0 Q → πn N . We now define D(n + 1) = coker(β), and construct a diagram

g0
D(0) → . . . → D(n) → D(n + 1) → M.

Using the octahedral axiom, we obtain a distinguished triangle

Q[n] → ker(g) → ker(g 0 ) → Q[n + 1],

and the associated long exact sequence of homotopy groups proves that ker(g 0 ) is (n + 1)-connective.
In particular, we conclude that for fixed m, the maps πm D(n) → πm M are isomorphisms for n  0, so
that the natural map lim D(n) → M is an equivalence of left A-modules, as desired.
−→
Using Proposition 4.7.13, we can give the following characterization of the ∞-category of (connective)
almost perfect modules over a connective A∞ -ring.
Corollary 4.7.14. Let A be a connective A∞ -ring, let C denote the full subcategory of ModA spanned by those
connective left A-modules which are connective and almost perfect, and let C0 ⊆ C denote the full subcategory
of C spanned by the objects {An }n≥0 . Let D be an arbitrary ∞-category which admits geometric realizations
for simplicial objects, and let Funσ (C, D) be the full subcategory of Fun(C, D) spanned by those functors which
preserve geometric realizations of simplicial objects. Then the restriction functor Funσ (C, D) → Fun(C0 , D)
is an equivalence of ∞-categories.
Proof. Let C0 be the smallest full subcategory of P(C0 ) which contains the essential image of the Yoneda
embedding and is stable under geometric realizations of simplicial objects, and let j : C0 → C0 be the
Yoneda embedding. Using Remark T.[Link], we conclude that composition with j induces an equivalence
Funσ (C0 , D) → Fun(C0 , D) for any ∞-category D which admits geometric realizations of simplicial objects.
In particular, the inclusion C0 ⊆ C extends (up to homotopy) to a functor F : C0 → C which commutes with
geometric realizations. To complete the proof, it will suffice to show that F is an equivalence of ∞-categories.
Using the fact that each An is a projective object of (ModA )≥0 , we deduce that F is fully faithful. Part (5)
of Proposition 4.7.13 implies that F is essentially surjective.
For a general connective A∞ -ring A, the t-structure on ModA does not restrict to a t-structure on the
full subcategory Modaperf
A . Roughly speaking, one would expect the heart of Modaperf A to be equivalent to
the ordinary category of finitely presented π0 A-modules. In general, this is not an abelian category. We can
correct this defect by introducing an appropriate hypothesis on A. We begin by recalling a definition from
classical algebra.
Definition 4.7.15. An associative ring R is left coherent if every finitely generated left ideal of R is finitely
presented (as a left R-module).
Example 4.7.16. Every left Noetherian ring is left coherent.
For completeness, we include a proof of the following classical result:
Lemma 4.7.17. Let R be a left coherent ring. Then:
(1) Every finitely generated submodule of Rn is finitely presented.

167
 (2) Every finitely generated submodule of a finitely presented left R-module is finitely presented.
(3) If f : M → N is a map of finitely presented left R-modules, then ker(f ) and coker(f ) are finitely
presented.
Proof. We first make the following elementary observations, which do not require the assumption that R is
left coherent:
(a) Suppose f : M → N is an epimorphism of left R-modules. If M is finitely generated and N is finitely
presented, then ker(f ) is finitely generated.
(b) Let 0 → M 0 → M → M 00 → 0 be a short exact sequence of left R-modules. If M 0 and M 00 are finitely
presented, then M is finitely presented.
We now prove (1) using induction on n. When n = 0 there is nothing to prove. Suppose that n > 0 and
that M ⊆ Rn is finitely generated. Form a diagram

0 / M0 /M / M 00 /0

    
0 / Rn−1 / Rn /R /0

where the vertical maps are monomorphisms. Then M 00 can be identified with a finitely generated left ideal
of R. Since R is left coherent, we conclude that M 00 is finitely presented. Using (a), we deduce that M 0 is
itself finitely generated. The inductive hypothesis now implies that M 0 is finitely presented, so that we can
use (b) to conclude that M is finitely presented.
We next prove (2). Suppose that f : M → N is a monomorphism, where N is finitely presented and M
is finitely generated. Choose an epimorphism g : Rn → N , and form a pullback diagram

K / Rn

 
M / N.

Then K can be identified with the kernel of the induced map Rn → N/M , and is therefore finitely generated.
Part (1) implies that K is finitely presented. The induced map K → M is an epimorphism, whose kernel is
isomorphic to ker(g) and is therefore finitely generated. It follows that M is finitely presented, as desired.
We now prove (3). It is clear that coker(f ) is finitely presented (this does not require the left coherence
of R). We next show that ker(f ) is finitely presented. The image of f is a finitely generated submodule of
N , and therefore finitely presented by (2). Consequently, we may replace N by im(f ), and thereby reduce to
the case where f is an epimorphism. We now apply (a) to deduce that ker(f ) is finitely generated. Invoking
(2) again, we conclude that ker(f ) is finitely presented as desired.
Definition 4.7.18. Let A be an A∞ -ring. We will say that A is left coherent if the following conditions are
satisfied:
(1) The A∞ -ring A is connective.
(2) The associative ring π0 A is left coherent (in the sense of Definition 4.7.15).
(3) For each n ≥ 0, the homotopy group πn A is finitely presented as a left module over π0 A.
Proposition 4.7.19. Let A be an A∞ -ring and M an A-module. Suppose that A is left coherent. Then M
is almost perfect if and only if the following conditions are satisfied:
(i) For m  0, πm M = 0.

168
 (ii) For all m ∈ Z, πm M is finitely presented as a π0 A-module.
Proof. Without loss of generality, we may assume that M is connective. Suppose first that M is almost
perfect. We will prove by induction on n that πn M is finitely presented as a π0 A-module. If n = 0, this
simply reduces to the observation that the compact objects of the ordinary category of left π0 A-modules are
precisely the finitely presented π0 A-modules. In particular, we can choose a finitely generated free A-module
P and a map α : P → M which induces a surjection π0 P → π0 M . Since A is coherent, the homotopy
groups πm P are finitely presented π0 A-modules. Let K = ker(α). Then K is connective by construction,
and almost perfect by Proposition 4.7.13. The inductive hypothesis implies that πi K is finitely presented
for 0 ≤ i < n.
We have a short exact sequence

0 → coker(πn K → πn P ) → πn M → ker(πn−1 K → πn−1 P ) → 0.

Using Lemma 4.7.17, we deduce that the outer terms are finitely generated, so that πn M is finitely generated.
Applying the same reasoning, we conclude that πn K is finitely generated, so that coker(πn K → πn P ) is
finitely presented. Using the exact sequence again, we conclude that πn M is finitely presented.
Now suppose that the connective left A-module M satisfies condition (ii). We will prove that M can be
obtained as the geometric realization of a simplicial left A-module P• such that each Pn is a free A-module
of finite rank. As in the proof of Proposition 4.7.13, it will suffice to show that M is the colimit of a sequence
f1 f2
D(0) → D(1) → D(2) → . . .

where each coker(fn )[−n] is a free A-module of finite rank. Supposing that the partial sequence
g
D(0) → . . . → D(n) → M

has been constructed, with the property that ker(g) is n-connective. If πn ker(g) is finitely generated as a
π0 A-module, then we can proceed as in the proof of Proposition 4.7.13. To verify this, we observe that D(n)
is almost perfect and therefore satisfies (ii) (by the first part of the proof). We now use the exact sequence

0 → coker(πn+1 D(n) → πn+1 M ) → πn ker(g) → ker(πn D(n) → πn M ) → 0

to conclude that πn ker(g) is finitely presented.

Proposition 4.7.20. Let A be a connective A∞ -ring. The following conditions are equivalent:
(1) The A∞ -ring A is left coherent.
(2) For every left A-module M , if M is almost perfect, then τ≥0 M is almost perfect.

(3) The full subcategories Modaperf

A ∩ Mod≥0 aperf
A and ModA ∩ Mod≤0 aperf
A determine a t-structure on ModA .
Proof. The implication (1) ⇒ (2) follows from the description of almost perfect modules given in Proposition
4.7.19. The equivalence (2) ⇔ (3) is obvious. We will show that (3) ⇒ (1).
Suppose that (3) is satisfied. We note that the first non-vanishing homotopy group of any almost perfect
A-module is a finitely presented module over π0 A. Applying (2) to the module A[−n], we deduce that πn A
is a finitely presented π0 A-module. To complete the proof, it suffices to show that π0 A is left coherent.
Let R = π0 A, and regard R as a discrete left A-module. Using condition (3), we deduce that R is almost
perfect. Let I ⊆ R be a finitely generated left ideal. Then I is the image (in the classical sense) of a map
f : Rn → R. Then ker(f ) (taken in the ∞-category ModA ) is almost perfect. We have a short exact sequence

0 → π0 ker(f ) → Rn → I → 0.

Since ker(f ) is almost perfect, condition (3) implies that π0 ker(f ) is finitely generated, so that I is finitely
presented as a π0 A-module. This completes the proof of (1).

169
Remark 4.7.21. Let A be a left coherent A∞ -ring, and regard Modaperf A as endowed with the t-structure
aperf
described in Proposition 4.7.20. Then ModA is right bounded and left complete, and the functor M 7→
π0 M determines an equivalence from the heart of Modaperf
A to the (nerve of the) category of finitely presented
left modules over π0 A.
We conclude with a few remarks about the interaction between finiteness and flatness conditions on
modules.

Proposition 4.7.22. Let A be a connective A∞ -ring, and let M be a connective left A-module. The following
conditions are equivalent:
(1) The left A-module M is a retract of a finitely generated free A-module.
(2) The left A-module M is flat and almost perfect.

Proof. The implication (1) ⇒ (2) is obvious. Conversely, suppose that M is flat and almost perfect. Then
π0 M is a left module over π0 A which is finitely presented and flat, and therefore projective. Using Proposition
4.6.22, we deduce that M is projective. Choose a map f : P → M , where P is a free module of finite rank
and the induced map π0 P → π0 M is surjective. Since M is projective, the map f splits, so that M is a
summand of P in hC. This proves (1).

In particular, if an almost perfect left A-module M is flat, then M is perfect. We conclude with a mild
generalization of this statement, where the flatness hypothesis is relaxed.
Definition 4.7.23. Let A be a connective A∞ -ring. We will say that a left A-module M has Tor-amplitude
≤ n if, for every discrete right A-module N , the homotopy groups πi (N ⊗A M ) vanish for i > n. We will
say that M is of finite Tor-amplitude if it has Tor-amplitude ≤ n for some integer n.

Remark 4.7.24. In view of Theorem 4.6.19, a connective left A-module M has Tor-amplitude ≤ 0 if and
only if M is flat.
Proposition 4.7.25. Let A be a connective A∞ -ring.
(1) If M is a left A-module of Tor-amplitude ≤ n, then M [k] has Tor-amplitude ≤ n + k.

(2) Let
M 0 → M → M 00 → M 0 [1]
be a distinguished triangle of left A-modules. If M 0 and M 00 have Tor-amplitude ≤ n, then so does M .
(3) Let M be a left A-module of Tor-amplitude ≤ n. Then any retract of M has Tor-amplitude ≤ n.
(4) Let M be an almost perfect left module over A. Then M is perfect if and only if M has finite Tor-
amplitude.

Proof. The first three assertions follow immediately from the exactness of the functor N 7→ N ⊗A M . It
follows that the collection left A-modules of finite Tor-amplitude is stable under retracts and finite colimits,
and contains the module A[n] for every integer n. This proves the “only if” direction of (4). For the converse,
let us suppose that M is an almost perfect of finite Tor-amplitude. We wish to show that M is perfect.
We first apply (1) to reduce to the case where M is connective. The proof now goes by induction on the
Tor-amplitude n of M . If n = 0, then M is flat and we may conclude by applying Proposition 4.7.22. We
may therefore assume n > 0.
Since M is almost perfect, there exists a free left A-module P of finite rank and a distinguished triangle
f
M 0 → P → M → M 0 [1]

170
where f is surjective. To prove that M is perfect, it will suffice to show that P and M 0 are perfect. It is
clear that P is perfect, and it follows from Proposition 4.7.13 that M 0 is almost perfect. Moreover, since f
is surjective, K is connective. We will show that M 0 is of Tor-amplitude ≤ n − 1; the inductive hypothesis
will then imply that M is perfect, and the proof will be complete.
Let N be a discrete right A-module. We wish to prove that πk (N ⊗A K) ' 0 for k ≥ n. Since the functor
N ⊗A • is exact, we obtain for each an exact sequence of homotopy groups

πk+1 (N ⊗A M ) → πk (N ⊗A M 0 ) → πk (N ⊗A P ).

The left entry vanishes in virtue of our assumption that M has Tor-amplitude ≤ n. We now complete the
proof by observing that πk (N ⊗A P ) is a finite direct sum of copies of πk N , and therefore vanishes because
k ≥ n > 0 and N is discrete.
Remark 4.7.26. Let A be a connective A∞ -ring, and let C be the smallest stable subcategory of ModA
which contains all finitely generated projective modules. Then C = Modperf A . The inclusion C ⊆ ModA
perf
perf
is obvious. To prove the converse, we must show that every object M ∈ ModA belongs to C. Invoking
Corollary 4.7.6, we may reduce to the case where M is connective. We then work by induction on the
(necessarily finite) Tor-amplitude of M . If M is Tor-amplitude zero, then M is flat and the desired result
follows from Proposition 4.7.22. In the general case, we choose a finitely generated free A-module P and a
map f : P → M which induces a surjection π0 P → π0 M (which is possible in view of Corollary 4.7.6). As
in the proof of Proposition 4.7.25, we may conclude that ker(f ) is a connective perfect module of smaller
Tor-amplitude than M , so that the ker(f ) ∈ C by the inductive hypothesis. Since P ∈ C and C is stable
under the formation of cokernels, we conclude that M ∈ C as desired.

Example 4.7.27. Let A be a discrete A∞ -ring, let Mod− A ⊆ ModA be the full subcategory consisting of
bounded-above objects, and let let M ∈ Mod− A . Using an inverse to the functor θ of Proposition 4.4.7, we
can identify any M ∈ Mod− A with a (bounded above) complex P• of projective left π0 A-modules. It follows
from Remark 4.7.26 that M is perfect if and only if P• can be chosen to have only finitely many terms, each
of which is finitely generated over π0 A.

171
References
[1] Arone, G. and M. Ching. Operads and Chain Rules for the Calculus of Functors. Preprint.

[2] Artin, M. Théorie des topos et cohomologie étale des schémas. SGA 4. Lecture Notes in Mathematics
269, Springer-Verlag, Berlin and New York, 1972.
[3] Artin, M. and B. Mazur. Étale Homotopy. Lecture Notes in Mathematics 100, Springer-Verlag, Berlin
and New York, 1969.

[4] Basterra, M. André-Quillen cohomology of commutative S-algebras. Journal of Pure and Applied Algebra
144 (1999) no. 2, 111-143.
[5] Behrend, K. and B. Fantechi. The intrinsic normal cone. Inventiones Mathematicae 128 (1997) no. 1,
45-88.
[6] Beilinson, A. , Bernstein, J. and P. Deligne. Faisceaux pervers. Asterisuqe 100, Volume 1, 1982.

[7] Bergner, J.E. A Model Category Structure on the Category of Simplicial Categories. Transactions of the
American Mathematical Society 359 (2007), 2043-2058.
[8] Bergner, J.E. A survey of (∞, 1)-categories. Available at [Link]/0610239
[9] Bergner, J.E. Rigidification of algebras over multi-sorted theories. Algebraic and Geometric Topoogy 7,
2007.
[10] Bergner, J.E. Three models for the homotopy theory of homotopy theories, Topology 46 (2007), 397-436.
[11] Bosch, Guntzer, U., and R. Remmert, R. Non-Archimedean Analysis: a Systematic Approach to Rigid
Analytic Geometry. Springer-Verlag, Berlin and Heidelberg, 1984.

[12] Bousfield, A.K. and D.M. Kan. Homotopy limits, completions, and localizations. Lecture Notes in Math-
ematics 304, Springer-Verlag, 1972.
[13] Cisinski, D-C and I. Moerdijk. Dendroidal sets as models for homotopy operads. Available for download
as arXiv:0902.1954v1.

[14] Dugger, D. Combinatorial model categories have presentations. Advances in Mathematics 164, 2001,
177-201.
[15] Eilenberg, S. and N.E. Steenrod. Axiomatic approach to homology theory. Proc. Nat. Acad. Sci. U.S.A.
31, 1945, 117-120.
[16] Eisenbud, D. Commutative algebra. Springer-Verlag, New York, 1995.

[17] Elmendorf, A.D., Kriz, I. , Mandell, M.A., and J.P. May. Rings, modules and algebras in stable homotopy
theory. Mathematical Surveys and Monographs 47, American Mathematical Society, 1997.
[18] Fulton, W. Algebraic curves. W.A. Benjamin, Inc., New York, 1969.
[19] Goerss, P. and J.F. Jardine. Simplicial Homotopy Theory. Progress in Mathematics, Birkhauser, Boston,
1999.
[20] Goodwillie, T. Calculus I: The rst derivative of pseudoisotopy theory. K-Theory 4 (1990), no. 1, 127.
[21] Goodwillie, T. Calculus. II. Analytic functors. K-Theory 5 (1991/92), no. 4, 295332.
[22] Goodwillie, T. Calculus III: Taylor Series. Geometry and Topology, Volume 7 (2003) 645-711.

172
[23] Grauert, H. and R. Remmert. Theory of Stein Spaces. Springer-Verlag, Berlin Heidelberg, 2004.

[24] Gunning, R. and H. Rossi. Analytic functions of several complex variables. Prentice-Hall, Englewood
Cliffs, N.J, 1965.
[25] Hatcher, A. Algebraic Topology. Cambridge University Press, 2002.
[26] Hook, E.C. Equivariant cobordism and duality. Transactions of the American Mathematical Society 178
(1973) 241-258.
[27] Hovey, M. Model Categories. Mathematical Surveys and Monographs 63, AMS, Providence, RI, 1999.
[28] Hovey, M., Shipley, B. and J. Smith. Symmetric spectra. Journal of the American Mathematical Society
13, 2000, no. 1, 149-208.

[29] Illusie, L. Complexe cotangent et déformations I. Lecture Notes in Mathematics 239, Springer-Verlag,
1971.
[30] Illusie, L. Complexe cotangent et déformations II. Lecture Notes in Mathematics 283, Springer-Verlag,
1972.

[31] Joyal, A. Notes on quasi-categories.

[32] Joyal, A. Simplicial categories vs. quasi-categories.
[33] Joyal, A. and M. Tierney. Quasi-categories vs. Segal Spaces. Preprint available at [Link]/0607820.
[34] Kerz, M. The complex of words and Nakaoka stability. Homology, Homotopy and Applications, volume
7(1), 2005, pp. 77-85.
[35] Klein, J. and J. Rognes. A chain rule in the calculus of homotopy functors. Geom. Topol. 6 (2002),
853887.
[36] Knutson, D. Algebraic spaces. Lecture Notes in Mathematics 203, Springer-Verlag, 1971.

[37] Laplaza, M. Coherence for distributivity. Coherence in categories, 29-65. Lecture Notes in Mathematics
281, Springer-Verlag, 1972.
[38] Laumon, G. and L. Moret-Bailly. Champs algebriques. Springer-Verlag, 2000.
[39] Lazard, Daniel. Sur les modules plats. C.R. Acad. Sci. Paris 258, 1964, 6313-6316.

[40] Lurie, J. Higher Topos Theory. Available for download at [Link] lurie/ .
[41] Lurie, J. Derived Algebraic Geometry I: Stable ∞-Categories. Available for download.
[42] Lurie, J. Derived Algebraic Geometry II: Noncommutative Algebra. Available for download.
[43] Lurie, J. Derived Algebraic Geometry III: Commutative Algebra. Available for download.

[44] Lurie, J. Derived Algebraic Geometry IV: Deformation Theory. Available for download.
[45] Lurie, J. Derived Algebraic Geometry V: Structured Spaces. Available for download.
[46] Lurie, J. Derived Algebraic Geometry VI: E[k]-Algebras. Available for download.

[47] Lurie, J. Derived Algebraic Geometry VII: Spectral Schemes. In preparation.

[48] Lurie, J. (∞, 2)-Categories and the Goodwillie Calculus I. Available for download.

173
[49] Lurie, J. (∞, 2)-categories and the Goodwillie Calculus II. In preparation.
[50] Lurie, J. Elliptic curves in spectral algebraic geometry. In preparation.
[51] Lurie, J. Toric varieties, elliptic cohomology at infinity, and loop group representations. In preparation.
[52] MacLane, S. Categories for the Working Mathematician. Second edition. Graduate Txts in Mathematics,
5. Springer-Verlag, New York, 1998.
[53] May, J.P. The Geometry of Iterated Loop Spaces.
[54] Mitchell, B. A quick proof of the Gabriel-Popesco theorem. Journal of Pure and Applied Algebra 20
(1981), 313-315.
[55] Neeman, A. Triangulated categories. Annals of Mathematics Studies, 148. Princeton University Press,
2001.
[56] Quillen, D. Homotopical Algebra. Lectures Notes in Mathematics 43, SpringerVerlag, Berlin, 1967.
[57] Rezk, C. A model for the homotopy theory of homotopy theory. Transactions of the American Mathe-
matical Society 35 (2001), no. 3, 973-1007.
[58] Rosicky, J. On Homotopy Varieties. Advances in Mathematics 214, 2007 no. 2, 525-550.
[59] Schwede, S. Spectra in model categories and applications to the algebraic cotangent complex. Journal of
Pure and Applied Algebra 120 (1997), pp. 77-104.
[60] Schwede, S. and B. Shipley. Algebras and Modules in Monoidal Model Categories. Proceedings of the
London Mathematical Society (80) 2000, 491-511.
[61] Schwede, S. and B. Shipley. Stable model categories are categories of modules. Topology 42, 2003, no. 1,
103-153.
[62] Serre, Jean-Pierre. Local algebra. Springer-Verlag, 2000.
[63] Shipley, B. A Convenient Model Category for Commutative Ring Spectra. Homotopy theory: relations
with algebraic geometry, group cohomology, and algebraic K-theory. Contemp. Math. volume 346 pp.
473-483, American Mathematical Society, Providence, RI, 2004.
[64] Spivak, D. Quasi-smooth Derived Manifolds. PhD dissertation.
[65] Srinivas, V. Algebraic K-Theory. Birkhauser, Boston, 1993.
[66] Toën, B. Champs affines. Available for download: [Link]/0012219.
[67] Toën, B. Vers une axiomatisation de la théorie des catégories supériures. K-theory 34 (2005), no. 3,
233-263.
[68] Toën, B. and G. Vezzosi. From HAG to DAG: derived moduli stacks. Available for download:
[Link]/0210407.
[69] Toën, B. and G. Vezzosi. Algebraic geometry over model categories. Available for download:
[Link]/0110109.
[70] Toën, B. and G. Vezzosi. “Brave New” Algebraic Geometry and global derived moduli spaces of ring
spectra. Available for download: [Link]/0309145.
[71] Toën, B. and G. Vezzosi. Segal topoi and stacks over Segal categories. Available for download:
[Link]/0212330.

174
[72] Toën, B. and G. Vezzosi. A remark on K-theory and S-categories. Topology 43, No. 4 (2004), 765-791

[73] Verity, D. Weak complicial sets, a simplicial weak omega-category theory. Part I: basic homotopy theory.
[74] Verity, D. Weak complicial sets, a simplicial weak omega-category theory. Part II: nerves of complicial
Gray-categories.
[75] Weibel, C. An Introduction to Homological Algebra. Cambridge University Press, 1995.

175