Patrick Lei

All notes

Topological cyclic homology

Lectures by Thomas Nikolaus and Achim Krause, YouTube, Spring 2025.

Abstract

These are notes taken live while watching the YouTube lectures about topological cyclic homology by Thomas Nikolaus and Achim Krause. Note that we use homological grading throughout because the lecturers are homotopy theorists.

Lectures on YouTube by Achim Krause and Thomas Nikolaus
Notes by Patrick Lei
  WHAT’S A PUBLISHER?

Motivation: trace methods

One motivation to study topological cyclic homology is its relation to crystalline cohomology and syntomic cohomology in arithmetic geometry. However, this is very involved, so we will not discuss it in depth. We will instead discuss the original motivation, which are trace methods in algebraic K-theory.

Let R be a ring. There is an invariant, called algebraic K-theory, which produces groups K∗⁢(R) for all ∗≥0. These are very important, but are almost impossible to compute or to understand structurally. For example, even K∗⁢(ℤ) is not very well-understood. They are known in all odd degrees, conjectured to vanish in degrees divisible by 4 (this is equivalent to the Kummer-Vandiver conjecture in number theory).

If you are a topologist, then one motivation is to study whether a retract X of a finite CW complex is itself a finite CW complex. The obstruction to this lies in

Another motivation is that by work of Whitehead and others (the s-cobordism theorem), the obstruction for a cobordism to be trivial lies in an algebraic K-theory group. There is now a higher version of this theorem which relates diffeomorphism groups to algebraic K-theory. A third motivation comes from the Baum-Connes conjecture, which has other applications in geometric topology.

For other applications, the group K0 was first invented by Grothendieck to give a statement of the Grothendieck-Riemann-Roch theorem, and algebraic K-theory is also related to special values of L-functions.

Definition.

For a ring R, the group K0⁢(R) is the Grothendieck group of isomorphism classes of finitely generated projective R-modules under the operation of direct sum.

Higher K-groups are defined as homotopy groups of the space obtained by group completing the category of finitely-generated projective R-modules. This can be made precise using higher algebra and actually produces a spectrum, and then we can take its homotopy groups. We can see why this is so complicated.

Example.

It is easy to see that K0⁢(field)=ℤ, but of course the higher K-groups are very complicated. Quillen computed them for finite fields, but if we leave this case this becomes extremely difficult.

Approximation of algebraic K-theory

We will attempt to approximate algebraic K-theory using more algebraic invariants. There is a Dennis trace map to Hochschild homology. There are refinements of this which fit into a diagram

Slogan.

The cyclotomic trace is often close to an isomorphism.

There are also relative K-groups K∗⁢(R,I) for any ideal I⊆R, which is simply the homotopy fiber of K⁡(R)→K⁡(R/I). There are also relative 𝖳𝖢 groups, which are defined in the same way. We can also define groups with coefficients like K∗⁢(R,ℤp) and 𝖳𝖢∗⁢(R,ℤp).

Theorem (Dundas-Goodwillie-McCarthy, Clausen-Matthew-Morrow).
  1. 1.

    If I⊆R is a nilpotent ideal, then

    is an isomorphism.

  2. 2.

    If R is commutative and I-complete, the same conclusion holds with p-adic coefficients. In other words, we have

  3. 3.

    If R is p-complete, then

Why the trace map is a trace

Let k be a field, R be a k-algebra, and P be a finitely generated projective right R-module. Then we have

We would attempt to go to R by an evaluation map

but this doesn’t actually land in R. Instead, we see that

The best we can do is to consider the quotient

as an abelian group, so in conclusion we have

Note that an R-R-bimodule is equivalently an R⊗kRop-module. Then, we in fact have

Deriving this expression, we will see later that

The Dennis trace will refine this trace in the sense that on K0, it will send a projective module P to the trace of the identity endomorphism.

1 ∞-categories and higher algebra

In this chapter, we give a user’s guide to ∞-categories and the homotopy-enriched algebra of spectra following the higher algebra playlist.

1.1 ∞-categories

Infinity-categories are a simultaneous generalization of both ordinary categories and of the homotopy theory of spaces which form a natural home for derived phenomena. We will not aim to cover the entire theory but instead to give an idea of how to work with and think about ∞-categories.

Convention 1.1.0.1.

We will resolve size issues by assuming the existence of a Grothendieck universe 𝒰. Elements of 𝒰 will be called small sets. We will take categories to have large object sets and large morphism sets. Therefore, for us we will take 𝖲𝖾𝗍 to mean the category of small sets.

1.1.1 ∞-categories

Definition 1.1.1.1.

An ∞-category is is a (large) simplicial set 𝒞 with the property that any diagram

admits a filling for all 0<i<n. Here, Λin is the i-th horn given by taking the boundary of Δn and removing the facet opposite to the i-th vertex.

A functor 𝒞→𝒟 is a map of simplicial sets.

It is not clear why this definition should be related to categories at all, but we will construct an example of an ∞-category from an ordinary category.

Example 1.1.1.2.

Let 𝒞 be an ordinary category. Define the nerve of 𝖢 to be the simplicial set given by

where [n] is the poset given by 0≤1≤⋯≤n.

Exercise 1.1.1.3.
  1. 1.

    Show that N⁢𝒞 is an ∞-category;

  2. 2.

    Show that functors N⁢𝒞→N⁢𝒟 are in bijection with functors 𝒞→𝒟.

Now let 𝒞 be an ∞-category. We will call 𝒞0 the set of objects and 𝒞1 the set of morphisms. Note that it is not clear what the composition of two morphisms should be. For any f∈𝒞1, we will view f as a morphism

Definition 1.1.1.4.

Two morphisms f,g:a→b are equivalent if there exists a 2-simplex σ∈𝒞2 of the form

In this case, we write f≃g.

This notion is extremely boring for the nerve of a 1-category. It is more interesting in the next example.

Example 1.1.1.5.

Let 𝒞 be a Kan complex. This means that all horns can be filled in, so in particular 𝒞 is an ∞-category. For example, if X is a topological space, then the singular simplicial set 𝖲𝗂𝗇𝗀⁡(X) is a Kan complex. Here, objects are points of X, morphisms are paths in X, and equivalence is simply homotopy of paths relative to the endpoints. If we pass to equivalence classes, we obtain the fundamental groupoid of X, so we can think of 𝖲𝗂𝗇𝗀⁡(X) as being an ∞-categorical version of the fundamental groupoid. Note here that composition of paths is only associative up to homotopy.

Definition 1.1.1.6.

Let f:a→b and g:b→c be morphisms. A composition of f and g is a 2-simplex σ of the form

In this case, we will write h≃g∘f.

Exercise 1.1.1.7.

Check that two different choices of compositions, h≃g∘f and h′≃g∘f, are equivalent.

The non-existence of uniqueness may seem like a defect, but if we consider coproducts in ordinary categories, being unique up to unique isomorphism is enough because it allows us to not have to think about a coproduct functor being associative (or maybe only in a weak sense).

Definition 1.1.1.8.

Let f:a→b be a morphism. It is an equivalence if there exists a morphism g:b→a such that ida≃g∘f and idb≃f∘g.

Recall that the functors between ordinary categories form a category. We will upgrade this to ∞-categories.

Definition 1.1.1.9.

Let 𝒞 and 𝒟 be simplicial sets. Then we define the simplicial set 𝖥𝗎𝗇⁡(𝒞,𝒟) by the formula

Proposition 1.1.1.10.

Let 𝒞 be an arbitrary simplicial set and 𝒟 be an ∞-category. Then 𝖥𝗎𝗇⁡(𝒞,𝒟) is an ∞-category.

Note that a morphism η:f→g is a functor 𝒞×Δ1→𝒟 which restricts to f and g on the boundaries. We will call these natural transformations. A natural transformation η is an equivalence in 𝖥𝗎𝗇⁡(𝒞,𝒟) if and only if it is objectwise an equivalence in 𝒟.

Definition 1.1.1.11.

A functor f:𝒞→𝒟 is an equivalence if there exists a functor g:𝒟→𝒞 and natural equivalences f∘g≃id𝒟 and g∘f≃id𝒞. In this case, we will write 𝒞≃𝒟.

We would now like to discuss something that is like the fact that equivalences of ordinary categories are fully faithful and essentially surjective. Unfortunately, we don’t really have the machinery to do this yet.

Definition 1.1.1.12.

Let a,b∈𝒞 be two objects in an ∞-category. We define the mapping space to be the simplicial set (in fact a Kan complex) which is the pullback in the diagram

Note that the 0-simplices in this mapping space are simply morphisms f:a→b and that 1-simplices are equivalences. Thus π0⁢𝖬𝖺𝗉𝒞⁢(a,b) is the set of equivalence classes of morphisms a→b.

Exercise 1.1.1.13.
  1. 1.

    Check that in N⁢𝒞 for a 1-category 𝒞, mapping spaces are discrete and agree with morphism sets, and that the composition map recovers ordinary composition.

  2. 2.

    Check that in 𝖲𝗂𝗇𝗀⁡(X), the mapping space

    is homotopy equivalent to the space of paths from a to b in X and that the composition map comes from path composition.

Definition 1.1.1.14.

A functor f:𝒞→𝒟 between ∞-categories is fully faithful if for any pair of objects a,b∈𝒞, the map

is a homotopy equivalence.

Definition 1.1.1.15.

A functor f:𝒞→𝒟 between ∞-categories is essentially surjective if for all d∈𝒟, there exists some c∈𝒞 and an equivalence d≃f⁡(c).

Proposition 1.1.1.16.

A functor f:𝒞→𝒟 between ∞-categories is an equivalence if and only if f is fully faithful and essentially surjective.

Recall the notion of a full subcategory. We will define an ∞-categorical version of this.

Definition 1.1.1.17.

Let S⊂𝒞0. We define the full subcategory 𝒞S⊆𝒞 to consist of all simplices with vertices in S.

Note that if we first saturate S under equivalence of objects to form S⊆S¯, we obtain an equivalence

We will now discuss a way to upgrade compositions to mapping spaces. Consider the pullback

which is a triple mapping space.

Lemma 1.1.1.18.

The triple mapping space 𝖬𝖺𝗉𝒞⁢(a,b,c) is a Kan complex and fits in a diagram

induced by the horn-filling

A choice of a homotopy inverse S gives us a composition map

This seems to depend on the choice of S, but it is unique up to unique homotopy. In fact, the space of all possible homotopy inverses is contractible.

Similarly, we can define 𝖬𝖺𝗉𝒞⁢(a,b,c,d), which parameterizes the associativity of composition. We can compare this to Segal categories, where data like this is an axiom and gives us categories weakly enriched in spaces. In our approach, we get this for free.

Until now, we have seen two kinds of ∞-categories: (nerves of) ordinary categories and Kan complexes (where all morphisms are invertible, sometimes called ∞-groupoids). We will now discuss another example, the ∞-category of spaces, which is the prototypical example of an ∞-category.

Recall that the category of Kan complexes is enriched in simplicial sets. This enrichment actually lands in Kan complexes, so we would like to form an ∞-category.

Construction 1.1.1.19.

For a finite, nonempty, totally ordered set J, define a simplicially enriched category

where objects are elements of J and for any i,j∈J, the mapping simplicial set is given by

Here, composition is induced by union of subsets of J.

Definition 1.1.1.20.

Define a simplicial set 𝒮 by the formula

The simplicies look like the follows:

  • •

    0-simplices are Kan complexes;

  • •

    1-simplices are simply morphisms of Kan complexes;

  • •

    2-simplices correspond to diagrams

Theorem 1.1.1.21.

This simplicial set 𝒮 is an ∞-category, and for all X,Y, there is a homotopy equivalence

Definition 1.1.1.22.

For an arbitrary simplicially enriched category 𝒞, we can define the homotopy-coherent nerve NΔ⁢(𝒞) by the formula

Proposition 1.1.1.23.

If 𝒞 is enriched in complexes, then NΔ⁢(𝒞) is an ∞-category and we have equivalences

1.1.2 Limits

We will again neglect to develop the full theory in favor of giving an idea of how to work with limits.

Notation 1.1.2.1.

From now on, we will denote the ∞-category of functors from a simplicial set I to an ∞-category 𝒞 by 𝒞I≔𝖥𝗎𝗇⁡(I,𝒞).

In this subsection, we will let I be a small simplicial set, 𝒞 be an ∞-category, and F:I→𝒞 be a functor.

Definition 1.1.2.2.

A cone over F is a pair consisting of an object y∈𝒞, together with a natural transformation

where

is the constant functor on y.

Construction 1.1.2.3.

Let (y,η) be a cone over F and x∈𝒞. We will construct a map

as the composite

This is well-defined up to a contractible choice.

Definition 1.1.2.4.

A cone (y,η) over F is a limit cone (or just limit) in 𝒞 if for all x∈𝒞, the map

is a homotopy equivalence. We will denote this by

Exercise 1.1.2.5.

If I→𝒞 is a functor of ordinary categories, check that limits of N⁢I→N⁢𝒞 correspond exactly to ordinary limits of I→𝒞.

Example 1.1.2.6.

Let I=∅. Then 𝖥𝗎𝗇⁡(I,𝒞)≃pt. Therefore, every y∈𝒞 is a cone, and it is a limit if

for all x∈𝒞. In other words, y must be a terminal object and we will denote it by ∗ or pt.

Example 1.1.2.7.

If I is a discrete set, then

Therefore, a functor I→𝒞 is simply a sequence {yi}i∈I, and a cone is an object y∈𝒞 with maps {πi:y→yi}i∈I. This is a limit cone if for all objects x∈𝒞, we have an equivalence

For example, in 𝒮, the products are given by products of Kan complexes.

Lemma 1.1.2.8.

Any two limits y,y′ of F are canonically equivalent.

Proof.

The space of maps y→y′ is simply

We define a map f:y→y′ by simply pulling back η. By construction, we have η′∘f≃η. Doing this in the other direction, we have g:y′→y such that η∘g≃η′. We see that

which implies that g∘f≃id. Applying this argument to f∘g yields the desired result. □

Remark 1.1.2.9.

We can strengthen the previous lemma to the statement that the ∞-category of limit cones over F:I→𝒞 is either empty or trivial.

Remark 1.1.2.10.

We have allowed I to be an arbitrary simplicial set, which may lack the lifting properties in the definition of an ∞-category. However, we can form a larger simplicial set I⊂I′ by gluing in fillers for horns which do not have fillers, and then we actually have an equivalence

which is even a Kan fibration.

We will now turn to some examples which have some coherence issues. The first are fiber products. Here, let

Then a functor F:I→C is simply a diagram

in 𝒞.

Lemma 1.1.2.11.

A natural transformation ca→F for some object a∈𝒞 is, up to equivalence, given by the data of maps i:a→b and j:a→c together with an equivalence

in 𝒞.

Proof.

Because I=Δ1⊔Δ0Δ1 is a pushout, the functor category

is a fiber product. Therefore a functor ca→F is given by diagrams

Gluing the two edges involving f together, we are done. □

To actually work with limits, we need results about the functoriality and compatibility of mapping spaces.

Lemma 1.1.2.12 (Workhorse lemma).

Let 𝒞 be a 𝖪𝖺𝗇-enriched category, F:I→NΔ⁢(𝒞) be a functor, z∈𝒮 be a Kan complex, and x∈𝒞 be an object. Then the internal hom

Furthermore, if 𝒞=𝒮, then this space is also equivalent to

Proposition 1.1.2.13.

A cone (y,η) over F in 𝒮 is a limit if and only if for all x∈𝒮, the map

is an isomorphism.

Proof.

We have an equivalence

for all z using the workhorse lemma. □

If we have a square

in 𝒮, it is a pullback if maps x→a are up to homotopy the same as maps x→b and x→c together with a homotopy between x→c→d and x→b→d.

Exercise 1.1.2.14.

Check that a pullback of a diagram

in 𝒮 can be computed as the ordinary limit of the diagram

of simplicial sets.

Proposition 1.1.2.15.

For any functor F:I→𝒮, the space

is the limit of F. In particular, 𝒮 has all (small) limits.

Proof.

We need to check that

This is true by using the exponential law to put cx×pt=cx on the RHS. To construct the cone, we simply put the identity map on the LHS. □

Example 1.1.2.16.

Let X be any simplicial set and let F=cx be the constant functor on a space x∈𝒮. Then

This tells us a lot more than a 1-categorical limit. For example, it sees all of the higher homotopy groups of x by inputting spheres.

Example 1.1.2.17.

Let I′=N⁢(ℕ,≤)op. This is an ∞-category. We will consider the simplicial set

Therefore, functors I→𝒮 are simply sequences

Then the limit of this diagram in 𝒮 is simply the mapping telescope.

We will now consider an arbitrary 𝖪𝖺𝗇-enriched category 𝒞, a functor F:I→NΔ⁢(𝒞), and an object x∈𝒞. We will consider the internal Hom functor

which induces a mapping space

Theorem 1.1.2.18.

A cone (y,η) over F is a limit cone if and only if the induced cone

is a limit cone in 𝒮 for all x∈𝒞. In fact, there is an equivalence

Remark 1.1.2.19.

We could try to take this as a definition, but it has two disadvantages:

  1. 1.

    We would have to define limits in spaces;

  2. 2.

    We would have issues with the functoriality of mapping spaces.

Proof.

Being a limit cone in 𝒞 is the same thing as an equivalence

This is now equivalent to having an equivalence

for any space z. But now this is equivalent to checking that

which is the condition about the induced cones. □

Remark 1.1.2.20.

In our definition, a cone over F is a functor

There is in fact a “smaller” model for this quotient called I⊲=Δ0∗I. Then 𝖬𝖺𝗉𝒞I⁢(cx,F) is the pullback

This model is the one that is used by Lurie and Joyal.

1.1.3 Colimits

Colimits are formally dual to limits. If 𝒞 is an ∞-category, there is an opposite ∞-category 𝒞op defined by

Then we can define the colimit F:I→𝒞 as the limit of the functor

Explicitly, we first define a cone under F to be an object y∈𝒞 together with a natural transformation η:F→cy. This is a colimit cone if for any z∈𝒞, the map

is an equivalence.

Proposition 1.1.3.1.

Extending an object y∈𝒮 to a colimit cone over a fixed F:I→𝒮 is equivalent to providing a natural isomorphism

where naturality is in z.

Example 1.1.3.2.

If I=∅, then a colimit over I is an initial object, which will be denoted by ∅. It is given by the property that

In 𝒮, this is actually given by the empty set.

Example 1.1.3.3.

If I is discrete, then a functor is F={yi}i∈I, and the colimit is called a coproduct and denoted by

It is characterized by the property that

In 𝒮, it is given by the disjoint union.

Example 1.1.3.4.

Suppose that

Then a functor is simply a diagram

The colimit, called a pushout and denoted b⊔ac, is characterized by the universal property that for all z∈𝒞, we have

where we are taking the homotopy fiber product in 𝒮.

We will now consider pushouts in the category of spaces. Given a diagram

of spaces, we form the simplicial set

which is usually called the double mapping cylinder. It corresponds to the diagram

Then we will replace this by a Kan complex d, and the resulting diagram

is a pushout.

Exercise 1.1.3.5.

Check that this is actually true.

We have seen that the category of spaces has all coproducts and pushouts, and we now want to show that it has all (small) colimits. One way is to write this is to consider

which is a formula for the colimit. Instead, we will show that any ∞-category with all coproducts and pushouts has all colimits.

Example 1.1.3.6.

A coequalizer is a colimit over

a coequalizer is simply a universal c which fits into a diagram

and is in fact the pushout

Exercise 1.1.3.7.

Prove this.

Example 1.1.3.8.

A pushout

is actually a coequalizer

We will now discuss sequential colimits. Here, we consider

Functors I→𝒞 and colimits are simply sequences

This is actually a coequalizer

Definition 1.1.3.9.

A simplicial set I is called finite if it has only finitely many non-degenerate simplices. Colimits over finite simplicial sets will be called finite colimits.

Theorem 1.1.3.10.

For an ∞-category 𝒞, the following are equivalent:

  1. 1.

    𝒞 admits an initial object and all pushouts;

  2. 2.

    𝒞 admits all finite coproducts and all coequalizers;

  3. 3.

    𝒞 admits all finite colimits.

Proof, but not really.

First, we note that coproducts a⊔b are pushouts

In the other direction, the initial object is the empty coproduct.

Now, we need to write arbitrary finite colimits using the basic colimits (coproducts, coequalizers, and pushouts). If I is a finite simplicial set, we write

and consider the attaching construction of I, which is

For any F:I→𝒞, we have an equivalence

Here, we use the fact that

This allows us to prove the existence of finite coproducts by inducting on the dimension of I, where the base case is simply the case of coproducts. □

Exercise 1.1.3.11.

Let I be a poset with a terminal object. Show that colimits over F:I→𝒞 are simply given by evaluating F at the terminal object.

We know already that the category of spaces has all finite colimits since we already wrote down finite coproducts and pushouts. However, we want all colimits.

Theorem 1.1.3.12.

Suppose 𝒞 admits all coproducts. Then the following are equivalent:

  1. 1.

    𝒞 admits all colimits;

  2. 2.

    𝒞 admits all pushouts;

  3. 3.

    𝒞 admits all coequalizers;

  4. 4.

    𝒞 admits all geometric realizations, or in other words colimits indexed over N⁢Δop.

Proof.

We already know that pushouts and coequalizers are equivalent. We now need to write arbitrary colimits using pushouts and coequalizers. We now write

Then we write

and sequential colimits are simply coequalizers.

We now want to construct pushouts from Δop. Here, we realize the pushout

as a colimit of the simplicial diagram

This proves the result. □

Corollary 1.1.3.13.

The ∞-category 𝒮 of spaces has all colimits.

We will discuss limits and colimits freely in some other ∞-categories.

Definition 1.1.3.14.

The ∞-category 𝒮∗ of pointed spaces is the homotopy-coherent nerve

where 𝖪𝖺𝗇∗ is the simplicially enriched category of pointed Kan complexes.

Definition 1.1.3.15.

The ∞-category 𝖢𝖺𝗍∞ of ∞-categories is given by the homotopy-coherent nerve

where 𝖢𝖺𝗍∞Δ is the simplicially enriched category where objects are ∞-categories and morphisms

are given by the maximal Kan complexes.

Theorem 1.1.3.16.

The ∞-categories 𝖢𝖺𝗍∞ and 𝒮∗ have all limits and colimits.

1.1.4 Derived catgories

We will construct an ∞-categorical refinement of the derived category 𝒟⁡(𝒜) of an abelian category 𝒜, which encode Ext groups in the mapping spaces. We will first describe an ∞-category where equivalence classes of morphisms are chain homotopy classes, and then we will use Dwyer-Kan localizations to localize at quasi-isomorphisms. The resulting category, 𝒟⁡(𝒜), will be characterized by the following:

  • •

    Objects will be (unbounded) chain complexes in 𝒜;

  • •

    Equivalence classes of morphisms will be given by the usual derived Hom.

Definition 1.1.4.1.

For an abelian category 𝒜, denote the 1-category of chain complexes by 𝖢𝗁⁡(𝒜). It is enriched over 𝖢𝗁⁡(ℤ) and is therefore a dg category.

Construction 1.1.4.2 (Dold-Kan).

There is a lax monoidal functor

Then we form the composite

Each functor in this composition is lax monoidal, so we can turn every dg category into a simplicially enriched category by applying K.

The functor K has two nice properties. First, it produces Kan complexes, not just simplicial sets. The homotopy groups are given by the non-negative homology groups of the complex, or in other words

for all n≥0.

Definition 1.1.4.3.

For a dg category 𝒞, define the dg-nerve of 𝒞 to be

where 𝒞Δ, where 𝒞Δ is obtained from 𝒞 by applying K to all Hom complexes.

Note that this produces an ∞-category.

Definition 1.1.4.4.

For an abelian category 𝒜, define

This actually has some very nice properties with respect to limits and colimits regardless of the abelian category 𝒜.

Proposition 1.1.4.5.

The ∞-category 𝒦⁡(𝒜) has all finite limits and colimits.

Proof.

First, note that 0∈𝒦⁡(𝒜) is the initial and terminal object. In particular, we see that

and that

for any E∈𝒦⁡(𝒜).

Next, let C,D∈𝒦⁡(𝒜). The coproduct will be C⊕D. Here, we need to check that

Similarly, C⊕D is also the product.

Pushouts are trickier. Let f:C→D and g:C→D′ be morphisms. Then we will construct the mapping cone C⁡(f,g) by setting the n-th element of the complex to be

with the differential

This gives us a pushout square

We will omit checking that this is actually a pushout. There is a similar description for fiber products. □

Note that 𝒦⁡(𝒜) is not the actual derived category of 𝒜. For example, there is a morphism

but not a morphism in the other direction. We want to localize at such quasi-isomorphisms, and the localization we will take is called Dwyer-Kan localization.

Definition 1.1.4.6.

Let 𝒞 be an ∞-category and W⊆𝒞1 be any subset of morphisms. A functor

is a Dwyer-Kan localization at W if

  • •

    f takes W to equivalences in 𝒞′;

  • •

    For any 𝒟, the functor

    is fully faithful with essential the full subcategory

    on those functors 𝒞→𝒟 which take W to equivalences. We will denote such a 𝒞′ by 𝒞⁡[W−1].

Example 1.1.4.7.

Let 𝒞=Δ1 and let W consist of the nontrivial morphism 0→1. Then

because if we invert the morphism, the two objects become equivalent. We will check the universal property now. Clearly, the functor Δ1→Δ0 takes the unique element of W to an equivalence. We now need to check that

is an equivalence. To prove that it is essentially surjective, we need to check that isomorphisms d≃d′ are equivalent to d→id⁢d, which is left as an exercise. To prove that it is fully faithful, we need to check that

is a fiber product of the diagram

This proof will also be omitted.

Exercise 1.1.4.8.

Show that every object

is equivalent to (d→idd).

Having defined Dwyer-Kan localizations, we still need to see that they actually exist.

Proposition 1.1.4.9.

For any ∞-category 𝒞 and any set of morphisms W, the Dwyer-Kan localization 𝒞⁡[W−1] exists and the functor

is essentially surjective.

Proof.

We can begin by trying the pushout

in 𝖢𝖺𝗍∞. If we apply 𝖥𝗎𝗇⁡(−,𝒟) to this square to obtain the square

We need to prove that this is a fiber product in 𝖢𝖺𝗍∞. Note that these functor categories (which are not Kan complexes) differ from mapping spaces in 𝖢𝖺𝗍∞, which are.

Before we do this, note that note that the essential image of

are those functors which send W to equivalences. By general properties of base change, the essential image of

consists of exactly those functors which take W to equivalences.

To prove the claim about fiber products of functor categories, we will show that for any pushout

the diagram

is a fiber product. Applying 𝖬𝖺𝗉𝖢𝖺𝗍∞⁢(ℱ,−), we first note that

This is because they are both the maximal Kan complex inside of

Therefore, we obtain the diagram

which is a fiber product. □

Therefore, we see that 𝖢𝖺𝗍∞ is really an (∞,2)-category, but we will not discuss this in detail here. We are now ready to define the derived category, however.

Definition 1.1.4.10.

Let 𝒲⊆𝒦⁢(𝒜)1 be the set of quasi-isomorphisms, namely morphisms which induce isomorphisms on homology. Now define the derived category 𝒟⁡(𝒜) by the formula

This is unfortunately extremely non-explicit. For example, it does not tell us how to work with mapping spaces in 𝒟⁡(𝒜).

1.1.5 Slices and filtered colimits

Our goal is now to give a description of mapping spaces in 𝒟⁡(𝒜) as colimits of mapping spaces in 𝒦⁡(𝒜). In many cases, they will even agree with mapping spaces in 𝒦⁡(𝒜). In general, the problem of describing mapping spaces in Dwyer-Kan localizations is very hard.

Definition 1.1.5.1.

Let 𝒞 be an ∞-category and x∈𝒞 be an object. We will define the slice category over x 𝒞/x by the fiber product

and the slice category under x by the fiber product

Note that objects of 𝒞/x are morphisms c→x and objects of 𝒞x⁣/ are morphisms x→c.

Definition 1.1.5.2.

For any chain complex C∈𝒦⁡(𝒜), define the subcategory

to be the full subcategory on all C′→C which are quasi-isomorphisms. Dually, we can define

Proposition 1.1.5.3.

We have the following description for mapping spaces in 𝒟⁡(𝒜):

Warning 1.1.5.4.

The mapping spaces as we have defined them are not functors 𝒞op×𝒞→𝒮. In this case, we can get around this by noting that both 𝒦⁡(𝒜) and 𝒮 are homotopy-coherent nerves of ordinary categories, but in general we will need the technique of straightening, which is beyond the scope of these notes.

Because quasi-isomorphisms become equivalences in 𝒟⁡(𝒜), we have a morphism

Every morphism in 𝒦⁢(𝒜)/C𝗊𝗂 induces an equivalence on mapping spaces in 𝒟⁡(𝒜), so the target of this morphism becomes

Exercise 1.1.5.5.

Check that the colimit of a constant functor (𝒦⁢(𝒜)/C𝗊𝗂)op→𝒮 is given by its value. Hint: first give a description of colimits of constant diagrams 𝒞→𝒮.

Definition 1.1.5.6.

A complex C∈𝒦⁡(𝒜) is K-projective if the mapping space functor 𝖬𝖺𝗉𝒦⁡(𝒜)⁢(C,−) sends quasi-isomorphisms to homotopy equivalences. Similarly, a complex C∈𝒦⁡(𝒜) is K-injective if 𝖬𝖺𝗉𝒦⁡(𝒜)⁢(−,C) sends quasi-isomorphisms to homotopy equivalences.

Corollary 1.1.5.7.

If C is K-projective or D is K-injective, then

Remark 1.1.5.8.

There is a fundamental lemma in homological algebra, which states that bounded below and levelwise projective chain complexes are K-projective. Similarly, bounded above and levelwise injective complexes are K-injective.

Given some C∈𝒦⁡(𝒜) and a quasi-isomorphism C′→C with C′ K-projective, then

This is the usual way that Ext groups are computed if 𝒜 has enough projectives. Dually, we can compute mapping spaces by taking a K-injective replacement of D.

Our next goal is to understand limits and colimits in 𝒟⁡(𝒜). Limits are easy to understand if there are enough K-projective objects and colimits are easy to understand if there are enough K-injective objects. In general, we will not need this, but at the cost of introducing filtered colimits.

Definition 1.1.5.9.

An ∞-category 𝒞 is filtered if for any finite simplicial set K, every map K→𝒞 extends over the simplicial set

This is a simplicial set obtained by taking a cone from K out to the right. Dually, we say that 𝒞 is cofiltered if 𝒞op is filtered.

Exercise 1.1.5.10.

A 1-category 𝒞 is filtered if and only if for every a,b∈𝒞, there exists an object c∈𝒞 and morphisms a→c and b→c and if for every pair of morphisms a⇉b there exists a morphism b→c such that the two composites are equal. Show that 𝒞 is filtered if and only if N⁡(𝒞) is filtered in the ∞-categorical sense.

We will call colimits indexed over filtered diagrams filtered colimits. Dually, we can also speak of cofiltered limits. Also, being filtered is a weaker property than having colimits.

Lemma 1.1.5.11.

If I is a filtered ∞-category and J is a finite simplicial set, then the diagram

commutes.

This result only holds for the ∞-category of spaces and is nontrivial even though we only need to check it for J being a fiber product square.

Exercise 1.1.5.12.

Prove that filtered colimits commute with finite products in the 1-category 𝖲𝖾𝗍.

Lemma 1.1.5.13.

If K∈𝒮 comes from a finite simplicial set, then the functor

commutes with filtered colimits.

The analogous result for finite sets follows by the previous exercise.

Exercise 1.1.5.14.

Prove the lemma using the fact that filtered colimits commute with finite limits.

Corollary 1.1.5.15.

The functor

commutes with filtered colimits.

Proof.

This follows from the fact that the functor

commutes with all colimits. □

Exercise 1.1.5.16.

Prove that

commutes with all colimits.

We will now apply this to the discussion of mapping spaces in 𝒟⁡(𝒜).

Lemma 1.1.5.17.

The categories 𝒦⁢(𝒜)C⁣/𝗊𝗂 and (m⁢c⁢K⁢(𝒜)/C𝗊𝗂)op are both filtered.

Proof.

We will instead note that they have all colimits. More precisely, we have that

is closed under colimits. We will not give the proof of this in full, but first we consider the cofiber functor

which sends a morphism C→C′ to its cofiber, which is the pushout of the diagram

Because it is itself a colimit, it commutes with colimits. Now, a morphism C→C′ is a quasi-isomorphism if and only if its cofiber is acyclic (in other words, quasi-isomorphic to 0), so we only need to check that acyclic objects in 𝒦⁡(𝒜) are closed under colimits, which is true by definition. A dual argument gives the result in the over category. □

Proposition 1.1.5.18.

The derived category 𝒟⁡(𝒜) has all finite limits and colimits and they are preserved by the functor 𝒦⁡(𝒜)→𝒟⁡(𝒜).

Proof.

First, we need to check that 0 is still initial and terminal in 𝒟⁡(𝒜). Note that 0 is K-injective and K-projective, so because mapping spaces can be computed in 𝒦⁡(𝒜), it is still initial and terminal.

Now consider a pushout

in 𝒦⁡(𝒜). Therefore, the square

is a fiber product in spaces. Then the diagram

is a filtered colimit of diagrams of mapping spaces in 𝒦⁡(𝒜) for E′∈𝒦⁢(𝒜)E⁣/𝗊𝗂. Because filtered colimits commute with finite limits in spaces, it is a fiber product. Dually, we see that fiber products are preserved.

Now we need to check that finite limits and colimits actually exist. Consider a diagram

in 𝒟⁡(𝒜). Because π0 commutes with colimits, we see that

is a filtered colimit of sets. Therefore, there exists a quasi-isomorphism C→C′ such that the morphism A→C→C′ comes from 𝒦⁡(𝒜). Therefore, we have a morphism

where the bottom row comes from 𝒦⁡(𝒜). Therefore, we can simply take the colimit of the bottom row. The same trick gives us fiber products. □

Warning 1.1.5.19.

The Dwyer-Kan localization producing 𝒟⁡(𝒜) is extremely nice. In general, this is not the case. We will see this in the following statements

Proposition 1.1.5.20.

The Dwyer-Kan localization of simplicial sets at weak equivalences

yields the ∞-category of spaces. Note that we have produced highly nontrivial mapping spaces by localizing a category with only discrete mapping spaces and that limits are usually not preserved by this procedure.

Proposition 1.1.5.21.

Inverting all quasi-isomorphisms from the 1-category of chain complexes yields the derived category of 𝒜, or in other words,

This is very hard to deal with, so the presentation we gave is better. Similarly, if we let W be the set of chain homotopy equivalences, we obtain

Therefore, we have in some sense performed a two-step localization.

1.1.6 Derived functors

We will discuss derived functors even when there are not enough injectives or projectives. Also, we will see how these are related to adjunctions and Kan extensions.

Assume that F:𝒜→ℬ is an additive functor, which means that F⁡(0)≅0 and

This induces a functor

given by levelwise applying F. More precisely, it is given by

and is an enriched functor with respect to the dg structure on 𝖢𝗁⁡(𝒜). Therefore, we may apply the dg nerve construction to obtain a functor

which commutes with finite limits and colimits. Our goal now is to induce a functor

One thing we might ask is that the square

is commutative in 𝖢𝖺𝗍∞. Unfortunately, this is too much to ask for in general.

Exercise 1.1.6.1.

Prove that the following are equivalent:

  1. 1.

    There exists a functor 𝒟⁡(𝒜)→𝒟⁡(ℬ) making the square commutative;

  2. 2.

    The functor 𝒦⁡(𝒜)→𝒦⁡(ℬ) preserves quasi-isomorphisms;

  3. 3.

    The functor 𝒜→ℬ is exact.

Of course, most functors we would like to derive are not exact, so we need to do something different in general. We could then try to ask for a weaker version of a commutative square, namely that there is a natural transformation

Unfortunately, even the zero functor works for L⁢F, so maybe we want some universality property.

Definition 1.1.6.2.

The left derived functor L⁢F of F is a functor

together with a natural transformation

such that for any other functor

the map

is an equivalence.

Warning 1.1.6.3.

The functor L⁢F might not always exist.

Proposition 1.1.6.4.

If L⁢F exists, it preserves finite limits and colimits.

Definition 1.1.6.5.

Given a pair (𝒞,W) and a functor G:𝒞→𝒟, the left derived functor L⁢G is given by a diagram

such that the induced map

is an equivalence for all H:𝒞⁡[W−1]→𝒟. We say that LF is the absolute left derived functor if for any T:𝒟→ℰ, the induced triangle

exhibits T∘L⁢G as the left derived functor of T∘G.

Of course, absolute left derived functors are even more rare than left derived functors.

Remark 1.1.6.6.

We have used the fact that

For a given G∈𝖥𝗎𝗇⁡(𝒞,𝒟), we have L⁢G∈𝖥𝗎𝗇W⁢(𝒞,𝒟) and

for H∈𝖥𝗎𝗇W⁢(𝒞,𝒟), which is an example of coreflection (or a right Kan extension).

The main result of this section is the following.

Theorem 1.1.6.7.

For any functor G:𝒦⁡(𝒜)→𝒟, we have

provided that this limit exists for each C∈𝒦⁡(𝒜). This limit is taken over the slice category 𝒦⁢(𝒜)/C𝗊𝗂.

Example 1.1.6.8.

For C=0, we obtain, we have

because 0 is initial in the slice category.

Example 1.1.6.9.

Assume that there is a K-projective object

This is in fact initial in 𝒦⁢(𝒜)/C𝗊𝗂 because for any other object C^^→C, the square

is a fiber product. Because C^^ is quasi-isomorphic to C, the right vertical arrow is an equivalence, so the mapping space we want is contractible. Therefore, we have

Remark 1.1.6.10.

If for every C there exists such a K-projective resolution C^, then

for any functor G. This implies that L⁢G is actually an absolute left derived functor, or in other words that every left derived functor is absolute.

Remark 1.1.6.11.

Often, people define derived functors by taking projective resolutions, but this of course runs into problems with functoriality of projective resolutions. In addition, our approach generalizes better to the nonadditive setting.

Example 1.1.6.12.

Assume that F:𝒜→ℬ is additive, 𝒜 is small, and ℬ has all infinite products and they are exact (Grothendieck’s AB4*). Then 𝒟⁡(ℬ) has all limits and the derived functor

exists and preserves all finite limits and colimits.

Exercise 1.1.6.13.

Prove that the large category 𝒟⁡(ℬ) has all small limits if B satisfies AB3* and AB4*, namely that ℬ has all small products and they are exact. Use as a black box the fact that if a large ∞-category 𝒞 has all small colimits, then 𝒞-indexed colimits in the category of large spaces commute with all small products, not just finite ones.

We have discussed left derived functors until now, but of course there is a dual notion of right derived functors. These are given by diagrams

with the appropriate universality property. The dual of our formula for L⁢G reads as

where the colimit is taken over 𝒦⁢(𝒜)C⁣/𝗊𝗂.

Example 1.1.6.14.

We have functors

Then the formulae for mapping spaces show that

This is usually written as

Unfortunately, we cannot derive in both variables at the same time, but we will be able to do this when we introduce nonabelian derived functors.

Sketch of proof of Theorem 1.1.6.7.

The first step is to prove that defining

defines a functor 𝒦⁡(𝒜)→𝒟.

The next step is to prove that it preserves weak equivalences. This seems obvious because we are taking a limit over weak equivalences, which requires that the pullback functor between slice categories along a weak equivalence has an adjoint.

As a consequence, there is a natural transformation

If G already preserves quasi-isomorphisms, then η is an equivalence.The last step is to show that

is an inverse for η∗. □

We will now discuss adjunctions and cofinality.

Definition 1.1.6.15.

Let 𝒞 and 𝒟 be ∞-categories with functors

A natural transformation ε:L⁢R→id𝒟 is a counit of an adjunction if for a given pair of objects c∈𝒞 and d∈𝒟, the induced map

is a homotopy equivalence. Dually, a natural transformation η:id𝒞→R⁢L is a unit of an adjunction if the induced mao

is an equivalence. In either case we say that L is left adjoint to R and write it as L⊣R.

Note that we have given two ways for L to be left adjoint to R, but we will see that they are actually equivalent.

Definition 1.1.6.16.

Given functors L,R as before and any natural transformations ε:L⁢R→id and η:id→R⁢L, the zig-zag identities hold if the composites

are equivalent to idL and idR, respectively.

Proposition 1.1.6.17.

If natural transformations ε and η satisfy the zig-zag identities, then they are unit and counit of an adjunction. Conversely, given a unit of an adjunction η, there is a unique counit ε such that the zig-zag identites hold.

Proof.

If ε and η satisfy the zig-zag identities, then

are inverse to each other. We omit the proof of the existence of a counit given a unit. □

Exercise 1.1.6.18.

Show that if F:𝒞→𝒟 is an equivalence, it is left and right adjoint to its inverse.

We now present some important facts about adjoint functors.

  1. 1.

    Given a functor L:𝒞→𝒟, then the right adjoint (R,η) is unique if it exists;

  2. 2.

    The composite of left adjoints is again a left adjoint with the right adjoint given by the composite of right adjoints. A more coherent statement is that we have an equivalence

    where 𝖢𝖺𝗍∞L is the ∞-category with ∞-categories as objects and left adjoint functors as morphisms and 𝖢𝖺𝗍∞R is defined similarly.

Now let I be a small ∞-category. Consider the functor

given by pulling back along I→pt. Assume that Δ has a left adjoint L. Then for any functor F:I→𝒞, the unit gives a map F→Δ⁡(L⁢F) such that the induced map

is an equivalence. In other words, we have

Proposition 1.1.6.19.
  • •

    If 𝒞 has all I-shaped colimits, then Δ admits a left adjoint.

  • •

    Assume that L:𝒞→𝒟 has for any d an object R⁢d with a map L⁢R⁢d→d such that for all c∈𝒞, the map

    is an equivalence. Then L admits a right adjoint R given pointwise by d↦R⁢d.

Exercise 1.1.6.20.

Prove the following statements.

  • •

    Consider the functor π0:𝒮→𝖲𝖾𝗍. This has a left adjoint given by the discrete topology.

  • •

    Given a functor forgetful functor 𝒞/x→𝒞 on a slice category, it is a left adjoint if 𝒞 has products.

Adjoint functors behave well with respect to limits and colimits as in the 1-categorical setting.

Proposition 1.1.6.21.

Left adjoints preserve colimits and right adjoint preserve limits.

Proof.

The proof is left as an exercise. □

Definition 1.1.6.22.

Given a functor p:𝒞→𝒞′ and any functor F:𝒞→𝒟, a triangle

exhibits R⁢𝖪𝖺𝗇⁢F as the right Kan extension of F along p if it is terminal among all such diagrams as in the definition of a left derived functor. In other words, the induced map

is an equivalence.

Corollary 1.1.6.23.

If the right Kan extension R⁢𝖪𝖺𝗇⁢F exists for every F:𝒞→𝒟, then the restriction functor

admits a right adjoint R⁢𝖪𝖺𝗇.

Therefore, we see that derived functors are right Kan extensions.

1.1.7 Nonabelian derived functors

In this section, we will discuss nonabelian derived functors and animation in order to derive non-additive functors. This will be technical, but it will be useful to understand these notions.

First, we will recast derived functors of additive functors F. Assume that 𝒦⁡(𝒜) has enough K-projectives. For example, this is true where 𝒜 is the category of modules over a ring. First, the functor

is an equivalence and i is left adjoint to the projection p. Using this equivalence, the left derived functor L⁢F:𝒟⁡(𝒜)→𝒟 of any functor F:𝒦⁡(𝒜)→𝒟 is simply the restriction along i. In particular, if the target is an abelian category ℬ, the derived functor

preserves connective objects, or in other words

Our goal is now to answer the following questions.

  • •

    How can we universally characterize p∘𝒦⁡(ℱ):𝒦⁢(𝒜)≥0→𝒟⁡(ℬ)?

  • •

    What if F is not additive?

Exercise 1.1.7.1.

Given a pointed functor 𝒜→ℬ between abelian categories, prove that if there exists a functor 𝒦⁢(𝒜)≥0→𝒟⁡(ℬ) given on chain complexes representing objects by applying F levelwise, F must already be additive.

Before we proceed, we need to discuss the Yoneda lemma in ∞-categories. Unfortunately, this will require discussing size issues. We will fix three Grothendieck universes

which is equivalent to fixing three strongly inaccessible cardinals. This becomes three nested models of ZFC where the set of all small sets is a large set and the set of very large sets is a very large set.

We may discuss categories and ∞-categories in all of these universes:

  • •

    Small categories have small object and morphism sets;

  • •

    Large categories have large object and morphism sets;

  • •

    Very large categories have very large object and morphism sets.

For example, the ∞-category 𝒮 of small spaces is a large ∞-category and the ∞-category 𝒮^ of large spaces is a very large ∞-category. For example, if 𝒞 is a large category, then 𝖬𝖺𝗉𝒞⁢(a,b)∈𝒮^. Another example of this is that if we let 𝖢𝖺𝗍∞𝗌𝗆𝖺𝗅𝗅 be the large ∞-category of small ∞-categories, we can also define 𝖢𝖺𝗍∞ to be the very large ∞-category of large ∞-categories. This approach avoids issues involving proper classes, but requires jumping between universes.

Some of our previous constructions can now be discussed more precisely. Here, if 𝒞∈𝖢𝖺𝗍∞, then the Dwyer-Kan localization is still large, or in other words,

Another example is that the formula

really lands in the ∞-category 𝒮^ of large spaces.

Definition 1.1.7.2.

A large ∞-category 𝒞 is locally small if for any pair a,b∈𝒞 of objects, the mapping space

is equivalent to an object to 𝒮⊆𝒮^, i.e. is essentially small. Equivalently, the functor

factors through 𝒮⊆𝒮^.

Construction 1.1.7.3.

For any large ∞-category 𝒞, we have a functor

which is given on objects by

called the Yoneda embedding. If 𝒞 is locally small, this factors as

Theorem 1.1.7.4 (Yoneda Lemma).
  1. 1.

    The functor j is fully faithful.

  2. 2.

    For any F:𝒞op→𝒮^ and any X∈𝒞, there is a natural equivalence

  3. 3.

    Every object F∈𝖥𝗎𝗇⁡(𝒞op,S^) is a (large) colimit of objects of the form X¯ for X∈𝒞.

Proof.

The proof should be the same as with ordinary categories, but machinery (straightening and unstraightening) is required to handle coherence issues. We will omit this. Of course, the second part implies the first by the equation

For the last item, we consider the pullback

The category 𝒞/F is actually large, not very large. We now claim that

Exercise 1.1.7.5.

For any F∈𝖥𝗎𝗇⁡(𝒞op,𝒮^), prove that

Because of our keeping track of size constraints, note that if 𝒞 is small, then every

is a small colimit of representables.

Now let 𝒞→𝒟 be a functor where 𝒞 is small and 𝒟 is possibly large.

Proposition 1.1.7.6.

Assume that 𝒟 admits all small colimits. Then

  1. 1.

    There exists an essentially unique colimit-preserving functor

    extending F. In other words, there is a left Kan extension of F.

  2. 2.

    If 𝒟 is locally small, then this is left adjoint to the restricted Yoneda embedding

Corollary 1.1.7.7.

There is an equivalence

In other words, 𝒫⁡(𝒞) is the universal ∞-category obtained from 𝒞 obtained from 𝒞 be freely adjoining colimits.

This is really just a restatement of the proposition.

Construction 1.1.7.8.

Let K be any class1 of small colimit shapes, for example all colimits, all finite colimits, all filtered colimits, or all geometric realizations. We form the category

as the smallest full subcategory which contains representables and which is closed under K-indexed colimits.

Proposition 1.1.7.9.

For any large ∞-category 𝒟 which admits K-indexed colimits, restriction along j:𝒞→𝒫K⁢(𝒞) is an equivalence

whose inverse is given by left Kan extension.

Example 1.1.7.10.

If 𝒞 is any ∞-category, we can define the Ind-category

We claim the following:

  1. 1.

    Objects in 𝖨𝗇𝖽⁡(𝒞) are given by functors F:I→𝒞, where I is filtered. These objects are denoted

  2. 2.

    For any pair F,G, we have

    In particular, if 𝒞 is a 1-category, so is 𝖨𝗇𝖽⁡(𝒞).

We are finally now able to return to nonabelian derived functors.

Proposition 1.1.7.11 (Dold-Kan, Quillen, …).

Assume that 𝒜 has enough compact projective objects (compact means that Hom𝒜⁡(A,−) preserves filtered colimits). Then we have an equivalence

Corollary 1.1.7.12.

A functor

which preserves geometric realizations and filtered colimits is uniquely determined by its restriction to 𝒜𝖼𝗉.

This is precisely a nonabelian derived functor, which is given by taking the Kan extension in the diagram

In other words, for any object in 𝒟⁢(𝒜)≥0, we resolve it by simplicial diagrams of compact projectives and then evaluate the functor termwise.

We conclude by giving the categorical incarnation of nonabelian derived functors.

Definition 1.1.7.13.

Let 𝒞 be a category which admits small colimits and is generated under small colimits by 𝒞𝖼𝗉. Here, an object x∈𝒞 is projective if 𝖬𝖺𝗉𝒞⁢(x,−) commutes with geometric realizations (if 𝒞 is a 1-category, we replace geometric realizations with split coequalizers). Then the animation 𝖠𝗇𝗂⁡(𝒞) is defined as

Example 1.1.7.14.

If 𝒞=𝖲𝖾𝗍, we obtain

1.2 Higher algebra

One of the most important objects in homotopy theory is the ∞-category of spectra. Studying spectra will give us a homotopy-enriched version of algebra.

1.2.1 Spectra

Let 𝒮∗ be the ∞-category of pointed spaces. There are functors

where Σ is given by the pushout

and Ω is given by the fiber product

Exercise 1.2.1.1.

Prove that Σ⊣Ω.

Definition 1.2.1.2.

A spectrum is a sequence of (pointed) spaces Xi for i≥0 together with homotopy equivalences

Definition 1.2.1.3.

The ∞-category 𝖲𝗉 of spectra is the infinite fiber product

where 𝖤𝗊⁡(𝒮∗)⊆𝒮∗Δ1 is the full subcategory on equivalences, the rightward morphisms 𝖤𝗊⁡(𝒮∗) take an arrow to its target, and the leftward morphisms take an arrow to the loop space of its source.

Note that objects in 𝖲𝗉 are exactly given by lists of

Maps in 𝖲𝗉 are lists of maps

together with homotopies

Remark 1.2.1.4.

We can write 𝖲𝗉 as a colimit

Using this, mapping spaces can be described as

Example 1.2.1.5.

If A is an abelian group, there is a spectrum H⁢A given by

By definition, we have equivalences

Example 1.2.1.6.

Let X∈𝒮∗ be a pointed space. Then there is a spectrum Σ∞⁢X given by

Here, we are using the fact that the loop space is a fiber product, so forming it commutes with filtered colimits. In particular, if X=S0, we define Σ∞⁢S0≕𝕊.

Lemma 1.2.1.7.

If X is a space, then

for any spectrum Y.

Proof sketch.

Consider a map

By the adjunction, it is adjoint to a morphism

Applying Ωk, we obtain a morphism

This yields a map

Going in the other direction is given by restricting to (Σ∞⁢X)0 and restricting to the first term in the colimit. □

Definition 1.2.1.8.

Define the functor Ω∞:𝖲𝗉→𝒮∗ by

Using this definition, the previous lemma becomes the statement that Ω∞ and Σ∞ are adjoint.

Example 1.2.1.9.

For a finite K∈𝒮∗, we have

For example, if X=S0, we see that

are the stable homotopy groups.

We will now see that 𝖲𝗉 is well-behaved.

Lemma 1.2.1.10.

The ∞-category 𝖲𝗉 has all limits, filtered colimits, and a zero object.

Proof.

The zero object is given by Xi=pt. Limits and colimits both commute with Ω, so we can take them pointwise and still get a spectrum. By the formula for mapping spaces, they satisfy the universal properties. □

Lemma 1.2.1.11.

The functor Ω:𝖲𝗉→𝖲𝗉 is an equivalence.

Proof.

Applying Ω to a spectrum is given by

This is just a shift, and its inverse is given by shifting down. □

Proposition 1.2.1.12.

The ∞-category 𝖲𝗉 has all colimits and pushout squares are the same as pullback squares (and vice versa).

Very rough proof sketch.

The proof is surprisingly elementary but quite tricky. The key idea is that ∞-category with all finite limits, a zero object, and where Ω is an equivalence has pushouts. These pushouts are determined by the fact that pushout squares are the same as pullback squares. □

We will call ∞-categories with zero objects and all finite limits and colimits where pushout and pullback squares are the same stable. Another example of a stable ∞-category is the derived category 𝒟⁡(𝒜).

Remark 1.2.1.13.

The suspension functor Σ:𝖲𝗉→𝖲𝗉 is inverse to Ω. Therefore, it is simply the left shift functor. Also note that Yi is equivalent to Ω∞⁢Σi⁢Y.

Example 1.2.1.14.

Note that

Similarly, we have

Therefore, we can obtain both cohomology and stable homotopy groups. Because 𝖲𝗉 is stable, we get long exact sequences in both cohomology and in stable homotopy.

Unfortunately, not every spectrum is a suspension spectrum. However, this is close to being true.

Lemma 1.2.1.15.

For all Y∈𝖲𝗉, we have

Proof.

We will prove that they have the same mapping spaces. In particular, we compute

□

Our goal is now to make sense of homology, To do this correctly requires us to talk about symmetric monoidal structures, but we can give a different construction.

Construction 1.2.1.16.

We will define a functor 𝖲𝗉→𝒟⁡(ℤ) given by

This functor is characterized by preserving colimits and taking Σ∞⁢X↦C∗⁢(X). In particular, we have

We want to view spectra as modules over 𝕊 and the homology functor as a base change to 𝒟⁡(ℤ). We will now attempt to make this precise.

Definition 1.2.1.17.

Define the category 𝖲𝗉𝖿𝗂𝗇⊆𝖲𝗉 of finite spectra to be the full subcategory of all spectra of the form Σ−i⁢Σ∞⁢K for K a finite space.

Exercise 1.2.1.18.

Prove that 𝖲𝗉𝖿𝗂𝗇 is closed under finite colimits.

Proposition 1.2.1.19.

We have 𝖲𝗉=𝖨𝗇𝖽⁡(𝖲𝗉𝖿𝗂𝗇).

Proof.

There are a natural functor 𝖨𝗇𝖽⁡(𝖲𝗉𝖿𝗂𝗇)→𝖲𝗉. The objects in 𝖲𝗉𝖿𝗂𝗇 are compact objects in 𝖲𝗉. This is because

Therefore, the functor is fully faithful. To prove that it is essentially surjective, we write a spectrum Y as

and then write Yi as a filtered colimit of finite spaces. □

Note that in the derived category 𝒟⁡(R), the subcategory 𝖯𝖾𝗋𝖿⁡(R) of perfect complexes consists exactly of the compact objects. This is analogous to the fact that every finite spectrum is a finite limit of shifts of 𝕊. In addition, we actually have

so we will think of 𝖲𝗉 as the derived category 𝒟⁡(𝕊) if such a thing existed (this requires a way to think about 𝕊 as a ring). However, a point of non-analogy is that the mapping space

is not discrete (and in fact it contains all stable homotopy groups of spheres, so is extremely complicated).

We will now make abelian groups a full subcategory of spectra.

Definition 1.2.1.20.

A spectrum X∈𝖲𝗉 is n-connective if

for all i<n. Dually, X is n-coconnective if πi⁢X=0 for all i>n. We will denote the full subcategories as 𝖲𝗉≥n and 𝖲𝗉≤n.

Proposition 1.2.1.21.

The inclusion

has a left adjoint τ≤n.

Proof.

Denote Xn≔X. We will build Xi as follows:

  • •

    Choose generators of πi⁢(Xi−1) and consider the corresponding map

  • •

    Considering the cofiber of this morphism, we obtain a pushout

  • •

    Let τ≤n≔colimiXi.

This satisfies the property that τ≤n⁢X∈𝖲𝗉≤n and that πi⁢τ≤n⁢X=πi⁢X for all i≤n. This uses the fact that 𝕊 is connective. Another property is that every map

factors through all of these pushouts because Y has no homotopy groups for i>n and therefore through τ≤n⁢X. □

Definition 1.2.1.22.

Let τ≥n⁢X denote the fiber of the morphism X→τ≤n−1⁢X.

Lemma 1.2.1.23.

If X∈𝖲𝗉≥n and Y∈𝖲𝗉≤n−1, then

In addition, there is an equivalence

Proof.

Any map X→Y factors through τ≤n−1⁢X→Y. But now τ≤n−1⁢X is weakly contractible, so it must be 0.

For the latter statement, note that πn gives a functor. In addition, note that mapping spaces are discrete because

for i>0. If πn⁢(X) is free, choose a map

which is an isomorphism on πn and therefore an equivalence on τ≤n. This gives us an isomorphism

If πn⁢(X) is not free, we choose a map

which is surjective on πn. The fiber has free πn, so we use the exact sequence for π∗⁢𝖬𝖺𝗉⁢(−,Y) to conclude. □

We will now discuss the Postnikov tower. For ever X, we have a tower

such that X is actually the limit of this sequence. Unfortunately, to actually reconstruct X from its truncations requires a lot of gluing data.

1.2.2 Symmetric monoidal ∞-categories

Our goal is to explain why 𝖲𝗉 and 𝒟⁡(R) have canonical symmetric monoidal structures.

Recall that an ordinary symmetric monoidal category consists of a category 𝒞 with a tensor product

a unit object 𝟏𝒞∈𝒞, and natural isomorphisms

satisfying a number of coherence conditions. One of them is the pentagon identity, which says that the diagram

commutes.

In this setting, a lax symmetric monoidal functor between symmetric monoidal categories 𝒞 and 𝒟 is a functor

together with natural morphisms

for all c,c′∈𝒞 and a natural morphism

These must be compatible with the associativity, unit, and symmetry isomorphisms. Such a lax symmetric monoidal functor is strong symmetric monoidal if the above natural morphisms are isomorphisms.

Also recall that a symmetric monoidal category (𝒞,⊗) is closed if for all c∈𝒞, the functor −⊗c admits a right adjoint Hom¯⁢(c,−) and in particular preserves colimits. Our goal is to generalize the following result to ∞-categories.

Proposition 1.2.2.1.

Let R be a commutative ring. Then the category 𝖬𝗈𝖽R of R-modules admits an essentially unique symmetric closed symmetric monoidal structure with unit R.

Proof sketch.

Existence is clear, so we will consider uniqueness. Note that because R is the tensor unit, we must have

Because the tensor product is assumed to preserve colimits, we obtain

For an arbitrary R-module N, we write N as the cokernel

of a morphism of free R-modules. This implies that N⊗M is the cokernel of

This must be equal to N⊗M. □

Exercise 1.2.2.2.

Give an actual proof of the proposition.

For any morphism R→S of commutative rings, we obtain a functor

which is symmetric monoidal.

In the land of ∞-categories, we will state the following theorem, which gives some examples of symmetric monoidal ∞-categories and symmetric monoidal functors.

Theorem 1.2.2.3.
  1. 1.

    The ∞-categories 𝖲𝗉 and 𝒟⁡(R) admit essentially unique closed symmetric monoidal structures with units 𝕊 and R, respectively. We will denote the tensor products by ⊗𝕊 and ⊗R𝕃, respectively.

  2. 2.

    The functors C∗:𝖲𝗉→𝒟⁡(ℤ), Σ+∞:𝒮→𝖲𝗉, and ⊗R𝕃S:𝒟⁡(R)→𝒟⁡(S) have canonical strong symmetric monoidal structures.

This will require us to define symmetric monoidal ∞-categories such that N⁢𝒞 is an example for any symmetric monoidal category and to define lax and strong symmetric monoidal functors.

The very rough idea of the proof of the theorem for 𝖲𝗉 is as follows. If X∈𝖲𝗉, then we must have 𝕊⊗X≅X. Secondly, tensoring with the suspension spectrum of a space gives

We then obtain (Σ−n⁢Y)⊗X≅Σ−n⁢(Y⊗X) because Σ is an equivalence and is clearly symmetric monoidal. Finally, we use the fact that every spectrum is a colimit of its constituent parts Σ∞−n⁢Yn. This determines the tensor product Y⊗X up to equivalence.

Let 𝖥𝗂𝗇∗ be the category of finite pointed sets ⟨n⟩={0,…,n} with 0 as a basepoint. For every 1≤i≤n there is a map

Definition 1.2.2.4 (First definition, Segal).

A symmetric monoidal ∞-category is a functor

such that the induced maps

are equivalences for all n=0,1,…

The underlying category of such an object is 𝒞≔𝒞¯⁢(⟨1⟩). The tensor product is simply the map

where m:⟨2⟩→⟨1⟩ is given by 0↦0 and 1,2↦1. The tensor unit is given by the map

which is equivalent to an object in 𝒞. The idea is that all of the higher values of the functor encode all coherences like associativity, and symmetry.

Exercise 1.2.2.5.

For an ordinary symmetric monoidal 1-category 𝒞, construct a functor

which takes a pointed set S⊔∗ to 𝒞×S.

Definition 1.2.2.6.

A strong symmetric monoidal functor 𝒞→𝒟 is a natural transformation 𝒞¯→𝒟¯.

In particular, we could define a full subcategory

The reader could take this as a definition of the ∞-category of symmetric monoidal ∞-categories, but it has some shortcomings. The main shortcoming is that we don’t really have a way to discuss lax symmetric monoidal functors.

Definition 1.2.2.7 (Lurie).

A symmetric monoidal ∞-category is a functor

satisfying the following conditions:

  1. 1.

    It is a cocartesian fibration.2 In particular, we can consider the fibers

    and compute them as ordinary fiber products of simplicial sets. In addition, for any morphism f:⟨n⟩→⟨m⟩ in 𝖥𝗂𝗇∗, we obtain a pushforward

  2. 2.

    The induced maps

    are equivalences for all n=0,1,…

We will call 𝒞≔𝒞⟨1⟩⊗ and the tensor product is given by

Theorem 1.2.2.8 (Lurie).

The two definitions agree in that for a symmetric monoidal ∞-category 𝒞⊗→N⁢𝖥𝗂𝗇∗, we get an induced functor N⁢𝖥𝗂𝗇∗→𝖢𝖺𝗍∞ given by ⟨n⟩↦𝒞⟨n⟩⊗ and vice versa.

We will discuss how to get a fibration from a copresheaf of ∞-categories, but only do it for an ordinary symmetric monoidal category (𝒞,⊗). We define 𝒞⊗ as follows:

  • •

    Objects are given by lists c1,…,cn in 𝒞 for ⟨n⟩∈𝖥𝗂𝗇∗;

  • •

    Morphisms from c1,…,cn to d1,…,dm are given by the data of

    • –

      A map f:⟨n⟩→⟨m⟩ in 𝖥𝗂𝗇∗;

    • –

      For each k∈⟨m⟩∖0, a map

  • •

    Composition is defined in the obvious way;

  • •

    The forgetful functor takes c1,…,cn to ⟨n⟩.

Taking the nerve of this gives a symmetric monoidal ∞-category.

We will now start defining lax symmetric monoidal functors. We will call a morphism

inert if the induced map

is a bijection.

Definition 1.2.2.9.

A lax symmetric monoidal functor between symmetric monoidal ∞-categories 𝒞⊗→N⁢𝖥𝗂𝗇∗ and 𝒟⊗→N⁢𝖥𝗂𝗇∗ is a functor

such that F⊗ sends cocartesian lifts of inert morphisms to cocartesian lifts.

Exercise 1.2.2.10.

Prove that for 1-categories, a lax symmetric monoidal functor 𝒞→𝒟 induces a functor 𝒞⊗→𝒟⊗.

1.2.3 𝔼n-algebras

𝔼n-algebras interpolate between associative (𝔼1-) algebras and commutative (𝔼∞-) algebras. In this section, 𝒞 will be a symmetric monoidal ∞-category.

Definition 1.2.3.1.

An associative algebra in 𝒞 is a symmetric monoidal functor

where objects of 𝖠𝗌𝗌𝖺𝖼𝗍⊗ are finite sets, morphisms are morphisms of finite sets with a chosen linear order on preimages of points, composition is given by the lexicographic order, and the tensor product is given by disjoint union. We will denote the ∞-category of associative algebras in 𝒞 by

Definition 1.2.3.2.

A commutative algebra in 𝒞 is a symmetric monoidal functor

where 𝖢𝗈𝗆𝗆𝖺𝖼𝗍⊗ is just the category of finite sets. We will denote the ∞-category of commutative algebras in 𝒞 by

Warning 1.2.3.3.

In a 1-category 𝒞, we have an inclusion

This is completely false in ∞-categories, even if there is a functor.

We will now interpolate between commutative and associative algebras.

Definition 1.2.3.4.

Let 0≤n<∞. We define a symmetric monoidal ∞-category (𝔼n⊗)𝖺𝖼𝗍 as the homotopy-coherent nerve of the topologically enriched category given by the following:

  • •

    Objects are disjoint unions Dn⊔⋯⊔Dn of (poly)disks Dn=(0,1)n;

  • •

    Morphisms from Dn⊔⋯⊔Dn to Dn⊔⋯⊔Dn are given by rectilinear embeddings. These are simply embeddings of topological spaces which on each disk is given by an affine linear map with matrix of the form

    An example would look something like Figure 1.1.

    123
    Figure 1.1: Morphism in (𝔼2⊗)𝖺𝖼𝗍.
  • •

    Composition is simply composition of embeddings.

This category is clearly topologically enriched and has a symmetric monoidal structure given by disjoint union.

Definition 1.2.3.5.

An 𝔼n-algebra in a symmetric monoidal ∞-category 𝒞 is a symmetric monoidal functor

We will denote the ∞-category of 𝔼n-algebras in 𝒞 by

Note that all mapping spaces in (𝔼n⊗)𝖺𝖼𝗍 are determined by

This has a map

given by evaluation at the center of each disk. Here, 𝖢𝗈𝗇𝖿k⁢(Dn)⊆Dn is the configuration space and is simply the locus where no two points agree.

Exercise 1.2.3.6.

Prove that the evaluation map we just defined is a homotopy equivalence.

As an example, it is clear that

The next example is

Now suppose we have an 𝔼n-algebra A¯:(𝔼n⊗)𝖺𝖼𝗍→𝒞 with underlying algebra A¯⁢(Dn)=A. We also have a morphism

and other morphisms. Restricting to the 0-cell, this is a multiplication morphism.

To try to understand 𝔼n-algebras, we may try to begin with n=1.

Proposition 1.2.3.7.

There is an equivalence

of symmetric monoidal ∞-categories which sends a finite set S to disjoint union ⨆SD1.

Proof.

First, we show that all mapping spaces in (𝔼1⊗)𝖺𝖼𝗍 are discrete. Up to homotopy, a map D1⊔⋯⊔D1→D1 is simply an ordering of the indexing set of connected components of the source. This is precisely the set of maps in 𝖠𝗌𝗌𝖺𝖼𝗍⊗. This functor sends disjoint unions to disjoint unions, and finally a symmetric monoidal functor between symmetric monoidal ∞-categories is an equivalence in the ∞-category of symmetric monoidal ∞-categories if and only if the underlying functor is an equivalence. □

Corollary 1.2.3.8.

We have an equivalence

Exercise 1.2.3.9.

Figure out what 𝔼0-algebras are in an arbitrary symmetric monoidal ∞-category. Next, figure out what 𝔼2-algebras in the ∞-category 𝖢𝖺𝗍 of 1-categories are (you should get braided monoidal categories).

There are functors

given by taking the Cartesian product with D1. These are symmetric monoidal functors, so we have induced functors

Definition 1.2.3.10.

The ∞-category of 𝔼∞-a;gebras is the inverse limit of the diagram

in 𝖢𝖺𝗍∞.

Theorem 1.2.3.11.

We have an equivalence

Sketch of proof.

First, note that

Proving that this colimit is equivalent to N⁢𝖢𝗈𝗆𝗆𝖺𝖼𝗍⊗ is left as an exercise. □

We will now state a very important result. The proof is very hard, but the statement should be understandable. Let (𝒞,⊗) be a closed symmetric monoidal category or at least have the property that −⊗c commutes with filtered colimits and geometric realizations. We will also assume that 𝒞 has all limits and colimits.

Theorem 1.2.3.12 (Lurie).
  1. 1.

    The ∞-category 𝖠𝗅𝗀𝔼n⁢(𝒞) admits all limits and colimits for any 0≤n≤∞. The forgetful functor

    preserves filtered colimits, geometric realizations, and all limits. Moreover, it detects equivalences.

  2. 2.

    When n=∞, the coproduct and tensor product coincide.

  3. 3.

    The ∞-category 𝖠𝗅𝗀𝔼n⁢(𝒞) admits a symmetric monoidal structure such that 𝖠𝗅𝗀𝔼n⁢(𝒞)→𝒞 is symmetric monoidal.

Exercise 1.2.3.13.

Do the 1-categorical analogue of the theorem in the category 𝖠𝖻 of abelian groups.

We will now discuss 𝔼n-algebras in the ∞-category of spaces 𝒮 equipped with symmetric monoidal structure given by the Cartesian product. Then 𝖠𝗅𝗀⁡(𝒮) is the category of “monoids” in 𝒮. For any such space X, the set of connected components π0⁢(X) is a monoid.

Definition 1.2.3.14.

We say X is grouplike if π0⁢(X) is a group. This is equivalent to saying that the shear map

is a homotopy equivalence. In addition, we say an 𝔼n-algebra is grouplike if the underlying 𝔼1-algebra is grouplike.

For every pointed space X∈𝒮∗, the iterated loop space

is canonically an 𝔼n-algebra in 𝒮.

Exercise 1.2.3.15.

Make this 𝔼n-algebra structure precise by giving a functor (𝔼n⊗)𝖺𝖼𝗍→𝒮 of topologically enriched categories.

Theorem 1.2.3.16 (Boardman-Vogt).

The functor

is fully faithful with essential image the grouplike 𝔼n-algebras for any 0≤n<∞.

Corollary 1.2.3.17.

For any 𝔼n-algebra A in 𝒮 and any connected pointed space X∈𝒮∗𝖼𝗈𝗇𝗇, we have that

Proof.

We can consider the subspace A′⊆A consisting of the unit component π0⁢(A′)⊆π0⁢(A). Then

Similarly, we have

Therefore, we may assume that A is grouplike. But this means that A=Ωn⁢Y for some pointed n-connected space Y. This implies that

Sketch of proof of Theorem 1.2.3.16.

We begin with the case of n=1. Here, we have the functor

We will construct an inverse using the bar construction, which is given by

The result we need is that Ω⁢𝖡𝖺𝗋⁢(G)≃G is an equivalence of spaces. This is usually done by constructing a fiber sequence

The general step is now to induct on n. For example, if we consider 𝒮∗2−𝖼𝗈𝗇𝗇, its image in 𝖠𝗅𝗀𝔼1𝖲𝗉⁢(𝒮∗𝖼𝗈𝗇𝗇). Therefore we have a chain

This is actually the same as 𝔼2-algebras, and the general case works the same way and the desired result follows from Theorem 1.2.3.18. □

Theorem 1.2.3.18 (Dunn additivity).

For any symmetric monoidal ∞-category 𝒞, we have

for all n,m≥0.

Note that the 1-categorical version of this is the Eckmann-Hilton argument.

Corollary 1.2.3.19.

Suppose that 𝒞 has sifted colimits and ⊗ commutes with them in both variables separately. Then

given by

is well-defined.

Proof.

Note that it is simply given by

1.2.4 p-adic completion

We will generalize the process of going from ℤ to ℤp in the world of spectra. In addition, we will see how spectra can be recovered from their p-completions and their rationalizations.

Recall that an abelian group A is p-complete if it is complete (and separated) with respect to the p-adic topology on A. This is the topology which has a neighborhood basis of 0 given by pn⁢A⊆A. Equivalently, this says that

Definition 1.2.4.1.

For a spectrum X∈𝖲𝗉 and an integer n∈ℤ, define X/n to be the cofiber of the map X→𝑛X given by the addition of identity morphism n times. Equivalently, we have

As we would expect, if n|m, then there is a canonical map

defined as the cofiber of

Definition 1.2.4.2.

For a spectrum X∈𝖲𝗉, we define its p-completion to be the spectrum

Note that there is a canonical map X→Xp∧ induced from the projections X→X/pn.

Remark 1.2.4.3.

All of these constructions are well-defined up to contractible choice.

Definition 1.2.4.4.

A spectrum X is p-complete if the map X→Xp∧ is an equivalence. Denote the ∞-category of p-complete spectra by 𝖲𝗉p∧.

Theorem 1.2.4.5.
  1. 1.

    For every spectrum X, the spectrum Xp∧ is p-complete.

  2. 2.

    The functor (−)p∧:𝖲𝗉→𝖲𝗉p∧ is left adjoint to the inclusion 𝖲𝗉p∧↪𝖲𝗉.

  3. 3.

    The ∞-category 𝖲𝗉p∧ has all limits and colimits.

Sketch of proof.

We will first prove that Xp∧ is p-complete. First, note that 𝖲𝗉p∧ is closed under limits and finite colimits. This is true because the functor −/p is a cofiber and so commutes with all limits and colimits and the limit functor commutes with limits and finite colimits. The second step is to prove that for any X, X/p is p-complete. This implies that X/pn is p-complete by using the fiber sequence

This implies that Xp∧ is p-complete.

First, note that the functor (−)p∧ is idempotent. This means that

If Y is a p-complete spectrum, we can construct an inverse to

by applying p-completion to everything.

We know already that limits and finite colimits exist. For an arbitrary colimit of a functor I→𝑋𝖲𝗉p∧, the colimit is simply given by

Exercise 1.2.4.6.

Prove that a spectrum X is p-complete if and only if

Moreover, prove that X/p is p-complete for any spectrum X.

Definition 1.2.4.7.

A map f:X→Y of spectra is a p-adic equivalence if the induced map

is an equivalence.

Note that if X and Y are p-complete, then f is a p-adic equivalence if and only if f is an equivalence.

Proposition 1.2.4.8.
  1. 1.

    A map f:X→Y is a p-adic equivalence if and only if the map

    is an equivalence.

  2. 2.

    A map f:X→Y between connective X,Y is a p-adic equivalence if and only if the induced map

    is an equivalence (in other words, we can check equivalence on homology).

Proof.

The proof is left as an exercise to the reader. □

Now we know that p-adic equivalences have nice properties, but it is not clear how to tell whether a spectrum is p-complete.

Proposition 1.2.4.9.

Let A be an abelian group. If A is p-complete, then so is H⁢A∈𝖲𝗉.

For example, we see that H⁢ℤp is p-complete. The converse unfortunately fails in general, but it does hold if A has bounded order of p∞-torsion.

Definition 1.2.4.10.

An abelian group A is derived p-complete if the Eilenberg-Maclane spectrum H⁢A is p-complete.

Remark 1.2.4.11.

There is also a notion of p-completeness in 𝒟⁡(ℤ). Everything works exactly as in 𝖲𝗉. Then A is derived p-complete if and only if it is p-complete as an object of 𝒟⁡(ℤ). This is equivalent to the condition that there is a quasi-isomorphism

Theorem 1.2.4.12.
  1. 1.

    A spectrum X is p-complete if and only if πn⁢(X) is derived p-complete for all n.

  2. 2.

    Assume that πn⁢(X) has bounded order of p∞-torsion for each n. Then

Example 1.2.4.13.

The most important example of a p-complete spectrum is the p-adic completion 𝕊p∧ of the sphere. We can compute

Another example is simply (H⁢ℚ)p∧=0 because multiplication by p is an isomorphism on ℚ.

There are some nice monoidal properties associated to p-complete spectra which are listed below. We will not prove these.

  • •

    The category 𝖲𝗉p∧ of p-complete spectra admits a closed symmetric monoidal structure given by

    In particular, the p-completion functor is strong symmetric monoidal.

  • •

    If R is an 𝔼n-algebra, then so is Rp∧ and R→Rp∧ is an 𝔼n-map exhibiting Rp∧ as the initial p-complete 𝔼n-algebra under R.

It is much easier to deal with p-complete spectra than to deal with spectra. However, this is only useful if we can recover spectra from their p-completions.

Question 1.2.4.14.

Can we recover any spectrum X from its p-completions Xp∧ as p ranges over all primes?

The answer to this question must be no by the example of H⁢ℚ. In order to even attempt to have an affirmative answer, we will also need to consider rationalization. Here, recall that an abelian group A is rational if it is uniquely divisible. Equivalently, it is a ℚ-vector space.

Definition 1.2.4.15.

A spectrum X is rational if πn⁢(X) is rational for each n. We denote the full subcategory of rational spectra by 𝖲𝗉ℚ⊆𝖲𝗉.

This is actually even better-behaved than p-completeness.

Theorem 1.2.4.16.
  1. 1.

    A spectrum is rational if and only if it admits the structure of an H⁢ℚ-module.

  2. 2.

    For any spectrum X, the spectrum X⊗𝕊H⁢ℚ≕Xℚ is rational and π∗⁢(Xℚ)=π∗⁢(X)ℚ.

  3. 3.

    The functor −⊗𝕊Hℚ:𝖲𝗉→𝖲𝗉ℚ is left adjoint to the inclusion 𝖲𝗉ℚ⊆𝖲𝗉.

  4. 4.

    The category 𝖲𝗉ℚ⊆𝖲𝗉 is closed under limits and colimits.

  5. 5.

    We have an equivalence

  6. 6.

    The category 𝖲𝗉ℚ is symmetric monoidal with tensor product

    and unit H⁢ℚ.

Exercise 1.2.4.17.

Prove this theorem using the fact that H⁢ℚ⊗𝕊H⁢ℚ=H⁢ℚ.

We can now refine our question.

Question 1.2.4.18.

Can we recover any spectrum X from its p-completions Xp∧ as p ranges over all primes together with its rationalization Xℚ?

Our slogan will be that the answer to this question is yes if we also include the data of a certain “gluing map.” This uses something known as the Hasse square.

For every spectrum we have maps

In addition, rationalizing the p-completion map, we have maps

By naturality properties of rationalization, we obtain a commutative square

This is called the Hasse square or the fracture square.

Theorem 1.2.4.19.

For any spectrum X, this square is a fiber product.

This theorem is proved by noting that p is invertible on the fiber of X→Xp∧. Therefore, the fiber of the top row is rational (because all primes are invertible), so it equals its rationalization, which is the fiber of the bottom row.

The theorem can be refined to the categorical level. In particular, we have a pullback diagram

of ∞-categories. This means that we can decompose the entire ∞-category of spectra.

Unfortunately, this morphism is not symmetric monoidal, but it at least behaves well with respect to 𝔼n-algebras. More precisely, there is a pullback square

of ∞-categories, where the arrows are given by restriction from the previous square.

1.2.5 Tate construction

Let G be a group object in 𝒮. In particular, this means that G is a group-like 𝔼1-space. These can in fact be modeled by topological groups, but the correct generality is that of group-like 𝔼1-spaces. There is a classifying space B⁢G∈𝒮 given by the bar construction. If 𝒞 is any ∞-category, we will let

denote the category of objects in 𝒞 with a G-action.

Construction 1.2.5.1.

We will consider G as an object in 𝒮G×G by letting G×G act on G by the action

Definition 1.2.5.2 (Klein).

We define a spectrum DG∈𝖲𝗉B⁢G called the dualizing spectrum of G as

with its remaining G=1×G-action.

The name dualizing spectrum needs a justification, which will be given in the remainder of this section. Before we do this, we give some examples.

Example 1.2.5.3.

Assume G is a finite group. Then

Computing the homotopy fixed points, we obtain

More precisely, the map

is an equivalence.

To see this, consider the homotopy fixed point spectral sequence

It is a classical result that

Therefore the spectral sequence degenerates and yields the result. This implies that DG=𝕊 with the trivial action.

Now suppose that G is a compact Lie group.

Theorem 1.2.5.4.

If G is a compact Lie group, then

where G acts by the adjoint representation.

For example, let 𝕋=U⁡(1)=S1. Then because the adjoint representation of an abelian group is trivial, we have D𝕋=𝕊1 with the trivial action.

Now assume that B⁢G≃(M,m0) for some closed smooth manifold M. For example, if G=ℤ, then B⁢G≃S1. Alternatively, we can consider any closed manifold (M,m0) and let G=Ω⁢M.

Theorem 1.2.5.5 (Atiyah duality).

The dualizing spectrum is given by

where we extend the construction 𝕊V for any representation of G to a virtual representation. As a functor

it sends

Remark 1.2.5.6.

For any space X (in place of B⁢G), we can define a dualizing spectrum

For example, we can define this functor on connected components because

In particular, we assign DGi to B⁢Gi.

Construction 1.2.5.7.

Let 𝒞 be a stable ∞-category which has all limits and colimits. For any E∈𝖲𝗉 and X∈𝒞, there is an object

defined such that the functor

preserves colimits and the composite

is given by the functor

Definition 1.2.5.8.

Let G∈𝖠𝗅𝗀𝔼1𝗀𝗉⁢(𝒮) and X∈B⁢G. We define the norm map

as the composite

Exercise 1.2.5.9.

Prove that

This is quite an abstract construction, so we will try to get our hands on it with some examples.

Example 1.2.5.10.

If G is finite and X∈𝒞B⁢G, then the norm map is

For example, if 𝒞=𝖲𝗉 and X=H⁢M for an abelian group M with a G-action, then this map is a map

and is induced from the classical norm map

from coinvariants to invariants. This follows from the fact that for a general X, the composite

is the sum over all g∈G of the maps ρg:X→X given by multiplication by g.

Exercise 1.2.5.11.

Prove that if G is finite, the composite

agrees with the map

Example 1.2.5.12.

When G=𝕋, we get a map

Theorem 1.2.5.13.

The norm map (DG⊗X)h⁢G→Xh⁢G is an equivalence provided that one of the following conditions holds:

  1. 1.

    B⁢G is a finite CW complex;

  2. 2.

    X is induced, which means that X≃Σ+∞⁢G⊗Y where G acts only on Σ+∞⁢G.

Sketch of proof.

In the first case, all limits in the interchange maps are finite. Because all functors in the composite are exact, we are done.

In the second case, the limits and colimits are still close enough to being finite, so we are done. □

Example 1.2.5.14.

Let B⁢G be a closed manifold M, 𝒞=𝖲𝗉, and X=H⁢ℤ with the trivial action of G. In this case, the norm is a map

The target has homotopy groups given by

The source is explicitly given by

where G acts on H⁢Z⁢[−n] non-trivially. This has homotopy groups

given by homology with coefficients in the orientation cover. On homotopy groups, this map

is simply Poincaré duality (at least if M is orientable).

If we replace H⁢ℤ by any spectrum, we get Poincaré duality in arbitrary cohomology theories. If X=B⁢G is a finite CW complex, the norm equivalence is a generalized version of Poincaré duality

This can be thought of as being analogous to Serre or Verdier duality in algebraic geometry. If DX is a parameterized sphere (DX=𝕊⁡[n] as underlying spectra), then this is equivalent to X being a Poincaré duality space.

Theorem 1.2.5.15 (Klein).

The natural transformation (called the assembly map)

exhibits the functor (DG⊗−)h⁢G as the universal functor over (−)h⁢G which preserves colimits.

Note that this uniquely determines the spectrum DG as an object of 𝖲𝗉B⁢G.

Definition 1.2.5.16.

For an object X∈𝒞B⁢G, we define the Tate object as the cofiber

Example 1.2.5.17.

If B⁢G is finite, then Xt⁢G=0 for all X. If G is finite and X=H⁢M for an abelian group M with a G-action, then (H⁢M)t⁢G has homotopy groups given by the Tate cohomology

of G with coefficients in M.

Exercise 1.2.5.18.

Work out what Tate cohomology is in terms of usual homology and cohomology.

Example 1.2.5.19.

Suppose that 𝒞 is a symmetric monoidal ∞-category such that the tensor product commutes with colimits in both variables separately. Then the functor

admits a (unique) lax symmetric monoidal structure such that the natural transformation

admits a refinement to a symmetric monoidal transformation.

Corollary 1.2.5.20.

If A∈𝖠𝗅𝗀𝔼n⁢(𝒞B⁢G)≃𝖠𝗅𝗀𝔼n⁢(𝒞)B⁢G, then Ah⁢G→At⁢G is a morphism of 𝔼n-algebras.

1.2.6 Tate diagonal

The Tate diagonal is a surprising feature of the world of spectra. Before we discuss it, we will motivate it by discussing abelian groups.

Let A be an abelian group. If we try to write a map of sets

this is not a homomorphism. For example, note that

Recall now that if Cp is the cyclic group with p elements, there is a norm map

Note here that the error term of the map x↦x⊗x is in the image of the norm map. Therefore, the “diagonal” x↦x⊗⋯⊗x induces a homormorphism

This homomorphism in fact exhibits

Exercise 1.2.6.1.
  1. 1.

    Prove that x↦x⊗p induces a homomorphism

  2. 2.

    Prove that this homomorphism factors through an isomorphism

We will now discuss the analogous construction for spectra.

Theorem 1.2.6.2.

There is a unique lax symmetric monoidal natural transformation

In the rest of this section, we will attempt to see where this comes from and understand some of its properties.

Lemma 1.2.6.3.

The Singer construction

is an exact functor.

This result is particularly surprising because the p-th tensor power functor is not even additive.

Proof.

First, suppose Y≃X⊕Z. By the binomial theorem, we have

On all of the non-pure terms, the action of Cp on the connected components is free, so these terms are all induced modules over Cp. These terms vanish after taking Tate fixed points.

In general, consider a cofiber sequence X→Y→Z as a “two-stage filtration” on Y of the form

where Z appears as the associated graded term. We then obtain a filtration on Y⊗p of the form

There is a monoidal structure on filtered objects called the Day convolution structure which produces this diagram from our filtered object. The associated graded terms are given by

Applying Tate fixed points, the associated graded terms are now of the form

This implies that we have a cofiber sequence

Remark 1.2.6.4.

The Singer construction commutes with suspension as

We can write

but the action of Cp involves a nontrivial action on 𝕊p, where we take the suspension spectrum of the one-point compactification of the regular representation of Cp. In fact, we have (𝕊p)t⁢Cp≃(𝕊1)t⁢Cp, where S1 has the trivial action.

Lemma 1.2.6.5 (Stable Yoneda Lemma).

Let 𝒞 be a stable ∞-category. We consider the stable ∞-category 𝖥𝗎𝗇𝖾𝗑⁢(𝒞,𝖲𝗉) of exact functors to 𝖲𝗉. Taking mapping spectra, there exists a natural equivalence

Proof sketch.

We will reduce this to the 1-categorical Yoneda lemma. A natural transformation 𝖬𝖺𝗉⁡(X,−)→F is simply a sequence of natural transformations on the n-th underlying spaces with some compatibility conditions. More precisely, we have natural transformations

which by the ordinary Yoneda lemma yields a sequence of points in

In other words, we have an equivalence of spaces

Natural transformations X→(X⊗p)t⁢Cp correspond to maps 𝕊→(𝕊⊗p)t⁢Cp≃𝕊t⁢Cp. There is a natural choice of map like this given by

Note that there are other maps 𝕊→𝕊t⁢Cp, for example the zero map. However, we need a map which preserves the lax symmetric monoidal structure, and the unique such map is the one we have just given.

Lemma 1.2.6.6.

The functor 𝗆𝖺𝗉⁡(𝕊,−) is initial among all lax symmetric monoidal exact functors 𝖲𝗉→𝖲𝗉.

We are now ready to discuss some properties of the Tate diagonal. In particular, we will not prove anything.

Theorem 1.2.6.7.

For a bounded below X, the morphism

exhibits (X⊗p)t⁢Cp as Xp∧. In particular, (X⊗p)t⁢Cp is bounded below.

In particular, we see that 𝕊t⁢Cp≃𝕊p∧. This is completely different from Tate cohomology for abelian groups and is a special form of the Segal conjecture. Proofs of these results rely on the Adams spectral sequence because the Tate spectral sequences is too messy.

The Tate construction also allows us to define a version of Frobenius on 𝔼∞-ring spectra.

Definition 1.2.6.8.

For an 𝔼∞-ring spectrum R, we have the Tate-valued Frobenius

using the fact that multiplication R⊗p→R is equivariant. This follows from the fact that a morphism (R⊗p)h⁢Cp→R is encoded in the definition of an 𝔼∞-algebra.

We should think about this as an analogue of the morphism

for an ordinary commutative ring R.

Exercise 1.2.6.9.

Prove that it is necessary that R is commutative and that the target is reduced by p. Also prove that x↦xp is a ring homomorphism R→R/p.

Applying the Tate-valued Frobenius to

for some space X and 𝔼∞-algebra E, we obtain power operations. For example, Steenrod operations and Adams operations arise in this way.

2 (Topological) cyclic homology

2.1 Classical Hochschild and cyclic homology

In this section, we discuss Hochschild homology, (negative, periodic) cyclic homology, the HKR theorem, and related topics.

2.1.1 Hochschild homology

Let k be a field and R be a k- algebra.

Definition 2.1.1.1.

The Hochschild homology of R is the homology of the complex

where the differential is given by

Remark 2.1.1.2.

This seems related to what happens when you multiply elements placed on an S1 as in Figure 2.1. In fact, we will see later that there is a connection to the actual S1.

a0a1an……
Figure 2.1: Components of a basic tensor placed along a circle.
Example 2.1.1.3.

For now, set R=k. In this case, we have

where the differentials are alternating sums of 1 and −1. In fact, we see that the complex is

and therefore the Hochschild homology is

Example 2.1.1.4.

Because the last differential is given by a⊗b↦a⁢b−b⁢a, we see that

In particular, if R is commutative, we have

In this case, we have

We will now recall the definition of the module of Kähler differentials.

Definition 2.1.1.5.

The module of Kähler differentials of a commutative ring R is the module ΩR/k1 generated by dx for all x∈R subject to the following relations:

  1. 1.

    For all x,y∈R, we have d⁡(x+y)=dx+dy;

  2. 2.

    For all x,y∈R, we have d⁡(x⁢y)=x⁢dy+y⁢dx;

  3. 3.

    For all x∈k, we have dx=0.

Therefore, we see that ΩR/k1 receives a universal k-linear derivation from R.

There is at most one map

Exercise 2.1.1.6.

Check that this is an isomorphism.

We will now prove some more properties of Hochschild homology.

Lemma 2.1.1.7.

If R is commutative, then 𝖧𝖧∗⁢(R) has the structure of a strictly graded-commutative ring. Here, this means that x2=0 for any x of odd degree, which is only relevant in characterisic 2.

Proof.

The Hochschild complex 𝖧𝖧⁡(R/k) in fact comes from a simplicial commutative ring

where the face maps are the components of the differential ∂. The desired result then follows by Theorem A.1.0.2. □

On the de Rham side, there is also a multiplicative structure.

Definition 2.1.1.8.

The de Rham complex is the exterior algebra

By extending from degree 1, we obtain a map ΩR/k∗→𝖧𝖧∗⁢(R/k).

Theorem 2.1.1.9 (Hochschild-Kostant-Rosenberg).

If R/k has cotangent complex concentrated in degree 0, then the map

is an isomorphism.

We will discuss the cotangnet complex later, but for now we may think that R is a smooth or ind-smooth k-algebra.

Remark 2.1.1.10.

The differential on ΩR/k∗ does not play a role here. Studying what it corresponds to on the Hochschild side will leade us to cyclic homology.

Exercise 2.1.1.11.

Find an explicit description of Ωk⁡[x1,…,xn]/k∗ and in particular show that it is concentrated in degrees at most n.

We will now extend to the case when k is an arbitrary commutative ring. This will lead to flatness issues, so we will fix this by taking derived tensor products. For now, we will let R be a dg-algebra over k which is K-flat. This is a flatness condition, but when R is bounded below, it agrees with levelwise flatness. Over ℤ, it always agrees with levelwise flatness. For such an R, we define the Hochschild complex by

If R is not K-flat, we will simply replace it by a K-flat resolution R♭ and define

This is well-defined up to quasi-isomorphism.

Proposition 2.1.1.12.

For a dg-algebra R, there is a quasi-isomorphism

Remark 2.1.1.13.

This is usually called Shukla homology, but we will not make this distinction here.

Proof.

Replacing R by a flat resolution, we consider the complex

The differentials are given by multiplying adjacent terms, but without the an⁢a0 term (think configurations on an interval). This complex is usually called the bar complex. Applying −⊗R⊗kRopR, the resulting total complex computes the desired complex R⊗R⊗kRop𝕃R, but it also computes 𝖧𝖧∗⁢(R/k). □

Lemma 2.1.1.14.

If R is a dg-algebra which arises from a simplicial commutative k-algebra, then 𝖧𝖧⁡(R/k) also comes from a simplicial commutative k-algebra. In particular, the Hochschild homology groups 𝖧𝖧∗⁢(R/k) form a strictly graded-commutative k-algebra.

Proof.

We replace R by a flat simplicial commutative k-algebra. Then 𝖧𝖧⁡(Rn/k) gives a bisimplicial commutative k-algebra. The total complex is then quasi-isomorphic to the diagonal, which is clearly simplicial. □

Lemma 2.1.1.15.

If A,B are dg-algebras, then

Remark 2.1.1.16.

If R is commutative, then setting A=B=R gives a multiplication map on 𝖧𝖧 via the multiplication of R.

Proof.

Because 𝖧𝖧⁡(A/k) and 𝖧𝖧⁡(B/k) are simplicial dg-algebras, the tensor product

is a bisimplicial dg-algebra. Taking the total complex in all three directions, the result is quasi-isomorphic to the diagonal, which is precisely

We are now ready to prove the HKR theorem for polynomial rings. As we will see later, we will reduce the general case to this one.

Proof of Theorem 2.1.1.9 for polynomial rings.

In the case of R=k⁡[x], we have

Then we only need to check that 𝖧𝖧⁡(k⁡[x]/k) vanishes in degree greater than 1. This is given by

We have the resolution

of k⁡[x], and this gives the desired result.

For R=k⁡[x1,…,xn] This is simply a tensor product

so because both sides of the HKR theorem commute with tensor products, we obtain the desired result.

Finally, if R=k⁡[xi|i∈I], we write R as a filtered colimit of finitely-generated polynomial algebras by writing the set {xi|i∈I} of generators as a filtered colimit of its finite subsets. Because both sides commute with filtered colimits, we are done. □

We will now turn to an extremely non-smooth example.

Proposition 2.1.1.17.

We have

where the ⟨x⟩ means the free divided power algebra generated by x.

Proof.

We will compute the derived tensor product

We will resolve 𝔽p by the dg-algebra

The tensor product is then given by

We will resolve 𝔽p as an A-algebra by

and the desired result follows by taking the tensor product. □

2.1.2 The Connes operator

We will describe an operator on 𝖧𝖧 which corresponds to the de Rham differential. Recall that the de Rham differential

is induced from the universial derivation d:R→ΩR/k1 by the Leibniz rule.

Definition 2.1.2.1.

Let R be an associative k-algebra. We define a k-linear map

by the formula

where we define

This formula is related to the cyclic objects introduced by Connes, and there is a way to think about it systematically in the topological setting, which we will discuss later.

Exercise 2.1.2.2.

Check that

  1. 1.

    B2=0;

  2. 2.

    dB+B⁢d=0.

The operator B equips 𝖧𝖧⁡(R/k) with the structure of a dg-module over the dg-algebra

Note that if 𝕋=U⁡(1) is the circle group, then we have

We then see that 𝖧𝖧∗⁢(R/k) is a graded module over k⁡[b]/b2, and in particular that there is an operator

satisfying B2=0.

Warning 2.1.2.3.

One might object that we only defined the Connes operator only for algebras concentrated in degree 0, but with some careful thought we can define B for dg-algebras R/k.

We need the following fact.

Lemma 2.1.2.4.

If R is commutative, then B is a derivation. In particular, it satisfies the graded Leibniz rule

Warning 2.1.2.5.

The Connes operator is not true on the complex 𝖧𝖧⁡(R/k). It is true up to coherent homotopy, but proving this is very hard. Instead, we will later use the topological enhancement to understand this.

Proposition 2.1.2.6.

The map

sends the de Rham differential d to the Connes operator B. In other words, the diagram

commutes.

Proof.

The map ΩR/k∗→𝖧𝖧∗⁢(R/k) is determined by its effect in degrees 0 and 1. In degree 0, it is given by the identity

and in degree 1, it is given by

Therefore, it is enough to check that the diagram

commutes. Note that for any r∈R, we see that all signs in the definition of B are +1, so we have

This corresponds to the element

We conclude by using the fact that the de Rham differential is uniquely determined by being a derivation and its effect on R. □

Remark 2.1.2.7.

Sometimes in the rest of these notes, we will denote the Connes operator by d. If this causes confusion, we will call it B.

Example 2.1.2.8.

For 𝖧𝖧∗⁢(𝔽p/ℤ)=𝔽p⁢⟨x⟩ with x in degree 2, the Connes operator acts trivially for degree reasons.

Question 2.1.2.9.

Does this mean that it acts trivially (maybe up to homotopy or quasi-isomorphism) on 𝖧𝖧⁡(𝔽p,ℤ)?

The answer to this question is in fact no, and we will see this by taking a (derived) reduction mod p. This means that we will consider

Letting A act trivially on 𝔽p, we obtain a structure of an A-module on 𝖧𝖧⁡(𝔽p)⫽p. We compute

This implies that

Proposition 2.1.2.10.

Writing the divided power γn⁢(x) as x[n], we have B⁡(x[n])=0 and B⁡(ε)=x.

Note that this implies that

Proof.

The map 𝖧𝖧⁡(𝔽p)→𝖧𝖧⁡(𝔽p)⫽p is compatible with B, and therefore we obtain B⁡(x[n])=0 in homology.

Now recall that we computed

by replacing 𝔽p by (Λℤ⁢(ε),∂ε=p), so we write the complex

Writing out the full double complex, we obtain

The class x is given by 1⊗ε−ε⊗1. Applying B, we obtain

After reducing mod p, we can replace all copies of ℤ by 𝔽p and ε becomes a cycle, and we obtain B⁡(ε)=x. □

2.1.3 Periodic and cyclic homology

In the anology between Hochschild homology and the de Rham complex, we will now discuss cyclic homology, which will be analogous to algebraic de Rham cohomology.

We will work with derived functors in the category 𝖣𝖦𝖬𝗈𝖽A. The homotopy theory we will work with is inverting quasi-isomorphisms, which means we will consider the ∞-category

Definition 2.1.3.1.

Let R be a k-algebra.

  1. 1.

    The cyclic homology of R is

    This will not play any role in these notes.

  2. 2.

    The negative cyclic homology of R is

    This is more important than cyclic homology, and topological cyclic homology will be an analog of this rather than of cyclic homology. Also, note that 𝖧𝖢∗−⁢(R/k) is a module over ExtA∗⁡(k,k)=k⁡[t], where |t|=−2.

  3. 3.

    The periodic cyclic homology (or periodic homology) of R is

Exercise 2.1.3.2.

Show that ExtA∗⁡(k,k)≅k⁡[t], where |t|=−2.

Note that the definition we gave is not the standard one in the literature, which involves double complexes. We may wonder if these definitions are equivalent.

Proposition 2.1.3.3.

For any k-algebra R, we have

where ∂+t⁢B is defined as

for x∈𝖧𝖧⁡(R/k) (this is basically saying that ∂(t)=0). Moreover, we have

Proof.

We will resolve k as an A-algebra by

where we declare that

Note that C also has the structure of a coalgebra with coproduct given by

This implies that

Here, we use the fact that ti is the dual of the element xi, and therefore because the differential on C is just b and lowers the x-degree by 1, dually it will be B and raise the t-degree by 1. The formula for 𝖧𝖯⁡(R/k) follows by inverting t. □

Remark 2.1.3.4.

This argument works more generally to obtain RHom⁡(k,H) for an arbitrary H∈𝖣𝖦𝖬𝗈𝖽A.

Remark 2.1.3.5.

A is a Hopf algebra with couint

and coproduct Δ:A→A⊗A given by

This implies that RHom¯A⁢(k,−) is a lax symmetric monoidal functor (here, note that k is the unit in the category of dg A-modules) and as such is given by

as a lax symmetric monoidal functor. Moreover, if C is an algebra object in 𝖣𝖦𝖬𝗈𝖽A, i.e. C is a dg algebra and B satisfies the Leibniz rule, then RHom¯A⁢(k,C) is a dg algebra.

Warning 2.1.3.6.

Even if R is commutative, B is not a derivation on 𝖧𝖧⁡(R), so we cannot simply argue that negative cyclic and periodic homology have products. This is true up to chain homotopy, however, but this is still not enough to obtain a product. We will see that B is a derivation up to coherent homotopy, but we run into issues with the formality of A, which will require working with S1 to resolve.

Instead of doing all of that, we will use the hack that (𝖧𝖧⁡(R/k)⁢⟦t⟧,∂+t⁢B) has a product up to chain homotopy. This implies that 𝖧𝖢∗−⁢(R/k) and 𝖧𝖯∗⁢(R/k) are graded commutative algebras.

Our goal now is to enhance the HKR theorem (Theorem 2.1.1.9). For this, we will need de Rham cohomology.

Definition 2.1.3.7.

Let R be a commutative k-algebra. The de Rham cohomology of R relative to k is defined as

Remark 2.1.3.8.

Since we are working with commutative rings (equivalently, affine schemes), every term in the de Rham complex is acyclic by vanishing of higher cohomology on affines, so we are just taking global sections. However, this invariant is more interesting for schemes X/k by upgrading the de Rham complex to a complex of sheaves.

Theorem 2.1.3.9.

Assume that ℚ⊆k and that LR/k is flat and concentrated in degree 0. Then there are natural isomorphisms

where |t|=−2 and

Exercise 2.1.3.10.

Describe the map 𝖧𝖢∗−→𝖧𝖯∗ in this case and check that it exhibits

Remark 2.1.3.11.

Because this is again relatively boring for rings, the interesting statement is that for a scheme X/k with LX.k being flat and concentrated in degree 0, we have

Remark 2.1.3.12.

We can drop the flatness assumption on R if we replace de Rham cohomology with a Hodge-completed derived de Rham cohomology following Bhatt-Morrow-Scholze, but this is of course extremely complicated.

Remark 2.1.3.13.

The statement of Theorem 2.1.3.9 is false if k is not characteric 0.

Proof.

We will give a classical proof. It is possible to give a fancier proof using nonabelian derived functors.

Consider a chain-level map

given by

Lemma 2.1.3.14.

This μ is an A-linear map of commutative dg algebras

Proof of Lemma.

We compute

Compatibility with B and d is given by the equation

We conclude with the following exercise.

Exercise 2.1.3.15.

Check that μ is multiplicative with respect to the shuffle product on the left and the wedge product of forms on the right. □

Using the lemma, we see that

is the identity, so the statement follows from Theorem 2.1.1.9. □

We are now ready to prove Theorem 2.1.3.9.

Proof of Theorem 2.1.3.9.

We simply compute

Applying the same strategy to negative cyclic homology, we obtain

We still want to deal with arbitary k (so not in characteristic 0).

Construction 2.1.3.16.

We have the Postnikov filtration

of 𝖧𝖧⁡(R/k) as an A-module, which is compatible with B and leads to a filtration of

𝖧𝖯⁡(R/k).

This leads to a multiplicative, conditionally convergent spectral sequence

The E3-page is given by

If R has flat cotangent complex, then this is

We will now return to the example of 𝔽p over ℤ. First, we need to upgrade divided powers to divided power series.

Definition 2.1.3.17.

Let R be a commutative ring. We define the divided power series algebra R⁢⟨⟨x⟩⟩ as the completion of R⁢⟨x⟩ at the filtration generated by the divided powers of x (note this is not an adic completion).

Proposition 2.1.3.18.

For R=𝔽p, we have

In addition, we have

where y has degree 0.

Remark 2.1.3.19.

Note that we have

which is obtained by adjoining divided powers of p to ℤp. Because ℤp already has divided powers of p, the element y[p]−ppp! is p-torsion. In fact, we have

where we should think of z as y−p.

What we have seen is that the negative cyclic and periodic cyclic homology of 𝔽p is very unpleasant. The situation will improve when we consider the topological version of everything. Also, in fact, 𝖧𝖯∗⁢(𝔽p/ℤ) is the 2-periodic derived de Rham cohomology of 𝔽p.

Proof.

Recall that 𝖧𝖧∗⁢(𝔽p/ℤ)≅𝔽p⁢⟨x⟩. We also recall the spectral sequence coming from the Postnikov filtration, which reads

The E2-page of the spectral sequence is given below, where the horizontal direction is powers of t and the vertical direction is divided powers of x. We also use dots to denote copies of 𝔽p.

−4−3−2−1001234∙∙∙∙∙∙∙∙∙

Because everything is in even degree, the spectral sequence collapses at the E2-page. Therefore, 𝖧𝖢∗−⁢(𝔽p/ℤ) has a filtration with associated graded given by 𝔽p⁢⟨x⟩⁢[t].

There is an additive extension problem and a multiplicative extension problem. We will obtain a copy of ℤp by using the diagonal (with slope 1) entries, and we also need to check that divided powers lift from the associated graded.

The connective cover of 𝖧𝖢−⁢(𝔽p/ℤ) can be represented by a simplicial commutative ring. Later, we will study everything using homotopy theory and see that the Connes operator comes from an S1-action on animated rings, so we can just take homotopy fixed points to obtain 𝖧𝖢−. In any case, the connective cover admits divided powers on positive degree homotopy groups. In particular, every choice of x,t∈𝖧𝖢∗−⁢(𝔽p/ℤ) yields a map

If we can find x,t such that 𝖧𝖢∗−⁢(𝔽p/ℤ), then this map induces an isomorphism on associated gradeds of the map

Because the filtration on 𝖧𝖢∗−⁢(𝔽p/ℤ) is complete, completing ℤ⁢⟨x⟩/(x⁢t−p) yields the desired result.

Now we will recall the computation of 𝖧𝖧⁡(𝔽p/ℤ). We first resolved 𝔽p by

In 𝖧𝖧⁡(𝔽p/ℤ), we have x=B⁢ε, ∂x=0, and B⁢x=0. It follows that in 𝖧𝖧⁡(𝔽p/ℤ), we have a cycle x representing x∈H⁢H∗⁢(𝔽p/ℤ). Using the fact that

we calculate

and therefore p=t⁢x in 𝖧𝖢∗−⁢(𝔽p/ℤ). □

Remark 2.1.3.20.

One can also deduce this entire computation using the fact that

Remark 2.1.3.21.

We have not talked at all about cyclic homology, but there is a long exact sequence

which computes 𝖧𝖢⁡(R/k). This comes from a cofiber sequence, or distinguished triangle.

2.1.4 The HKR theorem

We will now discuss the cotangent complex using nonabelian derived functors and prove the HKR theorem.

Let 𝒞 be a 1-category, which is generated by compact projective objects. Here, an object K in compact projective if

preserves filtered colimits and reflexive coequalizers, which are diagrams of the form

Note that this notion doesn’t really make sense for ∞-categories. Being generated by compact projective objects means that all

detect isomorphisms.

We can now pass from 𝒞 to the full subcategory on compact projective objects. If 𝒞=𝖲𝖾𝗍, then we get finite sets, and if 𝖢=𝖬𝗈𝖽R, then we get finitely generated projective modules. We can then construct the animation of 𝒞, which is given by

where 𝒮 denotes the ∞-category of anima (spaces or simplicial sets). Some properties of this include the following:

  • •

    The category 𝒞𝖼𝗉 of compact projective objects is a full subcategory of 𝖠𝗇𝗂⁡(𝒞), and so is the ind-category 𝖨𝗇𝖽⁡(𝒞𝖼𝗉).

  • •

    A general object of 𝒞 can be represented as the geometric realization of a simplicial diagram with entries in 𝖨𝗇𝖽⁡(𝒞𝖼𝗉).

  • •

    If X is compact projective object of 𝒞 and Yj are any ind-compact projective objects, then the mapping space

    is a simplicial set, so taking the geometric realization gives us a space.

Example 2.1.4.1.

If 𝒞=𝖲𝖾𝗍, functors are determined by their effect on the singleton set, so we just get 𝖠𝗇𝗂⁡(𝖲𝖾𝗍)=𝖲𝗉𝖺𝖼𝖾𝗌.

Example 2.1.4.2.

The non-negative derived category 𝒟⁢(𝒜)≥0 of an abelian category with enough compact projectives is equivalent to the animation 𝖠𝗇𝗂⁡(𝒜). One advantage of this description is that any functor out of 𝖠𝗇𝗂⁡(𝒜) preserving filtered colimits and reflexive coequalizers is simply a functor out of the 1-category 𝒞𝖼𝗉. For example, if we have a nonadditive functor

this will extend to a functor preserving filtered colimits and geometric realizations denoted by

Example 2.1.4.3.

For any commutative ring R, there is a functor

taking a module to its n-th exterior power. This is clearly nonadditive, but there is a derived functor

We cannot compute by resolving chain complexes, so instead we will represent an object by a simplicial diagram of projective modules and then applying ΛRn levelwise.

For example, consider R=ℤ. We will attempt to resolve ℤ/n, and this is resolved by the chain complex

Applying the Dold-Kan correspondence, we obtain a simplicial object

Exercise 2.1.4.4.
  1. 1.

    Find a levelwise projective simplicial abelian group whose complex is quasi-isomorphic to ℤ/n⁡[0].

  2. 2.

    Compute L⁢Λℤ2⁢(ℤ/n). Hint: compute Hi for i≤2 by hand and then find a different reason for the vanishing for higher i.

If we apply Λℤ2 to this, we obtain a simplicial diagram

which is simply given by

which is equivalent to ℤ/n⁡[1]. Note that this is not equivalent to applying Λℤ2 levelwise to the chain complex we started with, and that a functor commutes with Dold-Kan only if it is additive.

One advantage of this animation language is that we can discuss animation for categories which have no linear structure. For example, we have

Clearly we see that the ind-compact projective objects

are simply all polynomial rings. Therefore, we will represent objects by simplicial diagrams of polynomial rings. For example, we have

extracts the underlying simplicial set (in fact a simplicial abelian group). This give some version of non-negative complexes with ring structure.

Warning 2.1.4.5.

This does not agree with commutative ring objects in 𝒟⁡(ℤ).

The universal property is that

This allows us to define nonabelian derived functors by defining functors on polynomial rings which preserve filtered colimits.

Example 2.1.4.6.

The functor 𝖧𝖧(−/ℤ):𝖯𝗈𝗅𝗒→𝒟(ℤ)≥0 commutes with filtered colimits. Therefore, it extends to a functor

If we apply it to an ordinary commutative ring, this agrees with our previous definiion of Hochschild homology. In fact, it agrees for all animated rings by considering the associated dg-algebra.

Remark 2.1.4.7.

There is a way to define Hochschild homology in ∞-categories in much greater generality which does not require animated rings.

Question 2.1.4.8.

Is there a way to extend the HKR theorem to general rings using non-abelian derived functors?

Construction 2.1.4.9.

For any C∈𝒟⁡(ℤ), we can cut off homology groups using the functor τ≥n:𝒟⁡(ℤ)→𝒟⁡(ℤ). This produces a map

which has the following properties:

  • •

    The map is an isomorphism on Hi for i≥n;

  • •

    For all i<n, we have Hi⁢(τ≥n⁢C)=0.

One explicit construction of this is to replace everything in degree <n by 0 and to replace the degree n part with the cycles.

Exercise 2.1.4.10.

Construct the functor τ≥n:𝒟⁡(ℤ)→𝒟⁡(ℤ) using an adjunction.

We in fact have a map τ≥n+1⁢C→τ≥n⁢C, and the cofiber is given by

For polynomial rings R, we have a diagram

Therefore, we can try to compute Hochschild homology using the machinery of nonabelian derived functors by levelwise using the HKR theorem for this “filtration.”

Definition 2.1.4.11.

We define

as the geometric realization preserving extension of τ≥n𝖧𝖧(−/ℤ) from 𝖯𝗈𝗅𝗒.

We see that there is a cofiber sequence

for any animated ring R.

Lemma 2.1.4.12.

The nonabelian derived functor F𝖧𝖪𝖱n(−/ℤ) lands in 𝒟⁢(ℤ)≥n. More precisely, for any animated ring R, we have

Proof.

The lemma clearly holds for polynomial rings, and the extension to everything else follows from the fact that 𝒟⁢(ℤ)≥n is closed under filtered colimits. □

Lemma 2.1.4.13.

If a functor F:𝖼𝖱𝗂𝗇𝗀→𝖠𝖻 commutes with reflexive coequalizers, then

Proof.

We resolve R by a simplicial diagram of polynomial rings as

and note that R is the reflexive coequalizer of

Therefore, L⁢F⁢(R) is the total complex of

so taking H0, we obtain the reflexive coequalizer of

□

Exercise 2.1.4.14.

Show that the following functors commute with reflexive coequalizers:

  1. 1.

    The functor 𝖼𝖱𝗂𝗇𝗀→𝖲𝖾𝗍 given by R↦R×n;

  2. 2.

    The functor 𝖼𝖱𝗂𝗇𝗀→𝖠𝖻 given by R↦ℤ⁡[R×n];

  3. 3.

    The functor Ω−/ℤi:𝖼𝖱𝗂𝗇𝗀→𝖠𝖻. Hint: think about the generators and relations presentation for ΩR/ℤ1.

We are now ready to give an enhanced version of Theorem 2.1.1.9.

Theorem 2.1.4.15.

If L⁢ΩR/ℤn is concentrated in degree 0 for each n, then Theorem 2.1.1.9 holds for R.

Proof.

Using the long exact sequence in homology associated to

This implies that

is an isomorphism and that for all i>n, the induced map

is an isomorphism. Descending to R0, we obtain

We will omit checking that this is compatible with multiplication, and that in fact it is the map we had before. □

To better understand this, we need to better understand L⁢ΩR/ℤn. This will in fact agree with the value on R of the nonabelian derived functor

which is given by

for A∈𝖯𝗈𝗅𝗒/R. By definition, we have

Exercise 2.1.4.16.

Check that A↦R⊗AΩA/k1 takes compact projective objects in k−𝖠𝗅𝗀/R to compact projective objects in 𝖬𝗈𝖽R.

Applying nonabelian derived functors, we obtain

Proposition 2.1.4.17.

If L⁢ΩR/ℤ1 is concentraded in degree 0 and ΩR/ℤ1 is a flat R-module, then L⁢ΩR/ℤn is also concentrated in degree 0.

Proof.

If ΩR/ℤ1 is projective, then

In the general case, we use Lazard’s theorem, which states that every flat R-module is a filtered colimit of finitely generated projective modules. □

We are now able to state the correct form of Theorem 2.1.1.9.

Theorem 2.1.4.18 (Hochschild-Kostant-Rosenberg, ultimate version).

If L⁢ΩR/k1 is concentrated in degree 0 and ΩR/ℤ1 is a flat R-module, then

Remark 2.1.4.19.

For simplicity, we worked over ℤ, but everything still holds if we replace ℤ be an arbitrary commutative ring k.

2.1.5 Hochschild homology of schemes

Let k be a commutative ring and X be a scheme over k. We could in principle allow the base to be non-affine, but for simplicity we will ignore this. There are two approaches:

  1. 1.

    Extending from the affine case (by a right Kan extension);

  2. 2.

    Generalize the functor 𝖧𝖧(−/k) to k-linear dg categories (or stable ∞-categories) and then define define

    This is essentially an approach in the spirit of non-commutative geometry and is actually better in many cases (for example, we see that Hochschild homology is really an invariant of a stable ∞-category), but we will take the first approach in these notes since it is closer to the case of rings.

Definition 2.1.5.1.

For a scheme X, we define its Hochschild homology by

Note that if X=Spec⁡R is affine, then

because U=Spec⁡R is terminal in the poset of affine open subsets of Spec⁡R.

Remark 2.1.5.2.

We can also define Hochschild homology for stacks. For simplicity, we will not do this.

We will now discuss an example of a non-affine scheme. Let X=ℙk1. It has an open cover by

whose intersection is given by

We will hope for some kind of descent of Mayer-Vietoric sequence.

Theorem 2.1.5.3.

For any pair of open subsets U,V⊆X such that X=U∪V, the square

is a pullback in 𝒟⁡(k). In other words, Hochschild homology satisfies Zariski descent.

Remark 2.1.5.4.

Later, we will see that 𝖧𝖧(−/k) satisfies fpqc descent.

Corollary 2.1.5.5.

There is a long exact sequence

Lemma 2.1.5.6.

For any commutative ring R and x∈R, we have

Exercise 2.1.5.7.

Prove this.

In the example of ℙ1, we have a pullback

This becomes the long exact sequence

Therefore, it is easy to compute

Proof of Theorem 2.1.5.3.

We always have the square

For any open set U⊂X, a cofinality argument gives

Therefore, without loss of generality that X is affine. Unfortunately, affine schemes can have non-affine open subsets, but another cofinality argument allows us to assume that U and V are affine and are standard opens. Therefore, we have reduced to the case of

where 1∈(x,y). Using the fact that Hochschild homology commutes with localization, this becomes

Therefore, it suffices to prove that

which is a standard fact. This is true because

which is true because y acts invertibly on R⁡[x−1]/R. This is true because x is invertible on R/y. □

Our goal now is to prove an analog of Theorem 2.1.4.15 and Theorem 2.1.4.18 for schemes. We begin by defining a filtration on 𝖧𝖧⁡(X/k) by the formula

Remark 2.1.5.8.

Note that this cannot be done using the non-commutative approach.

Proposition 2.1.5.9.

This defines a complete filtration on 𝖧𝖧⁡(X/k), namely that

Moreover, the associated graded is given by

which is usually called the derived Hodge cohomology of X.

Note that if X is smooth, we just get the usual Hodge cohomology.

Proof.

Completeness follows from the fact that it is true for affines and that inverse limits commute with inverse limits. The assertion about (derived) Hodge cohomology follows again from the fact that cofibers commute with limits. □

Using the proposition, we obtain a spectral sequence

Corollary 2.1.5.10.

If ℚ⊆k, then there is a canonical isomorphism

Proof.

This is clearly true if X=Spec⁡k⁡[x1,…,xn]. Using non-abelian derived functors, this is also true for any affine X. The general case follows. □

Remark 2.1.5.11.

This isomorphism is in fact compatible with the B operator, so we get the desired results for periodic and negative cyclic homology.

2.2 Topological Hochschild and cyclic homology

2.2.1 Hochschild homology in ∞-categories

Hochschild homology can be constructed in an arbitrary symmetric monoidal ∞-category.

Recall that 𝖠𝗌𝗌𝖺𝖼𝗍⊗ is a 1-category with objects being finite sets and morphisms being maps of finite sets with a linear order on each preimage. Also recall that composition is given by the lexicographic order and that this category is symmetric monoidal with respect to the disjoint union.

In 𝖠𝗌𝗌𝖺𝖼𝗍⊗, the object ⟨1⟩ is an associative algebra object, which means it comes with a multiplication map

given by the left copy of ⟨1⟩ being smallest. The unit is the unique morphism from the empty set, and it is clear that multiplication is associative and unital. Therefore, we should think of 𝖠𝗌𝗌𝖺𝖼𝗍⊗ as a kind of free symmetric monoidal category on one associative algebra object, so as before, we define an associative algebra in an ∞-category 𝒞 to be a symmetric monoidal functor

Of course, the underlying object is simply the image of ⟨1⟩.

Exercise 2.2.1.1.

Prove that the category of symmetric monoidal functors 𝖠𝗌𝗌𝖺𝖼𝗍⊗→𝖠𝖻 is equivalent to the category of rings.

Recall that Hochschild homology was defined by writing a simplicial diagram. However, this is very hard to do by hand in an ∞-category, so we need a systematic way of handling ∞-categories.

Definition 2.2.1.2.

For a finite, nonempty totally ordered set S, define the set of cuts to be

We form the closely related set of cyclic cuts

Exercise 2.2.1.3.

Check that the functor Cut:Δop→𝖲𝖾𝗍 and Cut𝖢𝗒𝖼:Δop→𝖲𝖾𝗍 are represented by the simplicial sets Δ1 and S1.

The functor Cut𝖼𝗒𝖼 actually defines a functor Δop→𝖠𝗌𝗌𝖺𝖼𝗍⊗. For a map f:S→T of totally ordered sets and a cut (S0,S1)∈Cut𝖼𝗒𝖼⁡(S), the preimage

consists of all cuts on T “between” the images f⁡(S0) and f⁡(S1) if the cut is nontrivial. If (S0,S1) is trivial, then the preimage consists of all cuts outside of f⁡(S). To give this an order, in the first case we simply order the cuts from left to right. In the second case, we start with the cuts to the right of f⁡(S) and then go to the cuts to the left of f⁡(S).

Definition 2.2.1.4.

For an algebra A:𝖠𝗌𝗌𝖺𝖼𝗍⊗→𝒞, the Hochschild object is defined by

Lemma 2.2.1.5.

For an ordinary ring (or dg algebra), we obtain an algebra A in 𝒟⁡(ℤ). Then there is an equivalence

between the Hochschild complex of A as an algebra in 𝒟⁡(ℤ) and the Hochschild complex of A defined previously.

Proof.

The functor 𝖢𝗁⁡(ℤ)→𝒟⁡(ℤ) preserves tensor products of K-flat chain complexes. Therefore, if we take our dg algebra to be K-flat, the two sides of the lemma are computing the same object by the fact that in 𝒟⁡(ℤ), a colimit of a simplicial diagram is computed by a total complex. □

Exercise 2.2.1.6.

For an ordinary ring R, prove that the composite

agrees with the cyclic bar complex of R.

Definition 2.2.1.7.

For a ring spectrum A, define its topological Hochschild homology by

Recall that ordinary rings R have Eilenberg-Maclane spectra H⁢R, so we can take their topological Hochschild homology and denote it by 𝖳𝖧𝖧⁡(R).

Example 2.2.1.8.

Because 𝕊 is the unit in 𝖲𝗉, we have 𝖳𝖧𝖧⁡(𝕊)=𝕊 for the same reason that 𝖧𝖧∗⁢(ℤ)=ℤ.

This example suggests that we can view 𝖳𝖧𝖧 as 𝖧𝖧(−/𝕊), but it will take some work to make this precise. Before we do this, we will consider things like modules over algebras.

Definition 2.2.1.9.

Define the 1-category 𝖫𝖬𝗈𝖽𝖺𝖼𝗍⊗ in the following way:

  • •

    Objects are finite sets with elements labelled (“colored”) by {a,m};

  • •

    Morphisms are maps S→T together with a total ordering on preimages subject to the conditions that

    • –

      Preimages of a-colored elements are completely a-colored;

    • –

      Preimages of m-colored contain precisely one m-colored element and this element is the maximum.

We can define an analogous category 𝖱𝖬𝗈𝖽𝖺𝖼𝗍⊗ by the same definition, except the m-colored element of a preimage must be the minimum.

Definition 2.2.1.10.

A left module in a symmetric monoidal ∞-category 𝒞 is a symmetric monoidal functor

Note that if we only considered a-colored sets, we obtain the underlying algebra of the left module. Now recall that we can tensor left and right modules over a ring R to obtain an abelian group.

Definition 2.2.1.11.

Define a category 𝖫𝖱𝖬𝗈𝖽𝖺𝖼𝗍⊗ by the following:

  • •

    Objects are finite sets with colors given by {r,a,ℓ};

  • •

    Morphisms are maps f:S→T of finite sets with a total ordering on preimages such that

    • –

      Preimages of a-colored elements are entirely a-colored;

    • –

      Preimages of r-colored elements have precisely one r-colored element at the minimum and the rest are a-colored;

    • –

      Preimages of ℓ-colored elements have precisely one ℓ-colored element at the maximum and the rest are a-colored.

Because the r and ℓ-colored elements do not interact, we see that a symmetric monoidal functor 𝖫𝖱𝖬𝗈𝖽𝖺𝖼𝗍⊗→𝒞 is simply a pair of functors 𝖫𝖬𝗈𝖽𝖺𝖼𝗍⊗→𝒞 and 𝖱𝖬𝗈𝖽𝖺𝖼𝗍⊗→𝒞 which agree on 𝖠𝗌𝗌𝖺𝖼𝗍⊗.

Definition 2.2.1.12.

Define the tensor product N⊗AM to be the colimit of the composite

Here, the cut functor colors (∅,S) as r, (S,∅) as ℓ, and everything else as a.

There are variants of this theory, including bimodules. In this case, we relax the restriction on preimages of m-colored elements by dropping the condition that it must be either the minimum or maximum. Then Cut𝖼𝗒𝖼 factors through the category 𝖡𝗂𝖬𝗈𝖽𝖺𝖼𝗍⊗ by assigning the color m to the trivial cuts and the color a to everything else. Therefore, if M is a bimodule over A, we have

which is Hochschild homology of A with coefficients in M.

We can also discuss commutative algebras. Recall that 𝖢𝗈𝗆𝗆𝖺𝖼𝗍⊗=𝖥𝗂𝗇, which lets is define commutative algebras in 𝒞.

Lemma 2.2.1.13.

For a commutative algebra A, its Hochschild homology 𝖧𝖧⁡(A/𝒞) has a commutative algebra structure.

We will now study 𝖳𝖧𝖧 for ordinary rings. Eilenberg-Maclane spectra extend to a functor

coming from the Dold-Kan correspondence. It has the property that

This functor is lax symmetric monoidal and preserves colimits, so we obtain a canonical map

On homotopy groups, we obtain a morphism

The functor H canonically factors through an equivalence

which is symmetric monoidal. Therefore, we see that

Example 2.2.1.14.

If R is a ℚ-algebra, then the map 𝖳𝖧𝖧∗⁢(R)→𝖧𝖧∗⁢(R) is an isomorphism. This follows from the fact that

This implies that

so we obtain

Proposition 2.2.1.15.

For an ordinary ring R, the map

is an isomorphism for i≤2 and surjective for i≥3.

Proof.

The fiber of the map

is the geometric realization of a simplicial diagram of the form

The last term is 0 and the other terms are all 2-connective by a study of the effect of 𝕊→H⁢ℤ on homotopy groups. A technical argument (which we will not give here) shows that homotopy groups come from the last term, the suspension of the second-to-last term, the double suspension of the third-to-last term, and so on, so the geometric realization must be 3-connective. □

If we consider the ordinary ring 𝔽p, we see that

with generator x.

Theorem 2.2.1.16 (Bökstedt).

There is an isomorphism

In particular, we have no divided power structure. In addition, the map

is zero in degree ≥p.

2.2.2 Bökstedt periodicity

We will now study the spectrum 𝖳𝖧𝖧⁡(𝔽p).

Theorem 2.2.2.1 (Bökstedt).

As an 𝔼1-algebra over 𝔽p, 𝖳𝖧𝖧⁡(𝔽p) is free on one generator of degree 2. More explicitly, we have

Note that 𝖳𝖧𝖧⁡(𝔽p) is actually an 𝔼∞-algebra, but we are forgetting some of the structure in this theorem. Also, note that Ω⁢Σ⁢S2 is the free 𝔼1-algebra on S2, so

by a chain of adjunctions.

We should first prove that the two versions of Bökstedt’s theorem are equivalent. If we consider the element x∈𝖳𝖧𝖧2⁢(𝔽p), then we obtain a map of 𝔼1-algebras

This map is an equivalence if and only if it is an equivalence on homotopy groups. Finally, note that the homotopy groups of the source are simply

where x has degree 2. This is because

Using the Serre spectral sequence, this is 𝔽p in every even degree. Because Ω⁢S3 is an 𝔼1-algebra, we obtain the desired product structure.

Exercise 2.2.2.2.

Prove that π∗⁢(H⁢𝔽p⊗Σ+∞⁢X) agrees with H∗⁢(X,𝔽p) for any space X.

Lemma 2.2.2.3.

We have an equivalence

Sketch of proof.

Write R≃R⊗RR. This becomes

Therefore, we have

This construction identifies the rightmost and leftmost copies of R, so it gives us exactly the cyclic bar complex of R. □

We would like to apply this description to 𝔽p. First, we need to understand the spectrum

This is quite difficult (because the homotopy groups of spheres are extremely difficult), but it does have an interpretation as the homology groups of the Eilenberg-Maclane spectrum H⁢𝔽p, which form the dual Steenrod algebra.

Theorem 2.2.2.4 (Milnor).

As an algebra, we have

This is a very large description of the homotopy groups. Fortunately, there is a direct description of the spectrum H⁢𝔽p⊗𝕊H⁢𝔽p, which is an 𝔼∞-algebra over 𝔽p.

Lemma 2.2.2.5.

As an 𝔼2-algebra over H⁢𝔽p, H⁢𝔽p⊗𝕊H⁢𝔽p is free on one generator of degree 1. More explicitly, there is an equivalence

Proof of Theorem 2.2.1.16.

Write

The three terms in the tensor products come from the spaces pt and Ω2⁢S3, so we can form a bar construction with Ω2⁢S3 acting on two copies of a point. Therefore, we obtain

Because we formed the bar construction with an 𝔼2-algebra, we still have an 𝔼1-algebra structure. □

Remark 2.2.2.6.

The description of H⁢𝔽p⊗H⁢𝔽p is equivalent to Bökstedt’s theorem. This is because a map A→B of connected H⁢𝔽p-algebras is an equivalence if and only if the map

is an equivalence. We simply apply this to the map

which is given by specifying an element of π1.

Exercise 2.2.2.7.
  1. 1.

    Let A be a connected H⁢𝔽p-algebra. Prove that a map N→M of connective A-modules is an equivalence if

    is.

  2. 2.

    Let A→B be a map of connected H⁢𝔽p-algebras with an equivalence

    Prove that H⁢𝔽p⊗AB≃H⁢𝔽p and that A→B is an equivalence.

If p is odd, there are Dyer-Lashof operations

for any 𝔼∞-algebra over H⁢𝔽p. For any element of degree n, then

In addition, we have

whenever i<n2. Similarly, when i≤n2, we have

These should be viewed as coming from the homology of configuration spaces. On 𝔼2-algebras, we already have Qn2⁢x whenever n is even and Qn+12⁢x, β⁢Qn+12⁢x when n is odd. The first is simply the p-th power, but the existing power operations for odd n are an additional structure on 𝔼2-algebras.

Theorem 2.2.2.8 (Dyer-Lashof, Araki-Kudo).

We have

Note that this is very similar to Milnor’s result. All of the elements given by Dyer-Lashof have the same degrees as Milnor’s elements, so we might try to compare them.

Theorem 2.2.2.9 (Steinberger).

On π∗⁢(H⁢𝔽p⊗𝕊H⁢𝔽p), we have

and

Now it remains to try to construct a map realizing our desired isomorphism on homotopy groups.

Proof of Lemma 2.2.2.5.

We have a map

The target is simply Tor1ℤ⁡(𝔽p,𝔽p)≃𝔽p. This map is actually an isomorphism (because the map 𝕊→H⁢ℤ is 1-connected). Therefore, we obtain a map

of 𝔼2-algebras over H⁢𝔽p. This is an isomorphism on π1, but both sides are generated in the same way by the 𝔼2 Dyer-Lashof operations, so it is an isomorphism on all homotopy groups. □

Exercise 2.2.2.10.

Prove that π1⁢(H⁢𝔽p⊗𝕊H⁢𝔽p)≃𝔽p.

2.2.3 Properties of 𝖳𝖧𝖧

We will now consider formal properties of topological Hochschild homology. These include base change results, universality, and more.

Proposition 2.2.3.1.

The functor 𝖳𝖧𝖧:𝖠𝗅𝗀⁡(𝖲𝗉)→𝖲𝗉 is a (strong) symmetric monoidal functor. In particular, there is a natural equivalence

More generally, the functor

is symmetric monoidal.

Proof.

Consider the forgetful functor U:𝖠𝗅𝗀⁡(𝖲𝗉)→𝖲𝗉 as an object in the symmetric monoidal category 𝖥𝗎𝗇⁡(𝖠𝗅𝗀⁡(𝖲𝗉),𝖲𝗉). This is in fact an algebra object. Then we obtain

Corollary 2.2.3.2.

If R is an 𝔼n-algebra, then 𝖳𝖧𝖧⁡(R) is an 𝔼n−1-algebra.

Proof.

Recall that

Applying 𝖳𝖧𝖧 and using the fact that it depends only on the underlying 𝔼1-algebra structure, we land in 𝖠𝗅𝗀𝔼n−1⁢(𝖲𝗉). □

In particular, if R is an 𝔼∞-algebra, so is 𝖳𝖧𝖧⁡(R). However, there is a more canonical description. First recall that if A,B∈𝖢𝖠𝗅𝗀⁡(𝖲𝗉), then A⊗𝕊B is the coproduct in 𝖢𝖠𝗅𝗀⁡(𝖲𝗉).

Proposition 2.2.3.3.

For R∈𝖢𝖠𝗅𝗀⁡(𝖲𝗉), we have

where we are considering the constant diagram on R.

Exercise 2.2.3.4.

Prove this.

Warning 2.2.3.5.

This colimit is different from the colimit in the category of all spectra, which is given by

We will now prove a base change formula for 𝖳𝖧𝖧.

Proposition 2.2.3.6.

For any lax symmetric monoidal functor F:𝒞→𝒟 and A∈𝖠𝗅𝗀⁡(𝒞), we get a canonical map

If F is strong symmetric monoidal and preserves geometric realizations, then this map is an equivalence.

Proof.

We need to compare the cyclic bar constructions

and

The latter receives a map from

and this receives a map from the first colimit. The second part follows from the additional assumptions on F. □

Example 2.2.3.7.

Consider the Eilenberg-Maclane functor H:𝒟⁡(ℤ)→𝖲𝗉 which is lax symmetric monoidal. Therefore, we obtain a map

for any H⁢ℤ-algebra R.

Example 2.2.3.8.

If k→k′ is a map of commutative ring spectra, we have the functor

This is symmetric monoidal and preserves all colimits, so we have

We will now prove a more general version of Bökstedt periodicity. First, recall that an 𝔽p-algebra is perfect if the Frobenius morphism is an isomorphism. Examples include finite fields 𝔽pn and the ring

By definition, we have a map

inducing a map k→𝖳𝖧𝖧∗⁢(k) on homotopy groups. We can then precompose this with a map 𝖳𝖧𝖧⁡(𝔽p)→𝖳𝖧𝖧⁡(k), so we obtain a map

On homotopy groups, this gives us a map

Theorem 2.2.3.9 (Bökstedt periodicity for perfect rings).

For any perfect 𝔽p-algebra k, the map

is an isomorphism of commutative rings. Equivalently, 𝖳𝖧𝖧⁡(k) is the free 𝔼1-algebra on a degree 2 element x over H⁢k.

Proof.

For every perfect ring k, there is a commutative ring spectrum 𝕊W⁡(k) called the spherical Witt vectors such that

This should be thought of as a lift of k along 𝕊→H⁢𝔽p. This implies that

Therefore, it suffices to prove that

for all ∗≥1. This is left as an exercise. □

We will conclude with one more base change formula for 𝖳𝖧𝖧. Recall that if k is an 𝔼∞-ring, then 𝖳𝖧𝖧⁡(k)=k⊗S1, so we get a map 𝖳𝖧𝖧⁡(k)→k of 𝔼∞-rings by

along the projection S1→pt. In particular, the composition k→𝖳𝖧𝖧⁡(k)→k is the identity.

Remark 2.2.3.10.

This retract 𝖳𝖧𝖧⁡(k)→k does not exist if k is only 𝔼n for any finite n.

Theorem 2.2.3.11.

If k is a commutative ring spectrum and R is an associative k-algebra, then

Sketch of proof.

We can rewrite

2.2.4 The circle action on 𝖳𝖧𝖧

We will now consider the origin of the Connes operator on Hochschild homology from an actual action of §1 on topological Hochschild homology.

Recall that a group action on an object of a 1-category 𝒞 is a functor

where B⁢G is a category with a single object and automorphisms given by G.

Definition 2.2.4.1.

A group action of a group G on an object of a category 𝒞 is a functor

Note that N⁡(B⁢G) is a Kan complex with Ω⁡(N⁡(B⁢G))≃G. This means that the space associated N⁡(B⁢G) is the classifying space of G.

Example 2.2.4.2.

Note that B⁢ℤ≃S1.

Definition 2.2.4.3.

For any grouplike 𝔼1-algebra G in 𝒮 and an ∞-category 𝒞, the category of objects of 𝒞 with G-action is the functor category

Definition 2.2.4.4.

Define the homotopy orbits of an object X with G-action as the colimit

Similarly, define the homotopy fixed-points by the limit

In the category of spaces, if X has the trivial G-action, then we have

Exercise 2.2.4.5.

Prove this.

The homotopy orbits should be thought of as a derived version of a quotient construction. For example, if the action of G is free, then we get an actual quotient.

Proposition 2.2.4.6.

Objects in 𝒟⁡(ℤ) with an action of S1 are equivalent to A=Λℤ⁢(ε)-modules in 𝒟⁡(ℤ), where ε has degree 1.

Sketch of proof.

We use the fact that

We now need to compare C∗⁢(S1) to its homology A, or more precisely to prove that C∗⁢(S1) is formal. Here, we have maps

These maps are isomorphisms on homology in degrees 0 and 1 and they both factor through τ≤1⁢Free𝔼1⁢(ε). Because τ≤1 is lax symmetric monoidal on 𝒟⁢(ℤ)≥0, we are done. □

Remark 2.2.4.7.

This equivalence is not compatible with monoidal structures on both sides. It does not give the correct answer ∞-categorically when we consider 𝖧𝖢− and 𝖧𝖯, but it is fine on homology.

Theorem 2.2.4.8.

On 𝖧𝖧⁡(R/𝒞), there is a natural S1-action. More precisely, the Hochschild homology functor refines to a functor

which agrees with the S1-action on 𝖧𝖧⁡(R/ℤ) obtained from the Connes operator.

Definition 2.2.4.9.

The paracyclic category Λ∞ is a 1-category with objects given by totally ordered sets with a ℤ-action which are equivalent to sets of the form 1n⁢ℤ and morphisms given by equivariant order-preserving maps.

There is an S1≃B⁢ℤ-action on Λ∞ given by the map

acting trivially on objects and by the map

given by translation.

Definition 2.2.4.10.

The cyclic category Λ is the 1-category with the same objects as Λ∞ and morphisms given by

Lemma 2.2.4.11.

We have N⁡(Λ)=N⁢(Λ∞)h⁢S1.

This follows from the fact that the ℤ-action is free.

Lemma 2.2.4.12.

We have an equivalence

Proof.

We compute mapping spaces. We need to prove that

If we recall that mapping spaces were the maximal Kan complexes in functor categories, we can apply exponential laws. Therefore, it suffices to show that

which is just the definition of the colimit. □

Definition 2.2.4.13.

A cyclic object in 𝒞 is a functor N⁡(Λop)→𝒞. There is a functor Δ→Λ∞ sending [n−1] to the set ℤ×[n−1]≅1n⁢ℤ with the lexicographic order. The underlying object of a cyclic object is defined via the restriction

It is natural to wonder what the geometric realization of a cyclic object is.

Lemma 2.2.4.14.

The diagram

commutes. More precisely, the functor Δ→Λ∞ is cofinal.

Lemma 2.2.4.15.

For a cyclic object X, the object

admits an S1-action.

Proof.

Recall that

Taking the colimit, we obtain a functor

Exercise 2.2.4.16.

For an ∞-category 𝒞 with a trivial G-action, prove that

Proof of Theorem 2.2.4.8.

We now show that the cyclic bar complex is a cyclic object. Let

be the functor which sends a totally-ordered set S to the set of ℤ-equivariant cuts, which are just partitions of S of the form

such that each Si is a fundamental domain for the ℤ-action. We need to consider preimages along morphisms f:S→T in Λ∞. These are simply given by the set of all cuts between f⁡(S0) and f⁡(S1) ordered from left-to-right. The functor 𝖢𝗎𝗍ℤ is ℤ-invariant on morphisms, so it factors through

Restricting to Δop, we obtain 𝖢𝗎𝗍𝖼𝗒𝖼. □

Exercise 2.2.4.17.

Prove that 𝖢𝗎𝗍ℤ restricts to 𝖢𝗎𝗍𝖼𝗒𝖼.

2.2.5 𝖳𝖧𝖧 of the integers

In this slight digression, we will compute the topological Hochschild homology of the integers.

Theorem 2.2.5.1 (Bökstedt).

We have an isomorphism

In fact, 𝖳𝖧𝖧∗⁢(ℤ) is the homology of the dg algebra

where |x|=2, |e|=1, and the differential is given by ∂x=e and ∂e=0.

Note that we can compute the differentials as

Remark 2.2.5.2.

In fact, as an 𝔼1-algebra over H⁢ℤ, 𝖳𝖧𝖧⁡(ℤ) is the Eilenberg-Maclane functor applied to the above dg algebra.

As a consequence of this, we see that

is isomorphic on homotopy groups to the homology groups of the dg-algebra given by 𝔽p⁢[x]⊗Λ⁡(e) with ∂x=e and ∂e=0.

Exercise 2.2.5.3.

Compute the homology of the dg-algebra 𝔽p⁢[x]⊗Λ⁡(e) with ∂e=x and show that it has a nontrivial ring structure.

Definition 2.2.5.4.

For a ring spectrum R, define the p-completion 𝖳𝖧𝖧⁡(R,ℤp)≔𝖳𝖧𝖧⁢(R)p∧, the rationalization 𝖳𝖧𝖧⁡(R,ℚ)≔𝖳𝖧𝖧⁢(R)ℚ, and the ℚp version 𝖳𝖧𝖧⁡(R,ℚp)≔𝖳𝖧𝖧⁢(R,ℤp)ℚ.

By definition, we can use the Hasse square

to recover 𝖳𝖧𝖧⁡(R).

Lemma 2.2.5.5.

We have an equivalence

Moreover, we have

Proof.

First, we note that

The second statement follows for a similar reason. We calculate

As an example, we see that 𝖳𝖧𝖧⁡(ℤ,ℚ)=𝖧𝖧⁡(ℚ/ℚ)=H⁢ℚ. We then turn our attention to 𝖳𝖧𝖧⁡(ℤ,ℤp).

Theorem 2.2.5.6.

We have

This is the homology of the the dg-algebra ℤp⁢[x]⊗Λ⁡(e) with the differential ∂x=e.

Proof of Theorem 2.2.5.1 assuming Theorem 2.2.5.6.

Using Theorem 2.2.5.6, we get that 𝖳𝖧𝖧⁡(ℤ,ℚp)=H⁢ℚp. Therefore the fracture square takes the form

It is easy to see now that 𝖳𝖧𝖧0⁢(ℤ)=ℤ and

whenever n is nonzero. □

Definition 2.2.5.7.

We define an 𝔼∞-ring spectrum

To give a discrete commutative ring R the structure of an 𝕊⁡[z]-algebra is equivalent to giving an element π∈R (so we spend z→π). This is because any map 𝕊⁡[z]→H⁢R factors through H⁢ℤ⁢[x]. We will now consider ℤ as an 𝕊⁡[z]-algebra by sending z to p.

Warning 2.2.5.8.

If we consider a commutative ring spectrum R instead of a discrete ring, then an 𝔼∞-map 𝕊⁡[z]→R is not the same as an element z∈π0⁢R, or in other words 𝕊⁡[z] is not the free 𝔼∞ algebra on a single generator. However, it is free as an 𝔼1-algebra.

Theorem 2.2.5.9 (Relative Bökstedt periodicity).

We have

This has appeared in the work of Lurie, Bhatt-Morrow-Scholze, Antieau-Mathew-Nikolaus, and Krause-Nikolaus.

Proof.

We first note that

Now we study the long exact sequence

and deduce that the homotopy groups 𝖳𝖧𝖧∗⁢(ℤ/𝕊⁡[z],ℤp) are p-torsion-free and concentrated in even degrees. Therefore, there exists an element x∈𝖳𝖧𝖧2⁢(ℤ/𝕊⁡[z],ℤp) lifting x∈𝖳𝖧𝖧2⁢(𝔽p). This gives us a map

which is an isomorphism after mod-p reduction. Because these groups are p-torsion-free and p-complete, so this map is an isomorphism on the nose. □

Exercise 2.2.5.10.

Prove that 𝖳𝖧𝖧∗⁢(ℤ/𝕊⁡[z],ℤp) is p-torsion-free and concentrated in even degrees.

Recall that by the base change formula, we have

We can now consider the HKR filtration on 𝖳𝖧𝖧⁡(ℤ⁡[z]/ℤ), which induces a filtration on 𝖳𝖧𝖧⁡(R) and thus a multiplicative spectral sequence.

Proposition 2.2.5.11.

There is a convergent, multiplicative spectral sequence

Note that by definition, we have an equivalence

In addition, there is a similar spectral sequence in the p-complete world.

Proof of Theorem 2.2.5.6.

The spectral sequence takes the form

Using homological Serre grading, the only differentials are d2.

0123456011xx2x3d⁢zx⋅d⁢zx2⋅d⁢zx3⋅d⁢z

By the Leibniz rule, it is determined by

We see that 𝖳𝖧𝖧1⁢(ℤ,ℤp)=ℤp/α⁢ℤp. However, this is equal to

so α is a unit in ℤp. Up to changing x by a unit, we may assume that ∂x=dz. Because each degree has a single contribution, there are no extension problems. □

2.2.6 Negative topological cyclic homology

Definition 2.2.6.1.

We define negative topological cyclic homology as homotopy fixed points with respect to the S1-action on topological cyclic homology by the formula

Recall that an ∞-category is stable if it has finite limits and colimits and all pushout squares are also pullback squares.

Lemma 2.2.6.2.

If 𝒞 is stable, there is a unique lift

whoch is exact in both arguments.

In particular, objects in stable ∞-categories have mapping spectra.

Sketch of proof.

In 𝒞, we write

and therefore we have

which defines a spectrum. □

The proof of the following lemma is left as an exercise to the reader.

Lemma 2.2.6.3.

If I is a small ∞-category, then

  1. 1.

    The functor category 𝖥𝗎𝗇⁡(I,𝖲𝗉) is stable;

  2. 2.

    For any such functor, we have

We will now compare 𝖳𝖢− to the negative cyclic homology 𝖧𝖢−.

Proposition 2.2.6.4.

We have

Proof.

We will identify 𝒟⁡(ℤ) with 𝖬𝗈𝖽H⁢ℤ. In particular, we have

Unfortunately, the formality result we used does not behave well with respect to monoidal structures.

Lemma 2.2.6.5.

The functor (−)hS!:𝖥𝗎𝗇(BS1,𝖲𝗉)→𝖲𝗉 is lax symmetric monoidal and the §1-action on 𝖳𝖧𝖧⁡(R) (or 𝖧𝖧⁡(R)) is compatible with the product structure if R is an 𝔼∞-algebra. In particular, there is an 𝔼∞-algebra structure on 𝖳𝖢−⁢(R) (and 𝖧𝖢−⁢(R)) if R is an 𝔼∞-algebra.

If we want to do calculations, it would be helpful to understand how to compute with homotopy fixed points.

Lemma 2.2.6.6.

Let B⁢S1→𝖲𝗉 be an S1-action on H⁢A for an abelian group A. Then we have

with t of degree −2.

Proof.

The full subcategory of spectra on all Eilenberg-Maclane spectra (in degree 0) is a 1-category. In particular, we simply have

where we put the discrete topology on Hom⁡(A,B). Therefore, a functor B⁢S1→𝖲𝗉 which takes the basepoint to H⁢A is actually constant. Now we are looking for homotopy fixed points for a constant action, so they are given by

This has homotopy groups given by H∗⁢(B⁢S1,A), which gives the desired result. □

Exercise 2.2.6.7.

For a spectrum X, prove that

For a general spectrum, we can decompose it into its Postnikov tower. Unfortunately, we lose track of extension problems, but there is at least a spectral sequence.

Theorem 2.2.6.8 (Homotopy fixed points spectral sequence).

There is a multiplicative, conditionally convergent spectral sequence

Using negative Serre grading, this spectral sequence lives completely in the second quadrant and looks like the below.

−3−2−100123π0⁢(X)π1⁢(X)π2⁢(X)π3⁢(X)000π0⁢(X)⋅tπ1⁢(X)⋅tπ2⁢(X)⋅t000
Sketch of proof.

We consider the Whitehead tower

Applying homotopy fixed points, we obtain a sequence

which is not the same as taking the Whitehead tower of the homotopy fixed points. The homotopy cofibers are simply Eilenberg-Maclane spectra, so after taking homotopy fixed points they become of the form (H⁢πi⁢X)h⁢S1. Therefore we obtain a filtration on Xh⁢S1, which yields a spectral sequence. □

An S1-action on X gives us a map Σ+∞⁢S1⊗X→X. This uses the fact that S1 is a monoid in spaces, so Σ+∞⁢S1 is a ring spectrum. The canonical map

splits, so we have a decomposition

Restricting the action to the second summand, we obtain a map

or in other words a map

Lemma 2.2.6.9.

In the homotopy fixed points spectral sequence, the differential d2 is determined by the formula

for any α∈π∗⁢(X) and d2⁢t=η⁢t2.

Our goal is now to calculate 𝖳𝖢−⁢(𝔽p). The 𝔼2-page of the spectral sequence looks like

−4−3−2−1001234𝔽p𝔽p⋅x𝔽p⋅x2𝔽p⋅t𝔽p⋅t⁢x𝔽p⋅t⁢x2⋯⋯⋯

and therefore the spectral sequence degenerates at the 𝔼2-page. Therefore, 𝖳𝖢2⁢k−⁢(𝔽p) are complete filtered groups with associated graded a sequence of 𝔽ps.

Now choose lifts t~∈𝖳𝖢−2− and x~∈𝖳𝖢2− which lift t and x, respectively. Then 𝖳𝖢0−⁢(𝔽p)↠𝔽p is surjective and the kernel is generated as an ideal by t~⁢x~. Therefore, we have a relation

In order to compute a, we have a map, recall that we have a map

which is an isomorphism on τ≤2. Therefore we have an equivalence

The spectral sequence for 𝖧𝖢− showed that t~⁢x~=p, and thus a is a unit. Therefore, we may assume that a=1, so t~⁢x~=p, so we obtain a map

Theorem 2.2.6.10.

This map is an isomorphism.

Corollary 2.2.6.11.

We have

and is generated by x~k when k≥0 and t~−k when k≤0.

2.2.7 Topological periodic homology

We will now define a topological version of periodic homology. As a warning, this is not defined by inverting something and may not even be periodic.

Definition 2.2.7.1.

Define the topological periodic homology of a ring spectrum R as Tate fixed points by

Because taking Tate fixed points is lax symmetric monoidal in a way that is compatible with taking homotopy fixed points, we see that 𝖳𝖯⁡(R) is a 𝖳𝖢−⁢(R)-algebra if R is commutative. The homotopy groups of this spectrum can be computed using a spectral sequence.

Proposition 2.2.7.2.

For a spectrum X with G-action, there is a multiplicative, conditionally convergent spectral sequence

Proof sketch.

Consider the Whitehead filtration τ≥⁣∙⁢X on X. Applying Tate fixed points, it does commute with this limit even though it does not commute with limits or colimits in general. □

If we want to use this to calculate anything, we need to understand Tate fixed points of Eilenberg-Maclane spectra with respect to S1-actions. This can also be generalized to cyclic groups, but we will not pursue this direction.

For the spectrum H⁢ℤ, we can consider the cofiber sequence

Because the first term has only nonnegative homotopy groups and the second has only nonpositive homotopy groups, we can read off the homotopy groups of the Tate fixed points. In particular, we obtain

Lemma 2.2.7.3.

We have the ring structure

Proof.

Consider the Tate spectral sequence for H⁢ℤ⊗Σ+∞⁢S1. Because this has a free action, we have

The spectral sequence takes the form

−4−3−2−10123401ℤℤℤℤℤℤℤℤℤℤ

and so the only way for the spectral sequence to yield zero is for all maps d2 to be isomorphisms. Using the multiplicative structure, we do not get a ring but instead a module over the Tate construction on H⁢ℤ.

Fix generators xi of π2⁢i⁢H⁢ℤt⁢S1 with x−i=ti. Also fix generators σ0 and σ1 of H∗⁢(S1), so that the generator in degree (2⁢i,j) is xi⋅σj. We then compute

where the sign depends on our previous choices. This implies that

Therefore, up to a sign, we have xi⋅t=xi−1. We may choose the xi such that all signs are positive, but this now determines the ring structure. □

Exercise 2.2.7.4.

Using a similar approach to the one given here, compute π∗⁢(H⁢ℤt⁢Cm) as a ring. Hint: use a free action of Cm on a space with easy homology groups.

Remark 2.2.7.5.

If we run the same argument with H⁢A⊗Σ+∞⁢S1 as an H⁢ℤ-module, then we obtain

When A is a ring, this is also true as a ring.

Proposition 2.2.7.6.

If X is an object of 𝖥𝗎𝗇⁡(B⁢S1,𝖬𝗈𝖽H⁢ℤ) (equivalently a module over H⁢ℤ in 𝖥𝗎𝗇⁡(B⁢S1,𝖲𝗉)), then we have

In a weaker form, on homotopy groups we have

Sketch of proof.

We have an expression

as H⁢ℤh⁢S1-modules. Therefore, we need to prove that Xt⁢S1 is the same kind of colimit, or in other words that

This clearly works when X is an Eilenberg-Maclane spectrum. Both sides are compatible with fiber sequences, which gives us the case when X has finitely many nonzero homotopy groups. To extend to arbitrary X, we need some connectivity arguments, which we will omit here. □

Exercise 2.2.7.7.

Prove that

as H⁢ℤh⁢S1-modules.

Given that we are calling 𝖳𝖯 topological periodic homology, we would like to see how it is related to the usual periodic homology.

Proposition 2.2.7.8.

We have

Proof.

Because 𝖧𝖧⁡(R) is an H⁢ℤ-module, we see that 𝖧𝖧⁢(R)t⁢S1 is obtained from 𝖧𝖧⁢(R)h⁢S1=𝖧𝖢−⁢(R) by inverting t. This is exactly how we defined periodic cyclic homology. □

We will now return to topological periodic homology.

Proposition 2.2.7.9.

The Tate spectral sequence for Tate fixed points of S1-actions takes the form

Applying this to topological Hochschild homology gives us a spectral sequence

We are now ready to apply it to compute 𝖳𝖯⁡(𝔽p).

Theorem 2.2.7.10.

We have

Proof.

Note that the Tate spectral sequence is simply a periodized of the homotopy fixed points spectral sequence, at least on the 𝔼2-page. Therefore, it degenerates at the 𝔼2-page. The spectral sequence looks like

−4−3−2−10123401234𝔽p𝔽p⋅x𝔽p⋅x2𝔽p⋅t𝔽p⋅t⁢x𝔽p⋅t⁢x2𝔽p⋅t−1𝔽p⋅t−1⁢x𝔽p⋅t−1⁢x2⋯⋯⋯⋯⋯⋯

and therefore we have just periodized all of the extension problems. Using the map from 𝖳𝖢−, we see that

We will also choose a lift t−1~ of t−1. But now we see that

so it must be a unit. But then t~ is invertible and its negative powers t~−k generate π2⁢k⁢(𝖳𝖯⁡(𝔽p)). □

Warning 2.2.7.11.

If the Tate spectral sequence has differentials, then topological periodic homology may not be periodic. There are even examples with infinitely many differentials where 𝖳𝖯 becomes connective.

2.2.8 The cyclotomic structure

There is an additional structure on topological Hochschild homology which does now exist on ordinary Hochschild homology. This is a map

Construction 2.2.8.1.

Assume we have a fiber sequence

of pointed connected spaces. Because these are bar constructions of group objects, we write it as

Then there is a functor

which sends pt↦B⁢H which classifies B⁢G, i.e. G/H acts on B⁢H such that

Exercise 2.2.8.2.

For a normal subgroup H⊆G of a finite group, describe the G/H-action on B⁢H obtained from the fiber sequence

As a consequence, for any ∞-category 𝒞, we have an equivalence

With some additional work, this implies that the functors

induce functors

In particular, this means that for any normal subgroup H⊆G, if X has a G-action, then Xh⁢H and Xt⁢H have residual G/H-actions such that

Example 2.2.8.3.

Let G=𝕋=U⁡(1)=S1 and H=Cp⊆𝕋 be a cyclic group of prime order embedded as p-th roots of unity. For an object X∈𝒞B⁢𝕋, we see that Xh⁢Cp and Xt⁢Cp have residual 𝕋/Cp-actions. We identify 𝕋/Cp≃𝕋 by the map z↦zp.

Definition 2.2.8.4.

A cyclotomic structure on a spectrum X∈𝖲𝗉B⁢S1 is given by maps

which are 𝕋-equivariant for every prime p. We refer to φp as the cyclotomic Frobenius. A cyclotomic spectrum is a spectrum with a 𝕋-action and cyclotomic structure.

The main fact about cyclotomic spectra that we will discuss is the following.

Theorem 2.2.8.5.

For every ring spectrum R, 𝖳𝖧𝖧⁡(R) admits a natural cyclotomic structure.

Remark 2.2.8.6.

The cyclotomic structure on 𝖳𝖧𝖧⁡(R) is induced from the Tate diagonal

on spectra. It does not exist for 𝖧𝖧⁡(R/𝒞) where 𝒞 is an arbitrary stable symmetric monoidal ∞-category. In fact, one can prove that on 𝒟⁡(ℤ) there is no natural map

Therefore, 𝖧𝖧⁡(R/ℤ) does not admit any kind of cyclotomic structure.

We will now construct the cyclotomic structure. We will first give a construction in the case of an 𝔼∞-ring. First, we will need to discuss induction of G-objects. Let 𝒞 be an ∞-category with all small colimits. Then we have a forgetful functor

for any group G.

Proposition 2.2.8.7.

This functor has a left adjoint given by

with the G-action on G.

Proof.

Note that

Example 2.2.8.8.

Suppose that 𝒞=𝖲𝗉 and G is finite. Then

is the “free” object.

Example 2.2.8.9.

If G=𝕋, the free 𝕋-object is

Example 2.2.8.10.

This is the important example for the construction we want to perform. Let 𝒞=𝖢𝖠𝗅𝗀⁡(𝖲𝗉) and G be a finite group. If R∈𝖢𝖠𝗅𝗀⁡(𝖲𝗉), then

Proposition 2.2.8.11.

For R∈𝖢𝖠𝗅𝗀⁡(𝖲𝗉), the map

exhibits 𝖳𝖧𝖧⁡(R) as the free 𝔼∞-ring with an S1-action under R.

Proof.

By our general formula, the free object is given by

As discussed previously, this colimit is 𝖳𝖧𝖧⁡(R). We will omit the proof of the property that the S1-action agrees with the one previously constructed. □

Construction 2.2.8.12.

For every R∈𝖢𝖠𝗅𝗀⁡(𝖲𝗉), the map R→𝖳𝖧𝖧⁡(R) is a map of 𝔼∞-rings. The target has an action of 𝕋, so we obtain an induced Cp-action. Therefore, there is a unique extension a Cp-equivariant map

Taking Tate fixed points, we obtain a map

Also recall that the Tate diagonal is a map

of 𝔼∞ rings.

Exercise 2.2.8.13.

Check that the map R⊗p→𝖳𝖧𝖧⁡(R) agrees with the inclusion of the degree p part of the cyclic bar complex.

Using the universal property of 𝖳𝖧𝖧⁡(R), we obtain the cyclotomic Frobenius.

Proposition 2.2.8.14.

There is a unique 𝕋-equivariant map

of 𝔼∞-rings fitting into the diagram

We will now move on to the case when R is only an 𝔼1-algebra. Recall that by definition, we had

where 𝖳𝖧𝖧⁢(R)∙:Λ→𝖲𝗉 is a cyclic object. This can be written as

where R⊗n+1 carries an action of Cn+1. There is another cyclic object (sdp⁡𝖳𝖧𝖧⁡(R))t⁢Cp given by

and the cyclotomic structure on 𝖳𝖧𝖧⁡(R) arises by taking the levelwise Tate diagonal. The main complication comes from the way of viewing the target as a cyclic object.

Exercise 2.2.8.15.

For an ordinary ring R, explicitly describe a cyclic object whose underlying simplicial object has degree k part (R⊗(k+1)⁢p)Cp.

As a result, we obtain a map

2.2.9 The definition of 𝖳𝖢

We are finally ready to define topological cyclic homology. In fact, we will give two equivalent definitions, one of which is useful for theoretical purposes and one of which is useful for calcuations.

Definition 2.2.9.1.

The ∞-category of cyclotomic spectra is the fiber product

In particular, a map (X,φp)→(Y,φp′) is given by the data of

  1. 1.

    A 𝕋-equivariant map f:X→Y;

  2. 2.

    A 𝕋-equivariant homotopy filling the square

    for every prime p.

In the ∞-category of cyclotomic spectra, the mapping spaces are given by the equalizer

In fact, more is true. We will leave the following proposition as an exercise.

Proposition 2.2.9.2.

The ∞-category 𝖢𝗒𝖼𝖲𝗉 of cyclotomic spectra is a stable ∞-category and the mapping spectra are given by the equalizer

Moreover, 𝖢𝗒𝖼𝖲𝗉 has all limits and colimits. In addition, 𝖢𝗒𝖼𝖲𝗉 has a symmetric monoidal structure given by

In this monoidal structure, a commutative algebra in 𝖢𝗒𝖼𝖲𝗉 is given by

  1. 1.

    A commutative ring spectrum X∈𝖢𝖠𝗅𝗀⁢(𝖲𝗉)B⁢𝕋 with 𝕋-action;

  2. 2.

    A 𝕋-equivariant map of commutative algebras

    for all primes p.

In particular, if R is an 𝔼∞-ring, then 𝖳𝖧𝖧⁡(R) is a commutative algebra in 𝖢𝗒𝖼𝖲𝗉.

Now recall that negative topological cyclic homology was defined by

We will now give a similar description for 𝖳𝖢. In order to do this, we need to give 𝕊 the structure of a cyclotomic spectrum.

Example 2.2.9.3.

The sphere spectrum 𝕊 (with the trivial action) is naturally an 𝔼∞ cyclotomic spectrum.

  • •

    The underlying spectrum with 𝕋-action is 𝕊 with the trivial action;

  • •

    For any prime p, the map

    is the unit

Exercise 2.2.9.4.

Lift the diagram

to a diagram of 𝕋-equivariant maps.

Definition 2.2.9.5.

For a ring spectrum R, define the topological cyclic homology 𝖳𝖢⁡(R) as the mapping spectrum

If R is commutative, then this is an 𝔼∞-ring spectrum. As usual, we will denote the homotopy groups by π∗⁢(𝖳𝖢⁡(R))≕𝖳𝖢∗⁢(R).

More generally, for any cyclotomic spectrum X, we write

This means that 𝖳𝖢⁡(R)=𝖳𝖢⁡(𝖳𝖧𝖧⁡(R)).

We will now explicitly evaluate this formula. We compute

where φph⁢𝕋:Xh⁢𝕋→(Xt⁢Cp)h⁢𝕋 is the cyclotomic structure and can:Xh⁢𝕋=(Xh⁢Cp)h⁢𝕋→(Xt⁢Cp)h⁢𝕋 is the canonical map. In this calculation, we have used the following theorem.

Theorem 2.2.9.6.

Assume that X is bounded below with 𝕋-action. Then the canonical map

exhibits the right hand side as the p-completion of the left hand side. Moreover, if X is p-complete, then so is Xt⁢𝕋 (or in other worse the map is an equivalence).

As a result, if X is a cyclotomic spectrum whose underlying spectrum is bounded below, then

We can assemble all of the φp into a map

Corollary 2.2.9.7.

If X∈𝖢𝗒𝖼𝖲𝗉 is bounded below, then

In particular, we have

if R is a connective ring spectrum.

We will briefly discuss the non-affine situation. If X is a quasi-compact quasi-separated scheme, then we define

Therefore, we have

and

Finally, we define

Remark 2.2.9.8.

One can show that for an n-connective X∈𝖢𝗒𝖼𝖲𝗉 that 𝖳𝖢⁢(X)p∧≕𝖳𝖢⁡(X,ℤp) is (n−1)-connective. In particular, if R is a ring (or connective ring spectrum), then 𝖳𝖢⁡(R,ℤp) is (−1)-connective.

Proof of Theorem 2.2.9.6.

Note that both ((−)t⁢𝕋)p∧ and ((−)t⁢Cp)h⁢𝕋 are exact and commute with the Postnikov tower. Therefore, we can reduce to the case when X=H⁢A. Resolving A by a two-term complex of free groups, we may assume that A is free. Then the Tate cohomology H⁢At⁢Cp is p-torsion and therefore p-complete. This implies that (Xt⁢Cp)h⁢𝕋 is p-complete. If X itself is p-complete, then Xt⁢𝕋 is p-complete.

We now need to prove that (Xt⁢𝕋)p∧≃(Xt⁢Cp)h⁢𝕋. Because both sides are p-complete, we can check this after reducing mod p. Therefore, we can assume that A=⨁𝔽p is an 𝔽p-vector space. We will now carry out the argument in the case when A=𝔽p.

We need to compare the homotopy groups

with the homotopy groups

Recall that when p is odd, we have

In addition, we have

We will now use the homotopy fixed points spectral sequence to compute π∗⁢((H⁢𝔽pt⁢Cp)h⁢𝕋).

The spectral sequence for ((H⁢𝔽p)t⁢Cp)h⁢𝕋=H⁢𝔽ph⁢𝕋 is simply

It is given by

−4−3−2−10−4−3−2−101ete⁢tt2xx2

We see that 1 survives. However, 𝔽p⁢[t] is even, so e must be killed and thus d2⁢(e)=x. Applying the Leibniz rule, we obtain more differentials, which give us 𝔽p⁢[t]. The Tate spectral sequence is similar, except we need to invert t. It looks like

−4−3−2−10−4−3−2−1012341ete⁢tt2e⁢t−1t−1e⁢t−2t−2xx2

There is a map of spectral sequences from the homotopy fixed points spectral sequence to the Tate spectral sequence, so we can import differentials. Therefore, we see that d2⁢(e)=x, d2⁢(t)=0, and d2⁢(x)=0. Therefore, the spectral sequence converges to 𝔽p⁢[t±]. Because there are no convergence problems, we are done. □

2.2.10 𝖳𝖢 of 𝔽p

Continuing our running example, we will now compute 𝖳𝖢⁡(𝔽p). Recall that we had

for generators x in degree 2 and t in degree −2 and that

is equal to its profinite completion. Then the canonical map

is given by t↦t and x↦p⁢t−1. Our goal now is to understand the map

Lemma 2.2.10.1.

There exists a unit α∈ℤp× such that φ⁡(x)=α⁢t−1 and φ⁡(t)=α−1⁢p⁢t.

Remark 2.2.10.2.

A more careful argument shows in fact that α=1.

Proof.

We have the commutative diagram

We will now chase the element t through this diagram. In the map φ, we must have φ⁡(x)=α⁢t−1 and φ⁡(t)=μ⁢t. Of course our goal is to prove that α is a unit.

Chasing t∈𝖳𝖢−2−⁢(𝔽p) through the diagram clockwise, it must be zero because 𝖳𝖧𝖧⁡(𝔽p) is connective. If we recall that 𝔽pt⁢Cp≅𝔽p⁢[t±]⊗Λ⁡(e), the counterclockwise composition is [μ]∈𝔽p. Because the diagram commutes, we know that μ=p⁢μ′ for some μ′∈ℤp. Therefore, we see that

and therefore that α⁢μ′=1. □

Our goal now is to prove that 𝖳𝖢∗⁢(𝔽p) using the long exact sequence associated with the fiber sequence

This is the sequence

Up to a unit, all of the maps in degree not equal to zero are of the form 1−p⁢β:ℤp→ℤp.

Exercise 2.2.10.3.

Prove that the map 1−α⁢p:ℤp→ℤp is an isomorphism for any α∈ℤp.

Theorem 2.2.10.4.

We have

Remark 2.2.10.5.

As a ring, we must have 𝖳𝖢∗⁢(𝔽p)=ℤp⁢[ε]/ε2, where ε has degree −1. As an 𝔼∞ ring spectrum, we have

Remark 2.2.10.6.

Quillen proved that the p-completion of the algebraic K-theory of 𝔽p is

In particular, K⁡(𝔽p,ℤp) is the connective cover τ≥0⁢𝖳𝖢⁢(𝔽p). This is one important computation in the use of trace methods in algebraic K-theory.

We will now turn to the consequences of the computation of 𝖳𝖢⁡(𝔽p).

Corollary 2.2.10.7.
  • •

    There is an 𝔼∞-map H⁢ℤ→𝖳𝖢⁡(𝔽p);

  • •

    For any 𝔽p-algebra A, 𝖳𝖢⁡(A) is an H⁢ℤ-module spectrum.

Proof.

The first claim follows from the fact that the connective cover is a map H⁢ℤp→𝖳𝖢⁡(𝔽p) and the second claim follows from the fact that 𝖳𝖢⁡(A) is a module over 𝖳𝖢⁡(𝔽p). □

The following proposition is left as an exercise.

Proposition 2.2.10.8.

We have an adjunction

where Xtr has the trivial 𝕋-action and cyclotomic Frobenius given by φp:X→Xh⁢Cp→Xt⁢Cp. The left adjoint is symmetric monoidal and the right adjoint is lax symmetric monoidal.

Corollary 2.2.10.9.

We have a map of 𝔼∞-rings

In the relative case, we have that for any 𝔽p-algebra A, 𝖳𝖧𝖧⁡(A) is naturally an H⁢ℤtr-module.

Warning 2.2.10.10.

The map

is different from the obvious map

as maps of spectra. In fact, the second map cannot be made into a map of cyclotomic spectra.

To see this, we will show that postcomposing the two maps with

are different, or in other worse that r∘f≄r∘g.

For the map f, consider the diagram

Therefore, r∘f is equala to the composite

For the map g, consider the diagram

Therefore, the map r∘g is equivalent to the composite

Theorem 2.2.10.11.

The maps H⁢𝔽p→trH⁢𝔽pt⁢Cp and the Frobenius map H⁢𝔽p→H⁢𝔽pt⁢Cp differ by the Steenrod operations.

Corollary 2.2.10.12.

We have r∘f≄r∘g.

We will restate the theorem about being an H⁢ℤtr-module in a more algebraic way.

Definition 2.2.10.13.

We define the ∞-category 𝖢𝗒𝖼⁢𝒟⁢(ℤ) of cyclotomic chain complexes as the fiber product

The following characterization of 𝖢𝗒𝖼⁢𝒟⁢(ℤ) is left as an exercise.

Proposition 2.2.10.14.

We have an equivalence

Corollary 2.2.10.15.

For an 𝔽p-algebra A, 𝖳𝖧𝖧⁡(A) is naturally an object of 𝖢𝗒𝖼⁢𝒟⁢(ℤ).

Warning 2.2.10.16.

(Ordinary) Hochschild homology (of an ordinary ring) cannot be made into a cyclotomic chain complex.

2.2.11 𝖳𝖢 of perfect rings

Our goal is now to generalize the computation of 𝖳𝖢⁡(𝔽p) to any perfect ring of characteristic p. In this generality, we will need to replace ℤp by the Witt vectors. Recall that if k is a perfect 𝔽p-algebra, then

Recall that there is a commutative ring W⁡(k) called the p-typical Witt vectors of k which is p-torsion free, p-complete, and has W⁡(k)/p≃k. Moreover, it is uniquely determined by these properties.

Theorem 2.2.11.1.

Let R be any derived p-complete ring. Then the reduction mod p map

is a bijection.

Proof.

If R=lim←⁡R/pn is p-complete, then we use obstruction theory to inductively lift maps through the square-zero extension R/pn+1→R/pn. For more details, see Section A.2.2. If R is derived p-complete, then

as an animated ring. Therefore, as in the first step we can reduce to R⫽p. We conclude by using the fact that R⫽p→R/p is a square-zero extension. □

The following corollary is left as an exercise.

Corollary 2.2.11.2.

We have an equivalence of categories

Corollary 2.2.11.3.

There is a ring isomorphism

called the Witt vector Frobenius such that

is the Frobenius.

Proposition 2.2.11.4.

For a perfect 𝔽p-algebra k, we have natural isomorphisms

Remark 2.2.11.5.

Note that W⁡(𝔽p)=ℤp, so this matches our previous computation.

Proof.

Consider the homotopy fixed points spectral sequence for 𝖳𝖧𝖧⁢(k)h⁢𝕋. This is given by

−6−5−4−3−2−1001234561tt2t3xx2x3

where all dots are given by k. It now remains to solve the extension problems. There is a map of spectral seqeunces from the homotopy fixed points spectral sequence for 𝖳𝖧𝖧⁢(𝔽p)h⁢𝕋 induced from 𝔽p→k, so we must have

However, x⁢t cannot be a zero divisor because this is true on the spectral sequence, so 𝖳𝖢0−⁢(k) is p-torsion free and p-complete with mod-p reduction k. Therefore, we have

The rest follows because the spectral sequences are functorial. Periodizing the spectral sequences, we obtain the result for 𝖳𝖯. □

We will now need to understand the maps

We know that can⁡(x)=p⁢t−1, can⁡(t)=t, and can⁡(r)=r for r∈W⁡(k). We also know that φ⁡(x)=t−1 and φ⁡(t)=p⁢t. It remains to study its effect on W⁡(k).

Remark 2.2.11.6.

For any connective ring spectrum, the map π0⁢can:𝖳𝖢0−⁢(R)→𝖳𝖯0⁢(R) is an isomorphism for degree reasons. Thus we can consider π0⁢φ as an endomorphism of 𝖳𝖢0−⁢(R)=𝖳𝖯0⁢(R). This is one of the beginnings of the theory of prismatic cohomology.

Proposition 2.2.11.7.

The map π0⁢φ:W⁡(k)→W⁡(k) is the Witt vector Frobenius F.

Proof.

Consider the diagram

On π0, this diagram induces the diagram

Our goal is to understand the composite of the bottom row of arrows. We know that the composite

is the 𝔼∞ Frobenius. □

Exercise 2.2.11.8.

Prove that for a commutative ring R, the 𝔼∞ Frobenius R→Rt⁢Cp induces the usual Frobenius morphism R→R/p on π0.

Now we calculate 𝖳𝖢∗⁢(k) using the long exact sequence associated with the fiber sequence

The map can−φ:𝖳𝖢∗−⁢(k)→𝖳𝖯∗⁢(k) is given as follows:

  • •

    In degree 2⁢n for n≥0, it is the map

  • •

    In degree −2⁢n, it is the map

  • •

    In degree 0, it is simply id−F.

Note that these maps are isomorphisms whenever n≠0. We have thus proved the following theorem:

Theorem 2.2.11.9.

For a perfect 𝔽p-algebra k, we have

This formula can be evaluated in various examples using several facts:

  • •

    There is an isomorphism W⁢(k)F≅W⁡(kφ).

    Corollary 2.2.11.10.

    If k is a field, then 𝖳𝖢0⁢(k)=ℤp.

    Proof.

    Clearly we have kφ=𝔽p. □

  • •

    We have kφ=C0⁢(X,𝔽p) for a profinite space X. This is a version of Stone duality. In fact, X is the underlying topological space of Spec⁡kφ. Equivalently, it is the space of connected components of Spec⁡k.

  • •

    We have

    Exercise 2.2.11.11.

    Prove the third fact from the second.

    Corollary 2.2.11.12.

    For a perfect 𝔽p-algebra, we obtain

  • •

    The degree −1 part 𝖳𝖢−1⁢(k)=W⁢(k)F is p-torsion free and its mod-p reduction is kφ. Thus 𝖳𝖢−1⁢(k) is p-completely free of rank the dimension of kφ, or in other words

    where kφ=⨁S𝔽p.

    Corollary 2.2.11.13.

    If kφ=0, for example if k is an algebraically closed field, then 𝖳𝖢−1⁢(k)=0.

    As an example, we see that

  • •

    Suppose that kφ=𝔽p×⋯×𝔽p is a finite product of copies of 𝔽p. Then kφ is a kφ-module and can be described as

    for some 𝔽p-vector spaces V1,…,Vi. This implies that

    as a W⁢(k)F=ℤp×⋯×ℤp-module, where V~k is a p-complete, p-torsion free lift of Vi to ℤp.

2.2.12 Frobenius lifts and group rings

We will now study cyclotomic spectra with Frobenius lifts, i.e. spectra where the Frobenius factors through homotopy fixed points. Some examples are Xtr, but more interesting examples are spherical group rings. In these cases, we will give a more explicit formula for 𝖳𝖢.

Definition 2.2.12.1.

A p-cyclotomic spectrum with Frobenius lift is a p-cyclotomic spectrumm X with a 𝕋-equivariant factorization

of the Frobenius.

Example 2.2.12.2.

By definition, the cyclotomic spectrum Xtr with the trivial action has a Frobenius lift because φp was defined as the composite

This can be viewed as the map

If X has a Frobenius lift, the map φ:Xh⁢𝕋→Xt⁢𝕋 factors as

In particular, the map φ−can:Xh⁢𝕋→Xt⁢𝕋 factors as

We may now try to use this formula to study 𝖳𝖢.

Theorem 2.2.12.3.

In our setting, where X is a p-complete, bounded below, p-cyclotomic spectrum with Frobenius lift, then we have a pullback diagram

Proof sketch.

Recall that 𝖳𝖢 is the homotopy fiber of the map φ−can. Therefore, we have a diagram

so the square on the left is a pullback. The proof that the bottom square is a pullback is omitted. □

Corollary 2.2.12.4.

We have the formula

Exercise 2.2.12.5.

Prove this.

The spectrum fib⁡(Σ⁢𝕊h⁢𝕋→tr𝕊) is usually called Σ⁢ℂ⁢ℙ−1∞. Here, if we desuspend it, it has cells in even degrees starting in degree −2, hence the name.

We will now turn to the case of spherical group rings. Let G be an 𝔼1-group in 𝒮. Recall that Ω and B give an equivalence between these and pointed connected spaces. We will define the group ring to be the spectrum

In order to understand 𝖳𝖢, we will need to understand 𝖳𝖧𝖧.

Lemma 2.2.12.6.

We have the formula

where L denotes the free loop space.

Proof.

Because 𝖳𝖧𝖧⁡(𝕊⁡[G]) was defined by the cyclic bar construction, we can pull the suspension out of the colimit and obtain

We now compare 𝖡𝖺𝗋𝖼𝗒𝖼⁢(G) with the cyclic object 𝖥𝗎𝗇⁡(S∙,B⁢G), where Sn is the groupoid given by a cycle on {0,…,n}. Unfortunately, 𝖥𝗎𝗇⁡(Sn,B⁢G) is not G×n, so there is no levelwise isomorphism. However, it is equivalent to Gh⁢G×n×n, where G×n acts by conjugation on adjacent factors.

This implies that |Fun⁡(S∙,B⁢G)| is obtained from |𝖡𝖺𝗋𝖼𝗒𝖼⁢(G)| by taking orbits under

where the boundary maps are projections. This is contractible. Now we have

Exercise 2.2.12.7.

Prove that for a space X, the simplicial space

has contractible realization.

Proposition 2.2.12.8.
  1. 1.

    The 𝕋-action on 𝖳𝖧𝖧⁡(𝕊⁡[G])=Σ+∞⁢L⁢B⁢G comes from the 𝕋-action on L⁢B⁢G given by rotating loops.

  2. 2.

    The Frobenius morphism Σ+∞⁢L⁢B⁢G→Σ+∞⁢L⁢B⁢Gt⁢Cp admits a lift Σ+∞⁢L⁢B⁢G→Σ+∞⁢L⁢B⁢Gh⁢Cp coming from the Cp-invariant map L⁢B⁢G→L⁢B⁢G defined by precomposition with the degree p endomorphism of S1.

The proof of this uses the fact that the Tate diagonal 𝕊⁡[G]→(𝕊⁢[G]⊗p)t⁢Cp lifts through the actual diagonal G→G⊗p.

Proposition 2.2.12.9.

For an 𝔼1-group G such that π0⁢(G) is a p-group, the map

fits into a square

which is a pullback after p-completion.

Proof.

It suffices to p-complete the cofibers of the two rows and check that they are equivalent. They are given by (Σ+∞⁢L⁢B⁢G)h⁢ℕ≃Σ+∞⁢(ψp−1⁢L⁢B⁢G)h⁢ℤ and (Σ+∞⁢B⁢G)h⁢ℕ≃(Σ+∞⁢B⁢G)h⁢ℤ, respectively. Therefore, it suffices to prove that

is an equivalence on 𝔽p-homology.

To prove this, we will consider fiber sequences in spaces. More precisely, consider the sequence

of fiber sequences. In the colimit, we obtain the fiber sequence1

Now the homology H∗⁢(G0,𝔽p) is a connected Hopf algebra and p induces the map

This map is not necessarily zero, but it is degreewise nilpotent. □

Exercise 2.2.12.10.

Let A be a connected Hopf algebra over 𝔽p. Prove that the map

is degreewise nilpotent.

Using all of our results, we have proved the following.

Corollary 2.2.12.11.

For an 𝔼1-group G with π0⁢(G) a p-group, we have a pullback square

In particular, we have the formula

A Simplicial rings and cotangent complexes

A.1 Simplicial rings and divided power structures

Definition A.1.0.1.

The category 𝗌𝖼𝖱𝗂𝗇𝗀 of simplicial commutative rings is defined to be the functor category

More precisely, this is a diagram

satisfying the usual conditions for face and degeneracy maps.

Theorem A.1.0.2 (Eilenberg-Zilber).

The functor

sending a simplicial abelian group to the associated chain complex is lax symmetric monoidal with respect to the pointwise tensor product on simplicial abelian groups and the usual tensor product on chain complexes.

This implies that every simplicial commutative ring produces a commutative dg algebra. We will now make this product structure explicit.

Definition A.1.0.3.

An (m,n)-shuffle is a function

satisfying the following conditions:

  • •

    Both μ and ν are surjective and order preserving;

  • •

    The maps μ and ν jump at disjoint positions. For example, consider the case when m=n−2, as in Figure A.1.

    0m+n0m0n
    Figure A.1: A (2,2) shuffle. Red lines denote where μ jumps and blue lines denote where ν jumps.

    We can see that an (m,n)-shuffle is simply a way to shuffle m red and n blue cards together. There is a unique permutation which takes the configuration having all red cards on top to our given shuffle.

Proposition A.1.0.4.

Let R∈𝗌𝖼𝖱𝗂𝗇𝗀. Then there is a map Rm⊗Rn→⋅Rm+n given by the formula

where sμ⁢(x) and sν⁢(y) correspond to applying the degeneracies making up μ and ν to obtain elements of Rm+n. This gives R the structure of a strict cdga.

Proof.

We will not go through all of the combinatorics explicitly.

  • •

    To prove associativity, we can express both (x⁢y)⁢z and x⁡(y⁢z) as a sum

  • •

    To check commutativity, the fact that R was a simplicial commutative ring means we can exchange the roles of μ and ν up to

    which is exactly what we desire.

  • •

    A variant of this argument also gives strictness. When we compute x2, the summands corresponding to (μ,ν) and (ν,μ) appear with opposite signs, so we see that x2=0.

  • •

    The differential is given by the alternating sum of the face maps. Applying this to x⋅y will decompose the result into (m−1,n)-shuffles and (m,n−1)-shuffles, which yields the desired result. □

We will now discuss divided power structures. In fact, every simplicial commutative ring will give us one of these.

Proposition A.1.0.5.

Let R∈𝗌𝖼𝖱𝗂𝗇𝗀 be a simplicial commutative ring. Then there exist maps γk:Rn→Rn⁢k for all n,k≥1 such that

  • •

    We have γ1⁢(x)=x for all x∈R;

  • •

    We also have

    for all k,ℓ. This means that we should think that xk=k!⁢γk⁢(x);

  • •

    For all x,y∈Rn, we have

  • •

    Similarly, we have

  • •

    We have the identity

Proof.

If x is odd, we define γ1⁢(x)=x and γk⁢(x)=0 for all k≥2. This is fine because x2=0 already. If x is even, we have

We can permute the factors μ1,…,μk (all of the signs will become +1), and therefore for all σ∈Σk, (μ1,…,μk) and (μσ⁡(1),…,μσ⁡(k)) give the same summand. Choosing one element from each equivalent class, we obtain an element γk⁢(x) such that xk=k!⁢γk⁢(x).

We will omit checking that the properties are satisfied. □

We now need to make sure that this divided power structure behaves well with respect to the chain differential.

Proposition A.1.0.6.

Let R∈𝗌𝖼𝖱𝗂𝗇𝗀 be a simplicial commutative ring. Then

  • •

    If x is even, then dγk⁡(x)=γk−1⁢(x)⋅dx;

  • •

    The maps γk preserve boundaries;

  • •

    The γk give a well-defined divided power structure on H∗⁢(R).

Proof.

Suppose that x is a boundary. Then x=dy for x with di⁢y=0 for all i>0. Define

Here, for maps μi:[k⁢n]→[n], we set μi′:[1+k⁢n]→[1+n] to be the maps given by sticking a 1 in front. Then d0⁢γk′⁢(y)=γk⁢(x) and di⁢γk′⁢(y)=0 for all i>0, and therefore dγk′⁡(y)=γk⁢(x). □

A.2 Cotangent complexes and obstruction theory

A.2.1 Definition of the cotangent complex

Recall that we defined

as the nonabelian derived functor of

Lemma A.2.1.1.

The following constructions agree:

  1. 1.

    The functor

    evaluated on R;

  2. 2.

    The functor

    evaluated on R.

Proof.

If we begin with the second construction, we obtain

with Ai polynomial rings. Here, we have used the fact that identity morphisms are cofinal in the arrow category. □

The second functor has better categorical properties than the first. Namely, it commutes with coproducts. By definition, this implies that it preserves all colimits.

Example A.2.1.2.

If R=k⁡[x1,…,xn]/(f1,…,fm) is defined by a regular sequence, then the diagram

is a pushout in 𝖠𝗇𝗂⁡(k−𝖺𝗅𝗀/R). The functor L(R⊗∙Ω∙/k1) takes this to

where the top arrow is the Jacobian matrix of f1,…,fm. This is also a pushout in 𝒟⁢(R)≥0, and therefore L⁢ΩR/k1 has as H0 and H1 the cokernel and kernel of the Jacobian matrix.

Definition A.2.1.3.

We will call ΩR/k1 the cotangent complex of R relative to k and denote it by LR/k.

A.2.2 Applications to obstruction theory of rings

Let S be a ring and M be an S-module. We can form a split square-zero extension on the S-module S⊕M where

  • •

    If m1,m2∈M, we have m1⁢m2=0;

  • •

    Elements of s multiply as usual;

  • •

    If s∈S and m∈M, then s⁢m is given by the usual action of s on M.

We would like to understand when there is a lift in the diagram

These lifts correspond to φ-linear derivations R→M, which are simply R-module maps

Proposition A.2.2.1.

There exists a functor

which sends a projective module P to the square-zero extension S⊕P. We have equivalences of mapping spaces

We would now like to apply this to study ordinary rings. Given a surjective R~→R whose kernel I satisfies I2=0, we will call this a (not necessarily split) square-zero extension. One example of a non-split square-zero extension is

which does not even split as a morphism of abelian groups. Note that the condition that I2=0 implies that the R~-action on I factors through R.

It turns out that LR/R~ has H0=0 and

There is a tautological morphism

of R-modules which induces an isomorphism on H1. This corresponds to a map

of animated R~-algebras.

Proposition A.2.2.2.

R~ is the pullback of the diagram

in the category of animated R~-algebras.

We see that (not necessarily split) square-zero extensions R~→R along I correspond to maps

Maps S→R~ lifting a given S→R correspond to diagrams of the form

which are simply lifts

If we unwrap everything, we see that a lift exists if and only if the map

is nullhomotopic. Furthermore, lifts are in correspondence to choices of nullhomotopies. These form a torsor over π1⁢(𝖬𝖺𝗉⁡(LS/k,I⁡[1])). We therefore see that

  • •

    The possible square-zero extensions R~ are classified by

  • •

    For a given R~ and S→R, there is an obstruction living in

  • •

    If lifts do exist, they are parameterized by

Exercise A.2.2.3.

Let S be a k-algebra. Show that the following are equivalent:

  1. 1.

    For every square-zero extension R~→R and every S→R, there is a lift (this is usually called being formally smooth).

  2. 2.

    We have H1⁢(LS/k)=0 and H0⁢(LS/k) is a projective S-module.

We are now ready to apply this machinery.

Theorem A.2.2.4.

If R1 is a perfect 𝔽p-algebra (this means that Frobenius is an automorphism), then there exist flat ℤ/pn-algebras Rn, unique up to isomorphism, with

In particular, we have

so we have a tower

whose lift

is the unique flat, p-complete ℤp-algebra satisfying

This theorem tells us that there is a way to escape from the positive characteristic to the mixed characteristic setting under the assumption that R1 is perfect.

We will characterize Rn as a non-split square-zero extension of Rn−1 by R1. This is because if it exists, tensoring

with Rn gives us

These extensions are classified by

They also need to be compatible with the element of

classifying ℤ/pn→ℤ/pn−1.

Lemma A.2.2.5.

Isomorphism classes of such Rn form a torsor over

Lemma A.2.2.6.

If R1 is a perfect 𝔽p-algebra, then LR1/𝔽p≃0.

Proof.

For any perfect 𝔽p-algebra A, we will consider the effect of the Frobenius φ:A→A on the contangent complex. This induces the zero morphism on ΩA/𝔽p1 (in particular for polynomial rings). Therefore, if we resolve R1 by polynomial rings, φ acts by 0 on LR1/𝔽p. However, because we assumed that R1 is perfect, it must also act by an isomorphism. The desired result follows immediately. □

Putting together the two lemmas, we see that there is a unique isomorphism class of possible Rn, which proves Theorem A.2.2.4.

Definition A.2.2.7.

The R in Theorem A.2.2.4 is called the Witt vectors of R and denoted by W⁡(R0).

Remark A.2.2.8.

Usually the Witt vectors are given by an explicit construction and exist for any 𝔽p-algebra, not just perfect ones, but we believe this explanation is much more satisfying.

Example A.2.2.9.

Consider the finite field 𝔽pn. The Witt vectors W⁡(𝔽pn) is a ℤp-algebra with

One way to construct this is to consider 𝔽pn=𝔽p⁢[x]/f⁡(x) and take the ring

for some lift f~ of f.