AWS 2019

Topological Hochschild homology in arithmetic geometry

Matthew Morrow

v.1 (1 Feb.) – updates to follow during February.

v.2 (8 Feb.) – Sections 2 and 3 rewritten (except §2.4 and §3.4); several projects expanded.
v.3 (9 Feb.) – Fixed mistakes in Project C.
v.4 (19 Feb.) – Fixed typos; added details to some projects; minor changes to §4.
v.5 (27 Feb.) – Added a fun exercise to 2.18, some remarks on transitivity properties of HH, also a
section on the mixed characteristic story.
Many thanks to the following for corrections, comments, etc.: Liam Keenen, Shizhang Li, Zhouhang
Mao, David Mehrle, Shubhodip Mondal, Martin Speirs, Yuri Sulyma.

1 Introduction
This course is an introduction to Hochschild and cyclic homology and their topological variants, aimed
mainly at readers whose background is in number theory and arithmetic geometry rather than algebraic
topology and homotopy theory. Useful references to have at hand while reading these notes are the
following:

- [29] The standard textbook presenting cyclic homology in a classical fashion; available free on
SpringerLink if your institution subscribes.

- [28] Notes based on lectures given by Nikolaus on topological cyclic homology.

- [21] Notes from the Arbeitsgemeinschaft on topological cyclic homology at Oberwolfach, April 2018.

- [32] Nikolaus and Scholze’s new approach to topological cyclic homology.

- [8] The article constructing “motivic” filtrations on all variants of (topological) cyclic homology
and relating these theories to crystalline cohomology and p-adic Hodge theory.

Given a commutative base ring k and a k-algebra A, the Hochschild homology groups HHn (A/k),
arising as the homology groups of the explicit Hochschild complex HH(A/k), were introduced already in
the 1950s and are closely related to algebraic differential forms: to be precise, if A is commutative then
there exist maps ΩnA/k → HHn (A/k), which are even isomorphisms when A is smooth over k. In the
1980s it was realised by Connes and Feigin–Tsygan that the usual de Rham differential d : ΩnA/k → Ωn+1 A/k
was simply a shadow of the fact that the complex HH(A/k) could be equipped with an action by
the circle S 1 , in a sense which we will take time to explain in the course. One may then study the
homology, cohomology, and Tate cohomology of this S 1 -action on HH(A/k), giving rise respectively to
cyclic homology, negative cyclic homology, and periodic cyclic homology; these fit together into a fibre
sequence
HC(A/k)[1] −→ HC − (A/k) −→ HP (A/k).
In §2 we will discuss these constructions from the classical point of view of explicit double complexes
and the theory of mixed complexes, finally presenting a proof of the following classical description in the
case of smooth algebras over a characteristic zero base ring:

1
 Matthew Morrow

Theorem 1.1 (Connes, Feigin–Tsygan). If A is a smooth k-algebra and k ⊇ Q, then there are natural
decompositions Y Y ≥n
HP (A/k) = Ω•A/k [2n], HC − (A/k) = ΩA/k [2n]
n∈Z n∈Z

In short, periodic and negative cyclic homology encode the de Rham cohomology of A and its Hodge
filtration. The main interest of Connes was therefore to view these cohomology theories as replacements
for de Rham cohomology, even if A is non-commutative. In this course we will rather be interested in the
case in which A is commutative, but not necessarily smooth nor of characteristic zero. In particular our
first goal will be to explain the following generalisation of the previous theorem to finite characteristic:

Theorem 1.2. If A is a smooth k-algebra, where k is a field of characteristic p > 0, then HP (A/k) and
HC − (A/k) admit Z-indexed, complete filtrations whose nth graded pieces are given respectively by

Ω•A/k [2n], Ω≥n

A/k [2n].

In other words, the classical theorem of Connes and Feigin–Tsygan extends beyond characteristic
zero, at the expense of replacing the product decomposition by a possibly non-split filtration.
The techniques which we will use to prove Theorem 1.2 in §3 are radically different to its characteristic
zero predecessor. We will see that the cohomology theories involved can be determined by their behaviour
on large Fp -algebras, namely perfect rings and certain well-behaved quotients thereof (called quasiregular
semiperfect rings). This leads us to the study of these cohomology theories on quasiregular semiperfect
rings, and to the formalism of the quasisyntomic site.
The problem with Theorem 1.2 is that de Rham cohomology is not the most interesting cohomology
theory in characteristic p; we would prefer to replace it by cristalline cohomology. However, it is man-
ifestly impossible to build cristalline cohomology from HH(A/k) via any Z-linear construction: since
HH(A/k) is a complex of Fp -vector spaces, so too will be anything constructed from it Z-linearly. The
trick is therefore to leave the Z-linear world, i.e., D(Z), by passing to the world of spectra Sp in the sense
of homotopy theory; constructions in Sp are no longer Z-linear, merely S-linear where S is the sphere
spectrum.
The idea of developing Hochschild and cyclic homology in this “brave new world” of ring spectra is due
to Goodwillie and Wahldhausen. The subject has seen enormous progress in recent years, particularly
thanks to pioneering work of Hesselholt and then the refoundation of the subject by Nikolaus–Scholze,
whose framework we will adopt. To any ring A one may associated its topological Hochschild homology

THH(A)“ := HH(A/S)” ∈ Sp,

which is once again equipped with an action by the circle S 1 ; passage to homology, cohomology, and
Tate cohomology (which are no longer Z-linear constructions!) then yields an analogous fibre sequence
as classically
THH(A)hS 1 [1] −→ TC − (A) −→ TP (A)
In §4 we attempt to gradually familiarise the reader with the concept of what it means for S 1 to act
on an algebraic object like a chain complex, then categorify the construction of Hochschild homology in
order to construct topological Hochschild homology.
Section 5 presents the main results concerning topological Hochschild homology of Fp -algebras, after
which there will be a brief discussion of the mixed characteristic situation.

Notation
Both chain and cochain complex appear in this theory, which is an easy source of confusion. We follow
the standard convention of denoting the former by lower indexes (and often by leftwards arrows)

· · · ← Cn ← Cn+1 ← · · ·

and the latter by upper indexes

· · · → C n → C n+1 → · · · .

2
 AWS 2019 – Topological Hochschild homology

For shifting complexes we follow the convention C[1]n := C n+1 , whence for chain complexes we have
C[1]n = Cn−1 , i.e., given an abelian group A, the shift A[n] is supported in cohomological degree −n
and in homological degree n; the differentials on C[1] are given by the negatives of the differentials on
C. Under this convention, the Eilenberg–Maclane functor takes the shift [1] to the suspension Σ. There
are bound to be some mistakes involving signs of shifts...

2 Classical Hochschild and cyclic homology

In this section we give a classical presentation of Hochschild and cyclic homology and their variants,
namely negative cyclic homology and periodic cyclic homology. We will define these in terms of the
explicit double complexes which were introduced by Connes and Feigin–Tsygan and establish the pe-
riodicity and norm sequences relating the different theories. We then abstract the constructions by
introducing the classical concepts of cyclic objects and mixed complexes, which will be helpful when
discussing two important examples. In the final subsection we explain how the definitions should be
modified when treating non-flat algebras, which is really just an excuse to recall the cotangent complex.
A standard reference for this material is chapters 1–5 of the book of Loday [29], also [39, §9] and [30];
we will give quite a streamlined presentation of the results we need, and consultations of these sources
will probably be helpful for readers seeing the material for the first time.

2.1 Hochschild homology and differential forms

Let k be a commutative ring and A a flat k-algebra (the flatness hypothesis will be discussed and removed
in §2.4). Then the associated Hochschild complex, denoted by HH(A/k) or more classically by HH k (A),
is the chain complex of k-modules

b b b
A ←− A ⊗k A ←− A ⊗k A ⊗k A ←− · · ·

(where A lies in degree 0) with boundary maps traditionally denoted by b given by

b : A⊗k n+1 → A⊗k n , a0 ⊗ · · · ⊗ an 7→ a0 a1 ⊗ · · · ⊗ an

− a0 ⊗ a1 a2 ⊗ · · · ⊗ an
+ ···
+ (−1)n a0 ⊗ · · · ⊗ an−1 an
+ (−1)n+1 an a0 ⊗ · · · ⊗ an−1

For example, b(a ⊗ b) = ab − ba and b(a ⊗ b ⊗ c) = ab ⊗ c − a ⊗ bc + ca ⊗ b. The homology groups of the

Hochschild complex are denoted by HHn (A/k) are called the Hochschild homology of the k-algebra A.

Example 2.1 (Low degrees). (i) n = 0. The image of boundary map A ⊗k A → A is precisely the
k-submodule generated by elements of the form ab − ba, i.e., the commutator [A, A] of A; so
HH0 (A/k) = A/[A, A]. In particular, if A is commutative then HH0 (A/k) = A.

(ii) n = 1. Assume A is commutative. Then we have just seen that the first boundary map n the
Hochschild complex is zero, so that

HH1 (A/k) = A ⊗k A/hab ⊗ c − a ⊗ bc + ca ⊗ b : a, b, c ∈ Ai,

where h i denotes here the abelian group (or equivalently A-module) generated by the indicated
elements. The relation by which we are quotienting is precisely the Leibniz rule, and so we see that
there is an isomorphism of A-modules
'
HH1 (A/k) → Ω1A/k , a ⊗ b 7→ a db.

3
 Matthew Morrow

Example 2.2 (The case A = k). If A = k then the boundary maps in the Hochschild complex are
alternately zero and the identity, whence
(
k n=0
HHn (k/k) =
0 n > 0.

The higher Hochschild homology groups vanish more generally whenever A is a commutative étale k-
algebra, as will follow from the Tor description of Remark 2.4.
Remark 2.3 (Simplicial perspective). Recall that a simplicial object in a category consists of objects
An , for n ≥ 0, and morphisms

“face maps” di : An → An−1 , i = 0, . . . , n,

“degeneracy maps” si : An → An+1 , i = 0, . . . , n,

satisfying the rules di dj = dj−1 di for i < j, and si sj = sj+1 si for i ≤ j, and

sj−1 di
 i<j
di sj = id i = j or j + 1

sj di−1 i > j + 1.


Diagramatically, we tend to forget about the degeneracy maps and draw such an object as
o o
A 0 oo A 1 oo A2 ooo ···

A simplicial object is equivalently a functor to the category from ∆op , where ∆ is the simplicial category.
d d d
A simplicial object A• in any additive category gives rise to a chain complex A0 ← AP 1 ← A2 ← · · ·
n
where the boundary maps are given by the alternating sum of the face maps, namely d := i=0 (−1)i di :
An → An−1 ; this is known as the un-normalised chain complex associated to A• (and, by a common
abuse of notation, usually still denoted by A• ).
In particular, to our flat k-algebra A we may associate a simplicial k-module defined by An := A⊗k n+1
with face and degeneracy maps
(
a0 ⊗ · · · ⊗ ai ai+1 ⊗ · · · ⊗ an i < n,
di (a0 ⊗ · · · ⊗ an ) :=
an a0 ⊗ · · · ⊗ an−1 i = n,
si (a0 ⊗ · · · ⊗ an ) := a0 ⊗ · · · ⊗ ai ⊗ 1 ⊗ ai+1 ⊗ · · · ⊗ an ,

whose associated un-normalised chain complex is precisely the Hochschild complex HH(A/k) as defined
above. We will interchangeably use HH(A/k) to denote both this simplicial k-module and the associated
complex.
Remark 2.4 (Bar complex and Tor description). Don’t confuse HHn (A/k) with the bar complex
b0 b0
B• (A/k) := A ⊗k A ←− A ⊗k A ⊗k A ←− · · ·
n−1
X
0 ⊗k n ⊗k n−1
b :A →A , a0 ⊗ · · · ⊗ an−1 7→ (−1)i a0 ⊗ · · · ⊗ ai−1 ai ⊗ · · · ⊗ an−1
i=0

(i.e., same as the Hochschild complex except that the product of the first and last tensor factors does
not appear), which can alternatively be described simplicially as

di (a0 ⊗ · · · ⊗ an−1 ) := a0 ⊗ · · · ⊗ ai ai+1 ⊗ · · · ⊗ an

si (a0 ⊗ · · · ⊗ an−1 ) := a0 ⊗ · · · ⊗ ai ⊗ 1 ⊗ ai+1 ⊗ · · · ⊗ an−1 .

In fact, the bar complex B• (A/k) is acyclic except possibly in degree 0, with multiplication b0 = µ :
∼
A ⊗k A → A defining a quasi-isomorphism B• (A/k) → A; this follows formally from the existence of

4
 AWS 2019 – Topological Hochschild homology

the “extra degeneracy” s : A⊗k n → A⊗k n+1 , a0 ⊗ · · · ⊗ an−1 7→ 1 ⊗ a0 ⊗ · · · ⊗ an−1 , which serves as a
contracting homotopy for B• (A/k) as a simplicial object [29, 1.1.12].
The bar complex nevertheless plays a role in the theory in several ways. In particular, observe that it
consists of left A⊗k Aop -modules, under the rule (f ⊗g)·(a0 ⊗· · ·⊗an−1 ) := f a0 ⊗a1 ⊗· · ·⊗an−2 ⊗an−1 g,
∼
which are moreover flat since A is flat over k. Thus B• (A/k) → A serves as a resolution of A by flat
A ⊗k Aop -modules (where we view A as a left A ⊗k Aop -module). But one checks directly that there is
an isomorphism of complexes (even of simplicial objects)
'
A ⊗A⊗k Aop B• (A/k) → HH(A/k), a0 ⊗ (f ⊗ a1 ⊗ · · · ⊗ an ⊗ g) 7→ ga0 f ⊗ a1 ⊗ · · · ⊗ an

(the fact that f and g change order is an inevitable consequence of the fact that the tensor product
A⊗A⊗k Aop involves viewing A as a right A ⊗k Aop -module) whence we see that HH(A/k) is a model for
op
the derived tensor product A ⊗LA⊗k Aop A; in particular, HH∗ (A/k) ∼
= Tor∗A⊗k A (A, A).
Remark 2.5 (The case A commutative). In this remark we assume that A is a commutative k-algebra,
which is our main case of interest. Then the face and degeneracy maps in HH(A/k) are ring ho-
momorphisms, which are even A-linear if we declare A to act on each A⊗k n by multiplication on
the left-most tensor factor. In this way HH(A/k) becomes a simplicial A-algebra and the boundary
maps b are A-linear.
L It follows formally that the homology groups HHn (A/k) are A-modules, and that
HH∗ (A/k) = n≥0 HHn (A/k) has the structure of a graded-commutative A-algebra.
Similarly, B• (A/k) is a simplicial A ⊗k A-algebra and the identification A ⊗A⊗k Aop B• (A/k) ∼
=
HH(A/k) is compatible with multiplicative structure, whence HH(A/k) models the simplicial commu-
tative ring A ⊗LA⊗k A A.

Remark 2.6 (Étale extensions). Continue to suppose that A is a commutative, flat k-algebra, and let
A0 be an étale A-algebra. Then the canonical map HH(A/k) ⊗A A0 → HHn (A0 /k) is an isomorphism
for any n ≥ 0. This does not follow tautologically from the definition of Hochschild homology (even if
A0 is a localisation of A) but may be proved in several ways; here we present one approach, for another
see [40].
∼
Since A → A0 is flat, it suffices to check that HH(A/k) ⊗LA A0 → HH(A0 /k), which reduces via the
∼
Hochschild–Kostant–Rosenberg filtration of Proposition 2.28 to showing that LiA/k ⊗A A0 → LiA0 /k for
all i ≥ 0 (where L is the cotangent complex, which will be introduced in §2.4). But this follows easily
from the fundamental properties of the cotangent complex which will be presented in Proposition 2.27.
The cautious reader should be concerned about circular reasoning, given that the proof of Proposi-
tion 2.28 will Theorem 2.8, which in turn uses this remark, but in fact Proposition 2.28 only requires
Theorem 2.8 in the case of polynomial algebras, whereas this remark will be used to extend from poly-
nomial algebras to general smooth algebras.

Remark 2.7 (Transitivity). Suppose that we are given a flat morphism of flat k-algebras A → B, where
A is assumed to be commutative. Then we claim that the canonical map of simplicial A-algebras is an
isomorphism
'
HH(B/k) ⊗HH(A/k) A −→ HH(B/A).
Here we use that A is commutative so that the adjunction HH(A/k) → HH0 (A/k) is a map to A. Since
tensor products of simplicial rings are by definition computed degree-wise, the claimed isomorphism is
simply the assertion that
'
(B ⊗k · · · ⊗k B) ⊗A⊗k ···⊗k A,µ A → B ⊗A · · · ⊗A B

for any tensor power, where we base change along the multiplication map µ. But this is obvious: the
universal property of the right side yields the inverse map.

We finish this foundational subsection by discussing the case of smooth algebras. Assuming for
'
a moment that A is any commutative, flat k-algebra, then the identification Ω1A/k → HH1 (A/k) of
Example 2.1(i) induces a map of graded algebras Ω∗A/k → HH∗ (A/k), simply because HH∗ (A/k) is a
graded-commutative A-algebra by Remark 2.5. By explicitly examining the definition of the product

5
 Matthew Morrow

structure on the homotopy groups of a simplicial commutative rings (known as the shuffle product), one
can check that this map of graded algebras is given in each degree by the maps
X
εn : ΩnA/k −→ HHn (A/k), a db1 ∧ · · · ∧ dbn 7→ (−1)sign(σ) a ⊗ bσ(1) ⊗ · · · ⊗ bσ(n)
σ∈Symn

for each n ≥ 0 (however, we will try to avoid using this explicit formula). These are often known as the
anti-symmetrisation maps.
We may now present the fundamental description of Hochschild homology of smooth algebras:
Theorem 2.8 (Hochschild–Kostant–Rosenberg Theorem 1962). If A is a smooth k-algebra, then the
antisymmetrisation maps ΩnA/k → HHn (A/k) are isomorphisms of A-modules for all n ≥ 0.

Proof. The goal is to establish that the morphism of graded-commutative rings Ω∗A/k → HH∗ (A/k),
which is already known to be an isomorphism in degreesV≤ 1, is in fact an isomorphism in all degrees. It
∗
is therefore equivalent to prove that the canonical map A HH1 (A/k) → HH∗ (A/k) is an isomorphism;
by now appealing to the Tor description of Hochschild V∗ homology from Remarks 2.4 and 2.5, we see that
it is equivalent to check that the canonical map A TorA⊗ 1
kA
(A, A) → TorA⊗
∗
kA
(A, A) is an isomorphism,
A⊗k A
i.e., that the Tor algebra Tor∗ (A, A) is the free graded-commutative A-algebra generated in degree 1
µ
by TorA⊗1
kA
(A, A) = I/I 2
, where I = Ker(A ⊗k A − → A). This is well-known to be true if I is generated
by a regular sequence (just compute the graded Tor ring using a Koszul complex).
We have completed the proof whenever Ker µ is generated by a regular sequence, e.g., if A is a
polynomial k-algebra.
But A is smooth over k, so by arguing Zariski locally on Spec A (here we use that HHn (A/k)⊗A A[ f1 ] =
HHn (A[ f1 ]/k) for all f ∈ A, thanks to Remark 2.6) we may reduce to the case that A is étale over a
polynomial algebra; the result for A then follows by the polynomial case and another use of Remark 2.6.

Remark 2.9 (Detecting smoothness with HH). The HKR Theorem admits various converses; for ex-
ample, if A is a non-smooth, but finitely generated, algebra over a field k, then HHn (A/k) is non-zero
for infinitely many n ≥ 0 [3].
V∗
Question 2.10. Which E∞ -rings A have the property that THH∗ (A) = A THH1 (A)? This might be
known, I have not yet asked any experts.

2.2 Cyclic homology and its variants; cyclic objects and mixed complexes
The goal of this subsection is first to define classical cyclic homology and its variants, namely negative
and periodic cyclic homology; then we will axiomatise the formalism by summarising Connes’ theory of
cyclic objects and Kassel–Burghelea’s mixed complexes.
For the moment we continue to let A be a flat algebra over a commutative base ring k. The starting
point for cyclic homology is the observation that each tensor power A⊗k n+1 in the Hochschild complex
carries a k-linear action by the cyclic group Z/n + 1 via permutation, i.e., the generator tn = 1 ∈ Z/n + 1
acts1 as
tn · a0 ⊗ · · · ⊗ an := an ⊗ a0 ⊗ · · · ⊗ an−1 .
Following classical notation, we also introduce the k-linear maps
n
X
“norm” N := ((−1)n tn )i : A⊗k n+1 → A⊗k n+1 ,
i=0

“extra degeneracy” s : A⊗k n → A⊗k n+1 , a0 ⊗ · · · ⊗ an−1 7→ 1 ⊗ a0 ⊗ · · · ⊗ an−1 ,

and the composition

“Connes’ (boundary) operator” B := (1 − (−1)n tn )sN : A⊗k n −→ A⊗k n+1 .

1 Compared to the first version of these notes, the sign convention has changed.

6
 AWS 2019 – Topological Hochschild homology

When n is clear from the context one tends to write simply t instead of tn to simplify notation; with
this in
Pmind we will adopt the (non-standard) convention of writing ±t := (−1)n tn so that, for example,
n
N = i=0 (±t)i and B = (1 − ±t)sN .

Lemma 2.11. The above operators satisfy the following identities:

(i) (1 − ±t)b0 = n(1 − ±t) and b0 N = N b.

(ii) sb0 + b0 s = 1, B 2 = 0, and Bb = −bB.

Proof. We leave these to the reader as an exercise; see [29, Lem. 2.1.1].

Remark 2.12 (Connes’ operator vs. the de Rham differential). Connes’ boundary operator clearly
induces k-linear maps B : HHn (A/k) → HHn+1 (A/k) for n ≥ 0. When A is commutative, as we
assume in the rest of the remark, then this turns the graded algebra HH∗ (A/k) into a commutative
differential graded k-algebra [To do: insert proof of this.] Since HH0 (A/k) = A, the universal property
of Ω∗A/k (namely, it is the initial cdg k-algebra which is A in degree 0) induces a map of differential
graded k-algebras Ω∗A/k → HH∗ (A/k). However this is nothing other than the anti-symmetrisation map
constructed before Theorem 2.8, since the map

B : A = HH0 (A/k) −→ Ω1A/k = HH1 (A/k)

is easily seen to be precisely the de Rham differential (the reader should check).
In particular, it follows that the following diagrams commute

HHn (A/k)
B / HHn+1 (A/k)
O O
εn εn+1

ΩnA/k / Ωn+1
d A/k

i.e., B is a refinement of the de Rham differential. Part of Connes’ motivation when developing cyclic
homology was to consider B as a generalisation of the de Rham differential even when A was non-
commutative.

The classical double complex approach to defining cyclic homology and its variants are based on the
following (anticommuting) bicomplex of k-modules, which is horizontally 2-periodic

A⊗3 o A⊗3 o A⊗3 o A⊗3 o

N 1−±t N 1−±t
A⊗3

−b0 b −b0 b −b0

    
A⊗2 o A⊗2 o A⊗2 o A⊗2 o
N 1−±t N 1−±t
A⊗2

−b0 b −b0 b −b0

    
Ao Ao Ao Ao
N 1−±t N 1−±t
A

(the term in degree (0, 0) is underlined, a convention we will follow throughout this subsection), i.e.,
the even columns are the Hochschild complex HH(A/k) and the odd columns are the augmented bar
complex B• (A/k) → A from Remark 2.4. Restricting this bicomplex to the right half plane x ≥ 0 gives
a first quadrant bicomplex whose totalisation we define to be the cyclic homology HC(A/k). Similarly,
restricting to the left half plane x ≤ 0 gives a second quadrant bicomplex whose product totalisation is
defined to be negative cyclic homology HC − (A/k) (also denoted by HN (A/k) in the literature). Finally,
the product totalisation of the entire 2-periodic bicomplex is periodic cyclic homology HP (A/k).

7
 Matthew Morrow

However, these double complexes can be simplified by recalling from Remark 2.4 that the odd columns
in the above bicomplex are all acyclic, with contracting homotopy given by the extra degeneracy s;
this allows us formally to remove all odd columns from the bicomplex by mapping between adjacent
even columns via (1 − ±t)sN = B. More precisely, it follows that HP (A/k) is quasi-isomorphic to its
subcomplex given by totalising the BP -bicomplex

A⊗4 o A⊗3 o A⊗2 o

B B
A
B

b b b
  
A⊗3 o A⊗2 o
B B
A

b b
 
A⊗2 o
B
A

b

A

Similarly, HC − (A/k) and HC(A/k) are quasi-isomorphic to their subcomplexes given respectively by
totalising the restriction of the BP-bicomplex to x ≤ 0 or x ≥ 0, which we call the BC − -bicomplex and
BC-bicomplex

A⊗3 o A⊗2 o A⊗3 o A⊗2 o

B B B B
A A

b b b b
   
A⊗2 o
B
A⊗2 o
B
A A

b b
 
A A

For some further details on this process of discarding all all columns of the original bicomplexes, we refer
the reader to [29, §2.1.7] or [30, Prop. 1.5].
Over the next sequence of remarks we explain various formal properties of cyclic homology and its
variants, all of which follow by elementary examination of the double complexes defining them.
Remark 2.13 (Periodicity sequence for HC). Here we explain the periodicity sequence through which
cyclic homology is built from successive shifts of Hochschild homology.
Viewing HC(A/k) as the totalisation of the latter BC-bicomplex, the inclusion of the zero-th column
defines an inclusion I : HH(A/k) ,→ HC(A/k); the cokernel of I is the totalisation of what remains after
removing the first column from the BC-bicomplex, which is simply another copy of the BC-bicomplex,
but shifted by bidegree (1, 1). This gives a short exact sequence of complexes
I S
0 −→ HH(A/k) −→ HC(A/k) −→ HC(A/k)[2] −→ 0

8
 AWS 2019 – Topological Hochschild homology

with associated boundary map B : HC(A/k)[1] → HH(A/k). The projection map S : HC(A/k) →
HC(A/k)[2] from removing the zero-th column is known as Connes’ periodicity operator.
In other words, Fili := Ker(S i : HC(A/k) → HC(A/k)[2i]) defines an exhaustive, increasing, N-
indexed filtration on HC(A/k) whose ith graded piece is HH(A/k)[2i] for each i ≥ 0. For explanation of
our adjectives concerning filtrations, see Remark 2.29
Remark 2.14 (Periodicity sequences for HC − ). Similarly to Remark 2.14, projecting the BC − -bicomplex
to its zero-th column defines a short exact sequence of complexes
S h
0 −→ HC − (A/k)[−2] −→ HC − (A/k) −→ HH(A/k) −→ 0,
and the images of all powers of S define a complete, descending, N-indexed filtration on HC − (A/k) with
graded pieces HH(A/k)[-2i], i ≥ 0.
Remark 2.15 (Norm sequence). Continuing the theme of Remarks 2.13 and 2.14, the inclusion I of the
BC − -bicomplex into the BP -bicomplex has complement given by a shift of the BC-bicomplex, and there
are projections h from the BC − -bicomplex to the BC-bicomplex, and from the BP -bicomplex to the
BC-bicomplex. Using the sequences of the previous remarks, the summary is a commutative diagram of
short exact sequences as follows:
0 0

 
HC − (A/k)[−2] HC − (A/k)[−2]

S IS
 
0 / HC − (A/k) I / HP (A/k) Sh / HC(A/k)[2] /0

h h
 
0 / HH(A/k) / HC(A/k) / HC(A/k)[2] /0
I S

 
0 0
The connecting map HC(A/k)[1] → HC − (A/k) associated to the short exact sequence in the middle
is represented by the norm map N (this is easier to make precise using the original 2-periodic (b, b0 )-
bicomplexes).
Remark 2.16 (Periodicity of HP ). The restriction of the BP -bicomplex to any right half plane x ≥ −n
is of course a copy of the BC-bicomplex shifted by bidegree (−n, −n). Taking the inverse limit as n → ∞
shows that
HP (A/k) ' Rlimn→∞ HC(A/k)[2n],
where the transition maps are shifts of the periodicity operator S : HC(A/k) → HC(A/k)[2]. In
other words, periodic cyclic homology is constructed by forcing the periodicty operator to be a quasi-
∼
isomorphism, and in particular it is 2-periodic: S : HP (A/k) → HP (A/k)[2].
Next we axiomatise the essential properties of the Hochschild complex which were used in the above
constructions. However, we suggest that the reader first jumps ahead to the examples of §2.3 to see why
this axiomatisation is useful.
Definition 2.17 (Connes [14]). A cyclic object in a category C is a simplicial object X• in C such that
each Xn is equipped with an automorphism tn of order n + 1, and these are required to satisfy:
( (
di−1 tn−1 1 ≤ i ≤ n di−1 sn−1 1 ≤ i ≤ n
tn di = tn si =
dn i=0 sn t2n+1 i = 0.
Equivalently, a cyclic object is a functor Λop → C, where Λ is Connes’ cyclic category, characterised as
follows:

9
 Matthew Morrow

- Λ contains ∆ as a full subcategory having the same objects;

- AutΛ ([n]) = Z/n + 1 for each n ≥ 0 (whereas Aut∆ ([n]) = 1);

- The above relations between tn ∈ Z/n + 1 and the face and degeneracy maps hold.

Checking that there really exists such a category Λ characterised by these properties requires some mild
combinatorics which we omit; see [29, §6.1].

Example 2.18. The reader will easily observe that the simplicial k-module HH(A/k) is indeed a cyclic
object in k-modules, with the automorphisms tn as defined at the start of the subsection.
Assume now that A is commutative, whence HH(A/k) is even a cyclic object in k-algebras. In fact,
as an exercise prove the following universal characterisation of the HH(A/k): it is initial among the
collection of pairs (X• , A → X• ), where X• is a cyclic object in k-algebras and A → X• is a morphism
of simplicial k-algebras (i.e., the underlying simplicial object of X• is given the structure of a simplicial
A-module).

When viewed as a complex, the essential data on HH(A/k) was instead as follows:

Definition 2.19 (Burghelea [13], Kassel [25]). A mixed complex, or algebraic S 1 -complex, (A• , b, B) over
k is the data of a chain complex of k-modules
b b b
A• = · · · ← An ← An+1 ← · · ·

supported in homological degrees ≥ 0, together with maps of k-modules B : An → An+1 satisfying both
Bb = −bB (i.e., B is a morphism of complexes A → A[1]) and B 2 = 0.
Given a mixed complex (A• , b, B), its periodic cyclic homology HP (A• , b, B), negative cyclic homol-
ogy HC − (A• , b, B), and cyclic homology HC(A• , b, B), are the product totalisations of the following
bicomplex:

A3 o A2 o A1 o
B B B
A0

b b b
  
A0 o A1 o
B B
A0

b b
 
A1 o
B
A0

b

A0

resp. its restriction to the left half plane x ≤ 0, resp. its restriction to the right half plane x ≥ 0.

Example 2.20. (i) The Hochschild complex HH(A/k) is a mixed complex with respect to Connes’
operator B, by Lemma 2.11; in this case Definition 2.19 of course reduces to the earlier definitions
of cyclic, negative cyclic, and periodic cyclic homology.
B B B
→ Cn −
(ii) Given any cochain complex · · · − → C n+1 − → · · · supported in cohomological degrees ≥ 0,
we can form a mixed complex (C• , 0, B) by setting Cn := C n and declaring the b-differentials to be
zero. Then the double complex of Definition 2.19 is a product of infinitely many copies of shifts of

10
 AWS 2019 – Topological Hochschild homology

the cochain complex (C • , B); truncating to the left or right half plane involves naively truncating
this cochain complex. Taking care of the indexing, one sees that
Y Y Y
HC − (C• , 0, B) = C ≥i [2i], HP (C• , 0, B) = C • [2i], HC(C• , 0, B) = C ≤i [2i].
i∈Z i∈Z i≥0

Of particular interest will be the mixed complex (Ω•A/k , 0, d), where Ω•A/k is the usual de Rham
complex of a commutative k-algebra A.
(Warning: there is an unfortunate conflict of notation here: usually if we view a cochain complex
C • as a chain complex then it implicitly means that Cn := C −n with the same differential as in
C • ; but mixed complexes will appear sufficiently rarely that this conflict should not cause any
problems.)
(iii) At the other extreme to example (ii), one can also consider mixed complexes (A• , b, 0) in which the
B-differential is zero. Then the double complex of Definition 2.19 is a product of infinitely many
copies of shifts of the chain complex A• , and one sees that
Y Y Y
HC − (A• , b, 0) = A• [2i], HP (A• , b, 0) = A• [2i], HC(A• , b, B) = A• [2i].
i≤0 i∈Z i≥0

Remark 2.21. Remarks 2.13–2.16 extend verbatim to arbitrary mixed complexes, from which one
obtains the following sort of consequence (the reader should check if uncertain): given a morphism of
mixed complexes (A• , b, B) → (A0• , b0 , B 0 ) (the notion should be clear) such that A• → A0• is a quasi-
isomorphism of chain complexes, then HC(A• , b, B) → HC(A0• , b0 , B 0 ) is also a quasi-isomorphism, and
similarly for HC − and HP .
Now let A• be a cyclic object in k-modules. Then we claim that its un-normalised chain complex
n
b b
X
A• := A0 ←− A1 ←− · · · , b= di
i=0

admits the structure of a mixed complex. Indeed, defining the “extra degeneracy” s : An−1 → An to
be s := tn sn , we may then define the norm N and boundary operator B exactly as at the start of the
subsection and repeat the arguments of Lemma 2.11 verbatim to see that (A• , b, B) is a mixed complex.
There is moreover an analogue of the bar complex in this generality, namely the chain complex
n−1
b0 b0
X
A0 ←− A1 ←− · · · b0 = di ,
i=0

which is again acyclic since s serves as a contracting homotopy. Just as for Hochschild homology we may
therefore form the bicomplex

A2 o A2 o A2 o A⊗3 o
N 1−±t N 1−±t
A2

−b0 b −b0 b −b0

    
A1 o A1 o A1 o A1 o
N 1−±t N 1−±t
A2

−b0 b −b0 b −b0

    
A0 o A0 o A0 o A0 o
N 1−±t N 1−±t
A0

and argue that HP (A• , b, B), HC − (A• , b, B), and HC(A• , b, B) could instead have been defined (up to
quasi-isomorphism) as the totalisation of this bicomplex, resp. its truncation to x ≤ 0, resp. to x ≥ 0.
We will return to cyclic objects and mixed complexes, and their relation, from a more highbrow point
of view in §4.

11
 Matthew Morrow

2.3 Two examples: smooth algebras and group algebras

2.3.1 Smooth algebras in characteristic zero
The main classical example in cyclic homology which will be important to us (albeit only as motivation)
is the case of smooth algebras in characteristic zero, when it can be described in terms of de Rham
cohomology. For any commutative k-algebra A, we denote as usual by Ω•A/k its de Rham complex and
by
d d
Ω≤i
A/k := A −→ · · · −→ ΩiA/k −→ 0 −→ · · ·

its naive truncations; similarly define Ω≥i ≥i • ≤i−1

A/k , so that 0 → ΩA/k → ΩA/k → ΩA/k → 0 is a short exact
sequence of complexes.
Theorem 2.22 (Connes, Loday–Quillen, Feigin–Tsygan 1980s). Let k be a commutative base ring con-
taining Q and A a smooth k-algebra. Then there are natural equivalences
Y ≥i Y Y ≤i
HC − (A/k) ' ΩA/k [2i], HP (A/k) ' Ω•A/k [2i], HC(A/k) ' ΩA/k [2i].
i∈Z i∈Z i≥0

Proof. We define maps

π : A⊗k n+1 −→ ΩnA/k , a ⊗ b1 ⊗ · · · ⊗ bn 7→ a db1 ∧ · · · ∧ dbn

and leave it to the reader to check directly that the following diagrams commute:

A⊗ k n o / A⊗k n+1
b B
A⊗k n+1 A⊗ k n
π π π π
   
Ωn−1 o ΩnA/k n−1
ΩA/k / Ωn
A/k 0 (n+1)d A/k

In other words, the maps π define a morphism of mixed complexes from HH(A/k) to (Ω•A/k , 0, (•+1)d) :=
the mixed complex with zero b-differential associated to the cochain complex
2d 3d 4d
A −→ Ω1A/k −→ Ω2A/k −→ · · · .

ε Hn (π)
We also leave it to the reader to check that the composition ΩnA/k −→ n
HHn (A/k) −−−−→ ΩnA/k is simply
multiplication by n!, where εn is the anti-symmetrisation map which appeared in Theorem 2.8.
Now we use the hypotheses on A. Since it is smooth over k, Theorem 2.8 implies that εn is an
1
isomorphism; since k ⊇ Q, it follows that Hn (π) is also an isomorphism. Replacing π by n! πn constructs a
•
morphism of mixed complexes from HH(A/k) to (ΩA/k , 0, d), which is a quasi-isomorphism of underlying
complexes. Therefore it induces quasi-isomorphisms on HC, HC − , and HP by Remark 2.21, whence
Example 2.20(ii) completes the proof.

2.3.2 Group algebras

Let k be a commutative base ring and let G be an abelian group, written multiplicatively. Here we
describe some relations between the homology of G and the cyclic homology of the group algebra k[G].
The cyclic bar construction
o o
B cycl G = G oo G2 oo G3 ooo ···

is the cyclic set having face and degeneracy maps, and cyclic operator, as in the Hochschild complex,
i.e., the maps (g0 , . . . , gn ) 7→ g0 ⊗ · · · ⊗ gn define an inclusion of cyclic objects B cycl G ⊆ HH(k[G]/k).
This is clearly an equality after linearising B cycl G to turn it into a cyclic k-module:

k[B cycl G] = HH(k[G]/k).

12
 AWS 2019 – Topological Hochschild homology

To analyse B cycl G we decompose it as follows. Firstly, the simplicial set

o o
BG = 1 oo G oo G2 ooo ···
denotes the usual bar construction; it may be equipped with the structure of a cyclic set by declaring
the cyclic operator tn ∈ Z/(n + 1)Z to act on Bn G = Gn as
(g1 , . . . , gn ) 7→ ((g1 · · · gn )−1 , g1 , . . . , gn−1 ).
More generally, for any z ∈ G there is a “twisted” bar construction B(G, z) which is again a cyclic set:
as a simplicial set it is the same as BG, but the cyclic operator tn is now defined by (g1 , . . . , gn ) 7→
(z(g1 · · · gn )−1 , g1 , . . . , gn−1 ); obviously B(G, 1) = BG.
For any z ∈ G there is an inclusion of cyclic sets
B(G, z) ,→ B cylc G, (g1 , . . . , gn ) 7→ (z(g1 · · · gn )−1 , g1 , . . . , gn )
with image {(g0 , . . . , gn ) ∈ BncylcF
G : g0 · · · gn = z)}; these assemble to define a decomposition of the
cyclic bar construction B cylc G = z∈G B(G, z). Forgetting the cyclic structure, this decomposition may
be written as one of simplicial sets B cylc G = z∈G zBG, where zBG is the left translation of BG.
F
By then k-linearising we see that we have defined a direct sum decomposition of cyclic k-modules
M
HH(k[G]/k) = k[B(G, z)].
z∈G

Forgetting the cyclic

L structure, this may be written as a direct sum decomposition of simplicial modules
HH(k[G]/k) = z∈G zk[BG] = k[G] ⊗k k[BG], which we summarise in the next lemma:
Lemma 2.23. The natural map of simplicial k[G]-modules
k[G] ⊗k k[BG] −→ HH(k[G]/k), f ⊗ (g1 , . . . , gn ) 7→ f (g1 · · · gn )−1 ⊗ g1 ⊗ · · · ⊗ gn ,
where g1 , . . . , gn ∈ G and f ∈ k[G], is an isomorphism. Passing to homology yields a natural isomorphism
of graded k[G]-algebras
'
k[G] ⊗k H∗ (G, k) → HH∗ (k[G]/k).
Proof. The only statement still requiring justification is the homology calculation, which follows from
the fact that the simplicial k-module k[BG] models the group homology RΓ(G, k), where G acts trivially
on k.
The cyclic structure on the twisted bar constructions is somewhat subtle, so we focus now only on
the cyclic k-module k[BG], whose associated cyclic homology etc. we denote by HC(G/k) etc. It is
known that the S 1 -action on the geometric realisation of BG is trivial whence the cyclic homology and
its variants behave as in Example 2.20 (since we do not need the following proposition, we refer to [29,
§7.3.9 & Prop. 7.4.8] for further explanation):
Proposition 2.24. There are natural equivalences
Y Y Y
HC − (G/k) ' RΓ(G, k)[2i], HP (G/k) ' RΓ(G, k)[2i], HC(A/k) ' RΓ(G, k)[2i].
i≤0 i∈Z i≥0

2.4 The cotangent complex and Hochschild homology for non-flat algebras
We briefly summarise the theory of the cotangent complex; the classical standard references are Quillen
[35, 34], Illusie [23, 24], and M. André [1]. Let k be a commutative base ring and A a commutative
∼
k-algebra. Then there exists a simplicial resolution P• → A of A by polynomial k-algebras (possibly
in infinitely many variables); this resolution is unique up to homotopy, whence the same is true of the
simplicial A-module
LA/k := Ω1P• /k ⊗P• A
which is known as the cotangent complex of k → A. By a standard abuse, LA/k is often identified with
its associated complex via Dold–Kan in D(A) (whence the terminology). Similarly, the wedge powers of
Vi
the cotangent complex are LiA/k := A LA/k = ΩiP• /k ⊗P• A.

13
 Matthew Morrow

Remark 2.25 (Left Kan extension). Particularly in §3.1 it will be helpful to adopt a more modern
perspective on the above construction, namely that of left Kan extensions. Let D be any ∞-category
which admits sifted colimits (e.g., D(k) or Sp), and write k -algsΣ for the category of polynomial k-
algebras in finitely many variables. One says that a functor F : k -algs → D is left Kan extended (from
k -algsΣ ) if the canonical map F(A) → limP →A F(P ) is an equivalence for each k-algebra A, where the
−→
colimit in D is taken over all P ∈ k -algsΣ mapping to A.
Conversely, given a functor G : k -algsΣ → D, we may left Kan extend it to LG : k -algs → D
by using the above formula: G(A) := limP →A F(P ). We thus obtain an equivalence between functors
−→
k -algsΣ → D and left Kan extended functors k -algs → D.
Returning to the more concrete point of view, the left Kan extension is given as follows: first extend
G to all polynomial k-algebras (possibly in infinitely many variables) by taking filtered colimits, then
extend it to all k-algebras by the rule

LG(A) := |G(P• )| = geometric realisation of the simplicial D-object G(P• )

where P• → A is any simplicial resolution of A by polynomial k-algebras. For example, Li−/k is precisely
the left Kan extension of Ωi−/k : k -algsΣ → D(k) and could more correctly be denoted by LΩi−/k .

Example 2.26 (Low degrees). (i) For any k-algebra A and n ≥ 0, the augmentation LnA/k → ΩnA/k
'
induces π0 (LnA/k ) → ΩnA/k . Indeed, it is clearly surjective since P0  A induces Ω1P0 /k  Ω1A/k ;
but the differential d : P• → Ω1P• /k induces a derivative d : A → π0 (LA/k ) which in turn induces a
map Ω1A/k → π0 (LA/k ) splitting the surjection.

(ii) If A = k/I is a quotient of k, then π0 (LA/k ) = 0 and π1 (LA/k ) = I/I 2 . The vanishing of π0
follows from (i), so it remains to justify the description of π1 . Unfortunately there doesn’t follow
straight from the definitions: one either relates the cotangent complex to algebra extensions [23,
Corol. II.1.2.8.1], or else reads it off Quillen’s “fundamental spectral sequence” [34, Thm. 6.3].

Next we summarise the fundamental abstract properties of the cotangent complex:

Proposition 2.27. Let k be a commutative ring and A a commutative k-algebra.

(i) (Base change) For any base change k → k 0 (where k 0 is commutative) such that Torki (A, k 0 ) = 0
∼
for i > 0, then LA/k ⊗Lk k 0 → LA0 /k where A0 := A ⊗k k 0 .

(ii) (Künneth) For any other commutative k-algebra B such that Torki (A, B) = 0 for i > 0, then
LA⊗k B/k ' (LA/k ⊗Lk B) ⊕ (LB/k ⊗Lk A).

(iii) (Transitivity sequence) Given a commutative A-algebra B, the resulting sequence

LA/k ⊗LA B −→ LB/k −→ LB/A

is a fibre sequence.

(iv) (Localisation) If S ⊆ k is a multiplicative system, then LS −1 k/k ' 0

(v) (Étale) If A is étale over k, then LA/k ' 0.

∼
(vi) (Smooth) If A is smooth over k then LA/k is supported in degree zero, whence LA/k → Ω1A/k [0].
∼
Proof. For (i), just pick a simplicial resolution P• → A by polynomial k-algebras and observe that
∼
P• ⊗k k 0 → A0 is a resolution of A0 by polynomial k 0 -algebras (indeed, the failure of it to be a resolution
is precisely the higher Tors which we have assumed vanish); then use that Ω1Pn /k ⊗k k 0 = Ω1Pn ⊗k k0 /k0 for
each n ≥ 0.
∼ ∼
(ii): Let P• → A be as in the previous paragraph, and Q• → B similarly for B. Then P• ⊗k Q• is a
simplicial resolution of A⊗k B by polynomial k-algebras, whence LA⊗k B/k = Ω1P• ⊗k Q• ⊗P• ⊗k Q• (A⊗k B).
Rewriting this using Ω1Pn ⊗k Qn /k = Ω1Pn /k ⊕ Ω1Qn /k for each n easily gives the desired result.

14
 AWS 2019 – Topological Hochschild homology

To prove the transitivity sequence one first uses some standard properties of simplicial commutative
rings (or, more precisely, of their model structure; we refer to the proof of [34, Thm. 5.1] for details) to
build a commutative diagram
PO • B / Q• / Q• ⊗P• A
BB JJ :
BB JJJtttt
BB tJ
t J
B tt JJJ$ 
/ tt /B
k A
where P• is a simplicial resolution of A by free k-algebras, Q• is a simplicial resolution of B by a cofibrant
P• -algebra (this means that Q• is a retract of a free P• -algebra), and Q• ⊗P• A is a simplicial resolution
of B by free A-algebras (or perhaps only retracts of free A-algebras). Since the maps k → Pn → Qn
are all smooth, there is an exact Jacobi–Zariski sequence 0 → Ω1Pn /k ⊗Pn Qn → Ω1Qn /k → Ω1Qn /Pn → 0
for each n. Tensoring by − ⊗Qn B and rearranging some of the tensor products gives a fibre sequence
Ω1P• /k ⊗P• A → Ω1Q• /k ⊗Q• B → Ω1Q• ⊗P A/A ⊗Q• ⊗P• A B, as desired.
•
To prove (iv) and (v) we follow a clever argument apparently due to M. André. Given a multiplicative
system S ⊆ k, we have S −1 k⊗k S −1 k = S −1 k and so LS −1 k/k ⊗k S −1 k = LS −1 k/k ⊗S −1 k (S −1 k⊗k S −1 k) =
LS −1 k/k ; but part (i) shows that LS −1 k/k ⊗k S −1 k = LS −1 k⊗k S −1 k/S −1 k = LS −1 k/S −1 k = 0.
For (v), we recall that étale means that k → A is flat and that µ : A ⊗k A → A induces an open
immersion on Spec. Therefore, for any prime ideal p ⊆ A we have Ap = (A ⊗k A)q with q := µ−1 (p),
and so

LA/k ⊗A Ap = LA/k ⊗A (A ⊗k A) ⊗A⊗k A (A ⊗k A)q

= LA⊗k A/A ⊗A⊗k A (A ⊗k A)q (by part (i))
= L(A⊗k A)q /A (by transitivity and L(A⊗k A)q /A⊗k A = 0 by (iv))
= LAp /A
=0 (by (iv))

Since this holds for all prime ideals p ⊆ A, we have LA/k = 0.

To prove (vi) we then argue as follows: it is enough to prove the result Zariski locally on Spec A (using
(i)), so we may suppose that A is étale over k[t1 , . . . , td ] for some d ≥ 0. But then the result follows at once
from the following: Lk[t]/k = Ω1k[t]/k (since k[t] serves as a polynomial resolution of itself); LA/k[t] = 0
'
by (v); the transitivity sequence for k → k[t] → A; the well-known fact that Ωk[t]/k ⊗k[t] A → Ω1A/k since
k[t] → A is étale.
Our interest in the the cotangent complex stems from the following relation to Hochschild homology
Proposition 2.28 (Hochschild–Kostant–Rosenberg filtration). Let A be a commutative k-algebra.
(i) Then the Hochschild complex HH(A/k) (viewed as an object of D(A)) admits a natural, complete,
descending N-indexed filtration whose ith graded piece is equivalent to LiA/k [i], for i ≥ 0.

(ii) Similarly,
LHC(A/k) admits a natural, complete, descending N-indexed filtration whose ith -graded
i
piece is n≥0 LA/k [i + 2n], for i ≥ 0.
Remark 2.29 (Complete filtrations). Although we have already used terminology surrounding filtrations
in §2.2, we did not give definitions; we do so now. Let us begin with the case of an honest complex C
over a ring; by a descending, N-indexed filtration we mean a descending chain of subcomplexes C =
Fil0 C ⊇ Fil1 C ⊇ · · · . To say that it is complete means that the canonical map Cn → limi Cn / Fili Cn is
←−
an isomorphism for each degree n.
However, even when C is an explicit complex like HH(A/k), we always prefer to view C merely as an
object of the derived category (where it does not make sense to discuss subcomplexes or the individual
terms Cn ); in that case a descending, N-indexed filtration means simply that we are given complexes and
maps between then C = Fil0 C ← Fil1 C ← Fil2 C ← · · · , and complete means that the canonical map
C → Rlimi C/ Fili C is an equivalence (informally, we are asking that C can be completely recovered from
all the C/ Fili C). Here we write C/ Fili C for the cofiber of the map Fili C → C, so the completeness
condition is equivalent to asking that Rlimi Fili C ' 0.

15
 Matthew Morrow

Occasionally we will also encounter ascending filtrations, in which case exhaustive means that the
filtered colimit up the filtration coincides (up to equivalence) with the complex itself.

We have deliberately stated the proposition without assuming that A is necessarily flat over k; there-
fore we must first define Hochschild and cyclic homology in this greater degree of generality. Although
the definitions of §2.1–2.2 do work without assuming such flatness, the resulting theory would less closely
related to topological Hochschild homology (in which all tensor products will be automatically derived,
thereby automatically overcoming the flatness issues) and Proposition 2.28 would not hold.
To define the Hochschild and cyclic homologies of a k-algebra which is not necessarily flat, we es-
sentially replace ⊗k by ⊗Lk . More precisely, we define the Hochschild homology HH(A/k) to be the
diagonal of the bisimplicial k-module HH(P• /k), where P• → A is a simplicial resolution of A by
polynomial k-algebras. In other words (recalling that the diagonal of a bisimplicial object models the
geometric realisation), we are defining HH(−/k) : k -algs → D(k) to be the left Kan extension of
HH(−/k) : k -algsΣ → D(k). Note that if A is flat over k, then the augmentations P•⊗k n → A⊗k n are
equivalences for all n ≥ 0, whence the diagonal of the bisimplicial k-module HH(P• /k) is equivalent to
HH(A/k) as it was defined in the flat case, i.e., the two definitions agree in the flat case.
Similarly, we define HC(A/k) as the totalisation of the simplicial cochain complex HC(P• /k), or in
other words as the left Kan extension of HC(−/k) : k -algsΣ → D(k). This agrees with the old definition
if A is flat over k, and the periodicity sequence and increasing filtration of Remark 2.13 formally remain
valid for general A.
One can alternatively define HH(A/k) and HC(A/k) using a resolution of A by a flat differential
graded k-algebra; this approach was adopted in [37, 38], for which reason HH(A/k) of non-flat algebras
is sometimes called Shukla homology.

Proof of Proposition 2.28. (i): Since HH(A/k) is obtained by totalising the bisimplicial object HH(P• /k),
where P• → A is a simplicial resolution by free k-algebras, it follows formally that the Postnikov filtration
on each HH(Pn /k) naturally induces a complete, descending N-indexed filtration on HH(A/k) whose
ith graded piece is given by HHi (P• /k)[i], i ≥ 0. But each Pn is a polynomial k-algebra, so Theorem 2.8
implies that HHi (Pn /k) ∼ = ΩiPn /k ; therefore HHi (P• /k) ∼
= ΩiP• /k , which we may rewrite up to equivalence
as ΩiP• /k ⊗LP• P• = ΩiP• /k ⊗P• A since each ΩiPn /k is a free Pn -module; but the latter simplicial module
is precisely LiA/k .
In short, the desired filtration on HH(−/k) : k -algs → D(k) is the left Kan extension of the Postnikov
filtration on HH(−/k) : k -algsΣ → D(k).
(ii): We begin by supposing that A ∈ k -algsΣ , or more generally that A is smooth over k. In terms
of HC(A/k) as the totalisation of the BC-bicomplex, we define FiliHKR HC(A/k) to be its subcomplex
obtained by totalising

A⊗k i+3 o A⊗k i+2 o Ker b o

B B B
0
b b b
  
A ⊗k i+2 o B
Ker b o
B
0
b b
 
Ker b o
B
0
b

0

(i.e., canonical truncation on the columns). The ith graded step of this filtration is clearly the totalisation

16
 AWS 2019 – Topological Hochschild homology

of

0o A⊗k i+2 / Ker b o Ker b o

B B B
0
b b b
  
A⊗k i+2 / Ker b o Ker b o
B B
0
b b
 
Ker b o
B
0
b

0
i.e., n≥0 [· · · ← 0 ← Ker b ← A⊗k i+2 / Ker b ← 0 ← · · · ][i + 2n] where we underline the term in degree
L

0. But each of the complexes in the direct sum is equivalent to HHi (A/k) ∼ = ΩiA/k , whence the graded
piece is equivalent to n≥0 ΩiA/k [i + 2n] as desired.
L

To construct the analogous filtration on HH(A/k), with graded pieces n≥0 LiA/k [i+2n], for arbitrary
L

A there are two (essentially equivalent) ways to argue: either rewrite the above argument replacing all the
columns by HH(P• /k), or else just formally declare the desired filtration on HC(−/k) : k -algs → D(k)
to be the left Kan extension of the just-construction filtration on HC(−/k) : k -algsΣ → D(k).
It remains to check that the filtration on HC(A/k) is complete, i.e., that Rlimi FiliHKR HC(A/k) ' 0.
But this is a formal consequence of the fact that FiliHKR HC(A/k) is supported in homological degrees
≥ i by construction.

Remark 2.30. The antisymmetrisation maps ΩnA/k → HHn (A/k) from Theorem 2.8 continue to exist
even if A is not flat over k; indeed, they are precisely the edge maps in the spectral sequence which arises
from the filtration of Proposition 2.28.

Remark 2.31 (Transitivity in general). Let A → B be a morphism of k-algebras, where A is commu-

tative. Then there is a natural equivalence of simplicial A-algebras
∼
HH(B/k) ⊗LHH(A/k) A → HH(B/A),

generalising the isomorphism of Remark 2.7 in the flat case. This can be proved either by left Kan
extending the aforementioned isomorphism, or else picking compatible resolutions and arguing as in the
proof of Proposition 2.27(iii).

Example 2.32. The Hochschild homology groups of the Z-algebra Fp are given by
(
Fp n even
HHn (Fp /Z) ∼
=
0 n odd

Regarding multiplicative structure, HH∗ (Fp /Z) = Fp hui (divided power algebra on a single variable u)
with u ∈ HH2 (Fp /Z) any basis element. In particular, the powers u2 , . . . , up−1 serve as basis elements
for HH4 (Fp /Z), . . . , HH2(p−1) (Fp /Z) respectively, but up = 0 in HH2p (Fp /Z). This is usually viewed as
pathological behaviour, which will be fixed by topological Hochschild homology.
There are two ways to prove this. Either use the HKR filtration of Theorem 2.28 and a standard
calculation of the wedge powers of the cotangent complex for a ring modulo a regular element (see
Remark 3.9, which also shows that HH2 (Fp /Z) naturally identifies with π1 (LFp /Z ) = pZ/p2 Z; this gives
a preferred choice of u, namely the class of p mod p2 ), or use the standard resolution of Fp by the Koszul
p
complex K(p; Z) = [Z ←− Z] (which is a flat cdg Z-algebra) and compute HH(Fp /Z) from the point of
view of Shukla homology. We refer to [28, Prop. 2.6].

17
 Matthew Morrow

3 Hochschild and cyclic homology of Fp -algebras

In Theorem 2.22 we saw that the cyclic homology theories of a smooth algebra in characteristic zero
were given by de Rham cohomology and its Hodge filtration. The goal of this section is twofold; firstly,
we will present a proof of the following analogous result in characteristic p:
Theorem 3.1. Let k be a perfect field of characteristic p and R a smooth k-algebra. Then HC − (R/k),
HP (R/k), and HC(R/k) admit natural complete, descending, Z-indexed filtrations whose ith graded pieces
are respectively given by
Ω≥iR/k [2i], Ω•R/k [2i], Ω≤i
R/k [2i].

In short, Theorem 2.22 remains true in characteristic p except that the filtration need not be split.
Theorem 3.1 is far from the best possible generalisation of Theorem 2.22 beyond characteristic zero; in
fact, Antieau has generalised it to smooth algebras over arbitrary base rings [2] (and has even removed
the smoothness hypothesis at the expense of replacing de Rham cohomology by its derived version).
However, the second goal of this section is to introduce the reader to the methods of [8], in which
“motivic” filtrations are constructed on the cyclic homology and its topological variants of Fp - and p-
adic algebras. Therefore we will prove Theorem 3.1 following the main technique of [8], namely descent
to the case of quasiregular semiperfect rings.
In §3.1 we present this technique by explaining how suitable cohomology theories can be determined
by their behaviour on certain semiperfect rings. This technique is then applied in §3.2 in order to
prove Theorem 2.22. Finally in §3.4 the methods are abstracted and extended by introduction of the
quasisyntomic site.

3.1 Flat descent via quasiregular semiperfect algebras

Given a commutative base ring k, suppose that we have a functor

F : k -algs → D(Z) or D(k) or Sp or in general any ∞-category D with sifted colimits.

In Remark 2.25 we recalled the theory of left Kan extension, which allows us to make precise the idea
that F might be determined by its value on polynomial algebras, and so in particular by its value on
smooth algebras. Now we consider a converse problem: can we recover the behaviour of F on smooth
algebras from its values on some class of “large algebras” (where, perhaps perversely, we hope that F is
easier to understand)?
Definition 3.2. F is said to satisfy flat descent (or to be an fpqc ∞-sheaf, where fpqc means fidèlement
plat et quasi-compact) if, for every faithfully flat map A → B of k-algebras, the induced morphism
//

F(A) → lim F(B) // F(B ⊗A B) /// F(B ⊗A B ⊗A B) // · · ·

in D is an equivalence. Inside the bracket we are applying F termwise to the usual Cech nerve
// B ⊗ B // //
Cech(B/A) := B A / B ⊗A B ⊗A B // · · ·

(which is a cosimplicial object in D), and the recall that the limit may be interpreted as totalisation in
case D = D(Z) or D(k).
Now we suppose that k is a perfect field of characteristic p and explain how flat descent allows us
to answer the converse problem we have posed. Let R be a smooth k-algebra, and recall first that the
Frobenius morphism ϕ : R → R is therefore flat. Indeed, to prove this we may work locally on Spec R
and so assume that there exists an étale morphism k[t1 , . . . , td ] → R for some d ≥ 0; the étaleness implies
that the diagram
k[t1 , . . . , td ] /R
O O
ϕ ϕ

k[t1 , . . . , td ] /R

18
 AWS 2019 – Topological Hochschild homology

is a pushout, and so the desired flatness follows from the obvious flatness of ϕ on k[t1 , . . . , td ]. By taking
the colimit over iterations of ϕ, we see that the colimit perfection Rperf := limϕ R is a flat R-algebra.
−→
Moreover ϕ always induces a homeomorphism on Spec, so Spec Rperf → Spec R is a homeomorphism, in
particular surjective; so we have shown that R → Rperf is faithfully flat.
Assuming that F satisfies flat descent, we deduce that
∼ // F(R // //

F(R) → lim F(Rperf ) perf ⊗R Rperf ) / F(Rperf ⊗R Rperf ⊗R Rperf ) // · · · ,

which describes F(R) in terms of F(Rperf ) and all the F(Rperf ⊗R · · · ⊗R Rperf ). The tensor powers
Rperf ⊗R · · · ⊗R Rperf are no longer perfect, but they are locally quotients of perfect k-algebras by
regular sequences. Indeed, assuming as in the previous paragraph that there is an étale morphism
k[t1 , . . . , td ] → R, one obtains


Rperf ⊗R · · · ⊗R Rperf = k[t]perf ⊗k[t] · · · ⊗k[t] k[t]perf ⊗k[t] R,

1/p∞ 1/p∞
where k[t]perf = k[t1 , . . . , td ]. But the large bracketed term is indeed a perfect k-algebra modulo
a regular sequence, for example
∞ ∞ ∞ ∞ j j j j
k[t1/p ] ⊗k[t] k[t1/p ] ∼
= k[x1/p , y 1/p ]/(x − y), t1/p ⊗ 1 7→ x1/p , 1 ⊗ t1/p 7→ y 1/p .

For practical reasons, being locally a quotient of a perfect algebra modulo a regular sequence is not a
good class of rings, since regular sequences are not the right notion for non-Noetherian rings; it is better
to work instead with the following wider class, which avoids any finiteness hypotheses:

Definition 3.3. Quillen calls an ideal I of a ring A quasiregular if and only if I/I 2 is a flat A/I-module
and πn (LA/I /A ) = 0 for n > 1 (whence LA/I /A ' I/I 2 [1] by Example 2.26) [34, Thm. 6.13]. For
example, if I is locally generated by a regular sequence then it is quasiregular.
We say that an Fp -algebra A is quasiregular semiperfect (qrsp) if and only if it is semiperfect (i.e.,
the Frobenius ϕ : A → A is surjective) and LA/Fp is a flat A-module supported in homological degree 1.

Lemma 3.4. An Fp -algebra A is quasiregular semiperfect if and only if there exists a perfect Fp -algebra
S and a quasiregular ideal I ⊆ S such that S/I = A.

Proof. We first recall the following general result: if S 0 → S is any morphism between perfect Fp -algebras,
then LS/S 0 ' 0. Indeed, let P• → S be a simplicial resolution by polynomial S 0 -algebras and observe
that the absolute Frobenius ϕ induces the zero map on Ω1P• /S 0 , since ϕ(df ) = df p = pf p−1 dp = 0; but S
and S 0 are perfect, so this zero map ϕ : LS/S 0 → LS/S 0 is also an equivalence.
Both hypotheses in the statement of the lemma include that A is semiperfect, so let S be any perfect
Fp -algebra surjecting onto A, and let I denote the kernel; the standard choice is A[ := limϕ A, the inverse
←−
limit perfection (aka. tilt) of A. Then LS/Fp ' 0 by the previous paragraph, so the transitivity sequence
shows that LA/Fp = LA/S ; by definition this is a flat A-module supported in homological degree 1 if and
only if I is quasiregular.

In light of the previous lemma and definition, the prior discussion may therefore be summarised by
the following proposition:

Proposition 3.5. Let k be a perfect field of characteristic p and F : k -algs → D (where D = D(Z) or
D(k) or Sp) a functor satisfying flat descent. Then, for any smooth k-algebra R, we have
∼
// F(Rperf ⊗R Rperf )
// /// · · · ),
/ F(Rperf ⊗R Rperf ⊗R Rperf


F(R) → lim F(Rperf ) /

and each term Rperf ⊗R · · · ⊗R Rperf appearing in the Cech nerve is quasiregular semiperfect.

And then here is the resulting filtration on F(R) which we have promised to construct in Theorem 3.1
(we remark that in all the cases of interest, Fi will vanish on any qrsp algebra for i odd and so we will
discard the odd terms and reindex the filtration):

19
 Matthew Morrow

Corollary 3.6. Under the same hypotheses as Proposition 3.5, suppose that D = D(Z), D(k), or Sp.
Then F(R) has a natural, complete, descending Z-indexed filtration with ith -graded piece given by the
[i]-shift of the cosimplicial abelian group
// F (R // //
Fi (Rperf ) i perf ⊗R Rperf ) / Fi (Rperf ⊗R Rperf ⊗R Rperf ) // · · ·

(or more precisely by its totalisation in D, i.e., associated cochain complex via Dold–Kan), where
Fi (−) := πi F(−).
Proof. For each A = Rperf ⊗R · · · ⊗R Rperf , we equip F(A) with its natural complete, descending, Z-
indexed Postnikov/Whitehead filtration τ>i F(A). This formally induces a filtration on F(R) by Propo-
∼
sition 3.5, which is still complete since the limits F(A) → limi τ≤i F(A) may be commuted through the
totalisation. The graded pieces of this filtration on F(R) are precisely
//

lim Fi (Rperf ) // F (R
i perf ⊗R Rperf )
/// Fi (Rperf ⊗R Rperf ⊗R Rperf ) // · · · [i],

as desired.

3.2 Cyclic homology of quasiregular semiperfect algebras

We have seen in §3.1, especially Proposition 3.5 and Corollary 3.6, how flat descent provides a general
method for constructing filtrations, by descending the Postnikov filtration from quasiregular semiperfect
Fp -algebras. Now we wish to put this technique into practice to prove Theorem 3.1. The first step is to
check that cyclic homology and its variants satisfy the necessary axiom for the technique, namely:
Lemma 3.7. For any commutative ring k, the D(k)-valued functors

Li−/k , HH(−/k), HC(−/k), HC − (−/k), HP (−/k)

on k -algs satisfy flat descent.

Proof. The fact that Li−/k satisfies flat descent is due to Bhatt; his first proof is given in [4], and his
shorter proof may be found in [8, Thm. 3.1].
Since the property of satisfying flat descent is itself a derived limit and hence closed under derived
limits (including fibre sequences), Proposition 2.28 and the previous paragraph imply that HH(−/k) and
HC(−/k) satisfy flat descent. The complete filtration on HC − (−/k) from Remark 2.14 then similarly
implies that HC − (−/k) satisfies flat descent, then deduce it for HP (−/k) thanks to the norm fibre
sequence of Remark 2.15.
The second step is ideally to compute the cyclic homology and its variants of quasiregular semiperfect
algebras; we begin with the following coarse information:
Lemma 3.8. Let A be a qrsp Fp -algebra. Then HH∗ (A/Fp ), HC∗− (A/Fp ), HP∗ (A/Fp ), and HC∗ (A/Fp )
are all supported in even degrees. Moreover, the Fp -algebra HP0 (A/Fp ) admits a complete, descending,
N-indexed filtration by ideals such that

Fili ∼ −
= HC2i (A/Fp ), HP0 (A/Fp )/ Fili ∼
= HC2i−2 (A/Fp ), gri ∼
= HH2i (A/Fp ) ∼
= πi (LiA/Fp )
for all i ≥ 0.
Remark 3.9. In the setting of the lemma, the hypothesis is that LA/Fp = N [1] is given by a flat A-
module supported in homological degree 1. Standard results on divided wedge powers then imply that
LiA/Fp is given by a flat A-module supported in homological degree i, namely ΓiA (N ), compatibly with
multiplication as i varies [23, Prop. I.4.3.2.1].
Here Γ∗A (N ) = i≥0 ΓiA (N ) is the divided power algebra associated to the module N . Since N is
L

flat, each A-module ΓiA (N ) is isomorphic to SymiA (N ), but the algebra structure on Γ∗A (N ) is given by
(i+j)!
x · y := i!j! xy x ∈ ΓiA (N ), y ∈ ΓjA (N ),

where x · y represents the product in Γ∗A (N ) and xy represents the product in Sym∗A (N ).

20
 AWS 2019 – Topological Hochschild homology

Proof of 3.8. As mentioned in the remark, the fact that L1A/Fp = N [1] implies that LiA/Fp = ΓiA (N )[i].
Therefore the spectral sequence associated to the filtration in Proposition 2.28 degenerates, implying
that HH∗ (A/Fp ) is supported in even degrees and given as an A-algebra by HH2∗ (A/Fp ) = Γ∗A (N ).
The rest of the proof is quite general, we use only that k := Fp → A is a morphism whose HH is
supposed in even degrees. It follows formally that the various periodicity sequences in Remark 2.15 all
break into short exact sequence, and that HC∗ (A/k), HC∗− (A/k), and HP∗ (A/k) are also supported in
even degree:
I S
0 −→ HH2i (A/k) −→ HC2i (A/k) −→ HC2i−2 (A/k) −→ 0

− − S h
0 −→ HC2i+2 (A/k) −→ HC2i (A/k) −→ HH2i (A/k) −→ 0

− I Sh
0 −→ HC2i (A/k) −→ HP2i (A/k) −→ HC2i−2 (A/k) −→ 0
'
HP2i (A/k) = lim HC2i+2s (A/k), S : HP2i (A/k) → HP2i−2 (A/k)
←−
s wrt S

There is therefore an identification of graded rings HP∗ (A/k) = HP0 (A/k)[u±1 ], with u := S(1) ∈
HP2 (k/k) corresponding to the periodicity operator S. Furthermore, I : HC∗− (A/k) → HP∗ (A/k) is an
inclusion of graded rings which is an isomorphism in degrees ≤ 0; therefore
− −
Fili HP0 (A/k) := u−i I(HC2i (A/k)) = S n I((HC2i (A/k)),

for i ≥ 0, defines a complete decreasing filtration of HP0 (A/k) by ideals with quotients and graded pieces
given respectively by

h h
' '
HP0 (A/k)/ Fili → HC2i−2 (A/k), gri → HH2i (A/k).

Remark 3.10. The moral of the previous lemma and its proof is the following: since HH∗ (A/Fp ) is
supported in even degree, the data of HC − , HP , and HC are captured (up to extension problems) by
the Fp -algebra HP0 (A/Fp ) and its filtration coming from HC − .

The coarse information afforded by Lemma 3.8 will actually be sufficient to prove Theorem 3.1, but
conceptually it is better to first try to explicitly identify the Fp -algebra HP0 (A/Fp ), for any qrsp algebra
A. We now construct two possible candidates; they will both be equipped with a complete, descending,
N-indexed filtration by ideals.

Definition 3.11 (de Rham construction). The first construction comes from the theory of derived de
Rham cohomology. For each i ≥ 0, we let LΩ≤i
−/Fp : Fp -algs → D(Fp ) be the left Kan extension of
the naively truncated de Rham complex Ω≤i
−/Fp ; recall from Remark 2.25 that this concretely means the
∼
following: given an Fp -algebra A, we pick a simplicial resolution P• → A by polynomial algebras over Fp
(possibly in infinitely many variables), and define LΩ≤i
A/Fp to be the geometric realisation of the simplicial
cochain complex Ω≤i
P• /Fp . The Hodge completed derived de Rham complex of the Fp -algebra A is then
defined to be
c A/F := Rlimi LΩ≤i ,
LΩ p A/Fp

which is an algebra object in D(Fp ) (possibly unbounded in both directions).

From the fibre sequences LiA/Fp → LΩ≤i ≤i−1
A/Fp → LΩA/Fp we see using Remark 3.9 that, assuming A
is quasiregular semiperfect, then each LΩ≤i A/Fp is supported in degree 0 and the transition maps are
surjective as i increases; therefore LΩA/Fp is still supported in degree 0, given by an Fp algebra which is
c
complete with respect to its induced filtration by the ideals Ker(LΩc A/F  LΩ≤i ) for i ≥ 0.
p A/Fp

21
 Matthew Morrow

Definition 3.12 (Divided power construction). The second construction comes from the theory of
divided powers. Assuming that A is quasiregular semiperfect (though in fact the definition only really
requires that A be semiperfect), let A[ := limϕ A be its inverse limit perfection and I := Ker(A[  A),
so that LA/Fp = I/I 2 [1] (see the proof of Lemma 3.4). Let DA[ (I) denote the divided power envelope
of A[ along I, and D b S (I) its completion with respect to the divided power filtration (equipped with the
induced filtration).

The following is the promised improvement of Lemma 3.8 which completely describes the cyclic,
negative cyclic, and periodic cyclic homology groups of quasiregular semiperfect Fp -algebras:

Proposition 3.13. Let A be a quasiregular semiperfect Fp -algebra. Then there are natural isomorphisms
of filtered rings
HP0 (A/Fp ) ∼
=D b A[ (I) ∼ c A/F ,
= LΩ p

where HP0 (A/Fp ) is equipped with the filtration coming from HC − defined in Lemma 3.8.

Comments on the proof. We begin with a reality check that the isomorphisms are plausible, or more
precisely valid for the associated graded rings. By construction of the filtered ring LΩ c A/F , its graded
p
i
pieces are precisely πi (LA/Fp ) for i ≥ 0. The same is true of the filtered ring HP0 (A/Fp ), by Lemma
3.8. Meanwhile, it is a classical general result in the theory of divided powers that if I is generated
by a regular sequence of A[ , then the filtered ring DA[ (I) has associated graded ring Γ∗A[ (N ), where
N = I/I 2 . Since we know from Remark 3.9 that Γ∗A[ (N ) and π∗ (L∗A/Fp ) are naturally isomorphic, we
have indeed shown that the three rings of the proposition have isomorphic graded rings (at least if I is
generated by a regular sequence).
The obvious way to proceed is therefore to construct natural morphisms between the three rings of
the proposition respecting the filtrations and to check that they induce the aforementioned isomorphisms
on each graded piece; since the three rings are complete with respect to their filtrations, these morphisms
will therefore be the desired isomorphisms. Fortunately the ring DA[ (I) has a universal property since it
is a divided power envelope, so a priori we can construct maps out of it by showing that the target ring
has suitable divided powers. Unfortunately it seems hard to directly write down natural divided power
structures on HP0 (A/Fp ) and LΩ c A/F . See project F if you want to try.
p

Instead, Bhatt has constructed a comparison map LΩ c A/F → D b A[ (I), by left Kan extending Berth-
p
elot’s comparison from de Rham to crystalline cohomology, and shown that it is an isomorphism [5,
Prop. 3.25 & Corol. 3.40] (assuming I is generated by a regular sequence). Some tricks in the style of
the proofs of [8, Props. 8.12 & 8.15] should extend this to arbitrary quasiregular I.
To then compare to HP0 (A/Fp ) there are two options. The first of these is based on Theorem 3.1,
which we will prove in a moment without using the current proposition. Upon left Kan extending that
theorem to all Fp -algebras, it yields in particular natural isomorphisms HC2i (A/Fp ) ∼ ≤i
= LΩA/Fp ; letting
i → ∞ gives HP0 (A/Fp ) ∼ c A/F , as desired.
= LΩ p
¯
The second approach for comparing to HP0 (A/Fp ) is to use topological cyclic homology; indeed, we
will see in Corollary 5.5 that the topological periodic cyclic homology TP0 (A) is a p-adically complete,
p-torsion-free ring such that TP0 (A)/p = HP0 (A/Fp ). The p-adic completeness and p-torsion-freeness
mean that existence of divided powers on TP0 (A) (which would then induce divided powers on the
quotient HP0 (A/Fp )) becomes a condition on certain elements of the ring, rather than extra data to be
specified. This allows the approach proposed in the second paragraph to be carried out; we refer to [8,
Thm. 8.15] for the details.

3.3 Cyclic homology of smooth algebras; proof of Theorem 3.1

This subsection is devoted to indicating the main ideas of proof of Theorem 3.1, so we let R denote a
smooth algebra over a perfect field k of characteristic p. We will restrict to the central case of periodic
cyclic homology; the analogous results for HC − and HC follow by working with a suitable step of the
filtrations instead, or respectively the quotient by a step of the filtration.

22
 AWS 2019 – Topological Hochschild homology

First proof of Theorem 3.1. If we allow ourselves to use Proposition 3.13, then Theorem 3.1 follows for-
mally from the descent machinery as follows. Thanks to the general technique of Corollary 3.6, the veri-
fication in Lemma 3.7 that HP (−/k) satisfies flat descent, and the fact from Lemma 3.8 that HP∗ (−/k)
vanishes in odd degrees on any qrsp ring, we see that HP (R/k) has a complete, descending, Z-indexed
filtration with graded pieces given by the [2i]-shifts of the cosimplicial k-module HP2i (Cech(Rperf /R)/k)

HP2i (Rperf /k) // HP2i (Rperf ⊗R Rperf /k) /// HP2i (Rperf ⊗R Rperf ⊗R Rperf /k) /// · · ·
/
We wish to show that this (or rather its totalisation) is equivalent to the de Rham complex Ω•R/k ; we
may suppose that i = 0, since HP∗ is 2-periodic.
Proposition 3.13 shows that HP0 (Cech(Rperf /R)/k) is the same as

// LΩ // c //
LΩ
c R /k c R ⊗ R /k / LΩRperf ⊗R Rperf ⊗R Rperf /k // · · · .
perf perf R perf

c R/k = Rlimi LΩ≤i since the functor LΩ

But the totalisation of this is equivalent to LΩ c −/k satisfies flat
R/k
i
descent (indeed, it is complete with respect to a filtration whose graded pieces are L−/k , and these satisfy
flat descent by Lemma 3.7. But R is smooth over k, so the adjunction LΩ≤i ≤i
R/k → ΩR/k is an equivalence
c R/k ' Rlim Ω≤i = Ω• , as desired.
and therefore LΩ R/k R/k

Second proof of Theorem 3.1. We now sketch a proof of Theorem 3.1 which uses only the coarse infor-
mation of Lemma 3.8 and does not rely on Proposition 3.13; as we explained in the proof of the latter,
it can then be deduced via left Kan extension from the proof of the current theorem.
As in the first proof, the goal is to show that the totalisation of HP0 (Cech(Rperf /R)/k) is equivalent to
Ω•R/k . The filtration on HP0 of any qrsp ring from Lemma 3.8 induces a complete, descending, N-indexed
filtration on HP0 (Cech(Rperf /R)/k) with ith -graded piece
// π Li // //
πi LiR i R / πi LiRperf ⊗R Rperf ⊗R Rperf /k // · · ·
perf /k perf ⊗R Rperf /k

Each of the cotangent complexes appearing here is supported in degree i, so we may remove the πi and
instead shift by [i]; its totalisation is then precisely ΩiR/k [i], thanks to flat descent for Li−/k .
To summarise, the coarse information of Lemma 3.8 is enough to show that HP0 (Cech(Rperf /R)/k)
has a complete descending filtration with graded pieces R[0], Ω1R/k [1], Ω2R/k [2],. . . . It follows quite
formally2 that HP0 (Cech(Rperf /R)/k) is necessarily equivalent to an actual complex of the form R →
Ω1R/k → Ω2R/k → · · · , where each differential is the boundary map H i → H i+1 induced by a fibre
sequence coming from the filtration (see the footnote for details). But the complex is even a differential
2 Here is a precise statement; for more details, including compatibility with symmetric monoidal structures, see [8, §5.1]:
Lemma 3.14. Let R be a commutative ring and C ∈ D(R) a complex equipped with a descending, N-indexed filtration
C = Fil0 C ← Fil1 C ← · · · such that, for each i ≥ 0, the graded piece gri C is supported in degree i. Then C is naturally
equivalent to the actual complex
0 −→ H 0 (gr0 C) −→ H 1 (gr1 C) −→ H 2 (gr2 C) −→ · · ·
where the boundary maps are the Bocksteins of the fibre sequences gri+1 C → Fili C/ Fili+2 C → gri C; moreover, the
filtration on C corresponds to the naive upwards truncations of this actual complex, and this process is compatible with
symmetric monoidal structures.

Proof. By proving the result for each C/ Filn+1 C and then letting n → ∞ we may suppose that the filtration on C is
finite of length n + 1, and by induction we may suppose that the result has been proved for any complex with a filtration
of length n. Then we have a fibre sequence and equivalence
H n (grn C)[−n] = Filn C −→ C −→ C/ Filn C ' [H 0 (gr0 C) → H 1 (gr1 C) → · · · → H n−1 (grn−1 C) → 0],
whence C is the homotopy fibre of
β
[H 0 (gr0 C) → H 1 (gr1 C) → · · · → H n−1 (grn−1 C) → 0] −→ H n (grn C)[−n + 1]
But β is precisely given by the desired Bockstein H n−1 (grn−1 C) → H n (grn C) and the homotopy fibre is given in the
desired way.

23
 Matthew Morrow

graded k-algebra since the above construction is compatible with product structures, so it is enough so
check that R → Ω1R/k is the usual de Rham differential. By picking an element of R and looking at the
corresponding map k[t] → R, functoriality reduces this to the case R = k[t], in which case one computes
explicitly to check it.

3.4 The quasisyntomic site

This subsection can be omitted when reading the notes for the first time. In fact, in the current version
of the notes it serves no purpose.
The arguments of §3.1–3.2 depended crucially not only on the cohomologiy theories of interest satis-
fying flat descent, but also on existence of the flat cover R → Rperf , for the smooth k-algebra R, having
the following properties:

(a) All terms of the Cech resolution Rperf ⊗R · · · ⊗R Rperf are qrsp.

(b) The flat cover Rperf is functorial in R (to ensure naturality of the filtrations we constructed).

Although we argued directly to prove (a), it is also easy to check (we will do it in Lemma 3.16)
that it is a consequence of LRperf /R being supported in homological degree 1, where it is given by a flat
Rperf -module. There do exist other rings admitting such a nicely behaved flat cover:

Definition 3.15. Given a morphism A → B of Fp -algebras, we say that it is quasisyntomic if it is flat

and LB/A has cohomological Tor amplitude in [−1, 0], i.e., for any B-module M , the complex LB/A ⊗LB M
is supported in cohomological degrees [−1, 0]. We say that it is a quasisyntomic cover if, in addition, it
is faithfully flat.
We say that an Fp -algebra A is quasisyntomic if Fp → A is quasisyntomic in the above sense. Let
QSyn denote the category of quasisyntomic Fp -algebras (and all morphisms, not just quasisyntomic
ones).

Quasisyntomic rings are precisely those which have a cover behaving similarly to R → Rperf :

Lemma 3.16. An Fp -algebra A is quasisyntomic if and only if there exist a qrsp Fp -algebra S and a
quasisyntomic cover A → S. When this holds, all terms S ⊗A · · · ⊗A S of the Cech nerve are qrsp.

To prove Lemma 3.16 it is helpful to first note the following general properties, which follow from
the basic properties of the cotangent complex, especially transitivity (Proposition 2.27), so we leave the
verifications to the reader:

Lemma 3.17. (i) Let A → B be a quasisyntomic cover of Fp -algebras; then A is quasisyntomic if and
only if B is quasisyntomic.

(ii) A composition of quasisyntomic morphism (resp. quasisyntomic covers) again a quasisyntomic

morphism (resp. quasisyntomic cover).

(iii) A pushout of a quasisyntomic morphism (resp. quasisyntomic cover) along an arbitrary morphism
is again a quasisyntomic morphism (resp. quasisyntomic cover).

Proof of Lemma 3.16. The implication ⇐ is a consequence of Lemma 3.17(i). For the converse, suppose
that A is quasisyntomic, let Fp [xi : i ∈ I]  A be a surjection from a polynomial algebra and let
1/p∞
Fp [xi ] = Fp [xi : i ∈ I]perf be its colimit-perfection, which is easily seen to be a quasisyntomic cover
1/p∞
of Fp [xi ]; finally let S := A ⊗Fp [xi ] Fp [xi ] be the resulting pushout. Then A → S is the pushout of a
quasisyntomic cover, hence is a quasisyntomic cover (by Lemma 3.17); finally note that S is semiperfect,
1/p∞
since it is quotient of Fp [xi ].
To prove the Cech statement, let A → S be any quasisyntomic cover where S is qrsp. By taking
repeated pushouts and compositions, Lemma 3.17(ii)&(iii) shows that A → S ⊗A n is quasisyntomic
whence S ⊗A n is quasisyntomic; but it is also clearly semiperfect since S is semiperfect.

24
 AWS 2019 – Topological Hochschild homology

Remark 3.18. A stronger condition than quasisyntomic is quasismooth, which means that A → B is flat
and LB/A has Tor amplitude in [0, 0], i.e., LB/A is supported in degree 0 and Ω1B/A is a flat B-module.
Just as in Lemma 3.17(ii)&(iii), compositions and pushouts of quasismooth maps are again quasismooth.
Example 3.19. If k is a perfect field (in fact, any perfect ring) of characteristic p and A is a smooth
k-algebra (resp. perfect k-algebra), then A is quasismooth; if I ⊆ A is a quasiregular ideal (e.g., generated
by a regular sequence, then A/I is quasisyntomic). These are our main examples of interest, but the
advantage of adopting an abstract definition in terms of the cotangent complex is that it avoids any
finiteness hypothesis (which allows us to use, for example, polynomial rings in infinitely many variables,
possibly over non-Noetherian perfect rings).
It follows from Lemma 3.17 that the opposite of QSyn is a site, with covers declared to be qua-
sisyntomic covers. In other words, we have cut down the flat site on all Fp -algebras by restricting to
quasisyntomic algebras and by insisting that flat covers also be quasisyntomic. Lemma 3.16 gives us
interesting covers with the same properties as (a), but without any naturality properties as in (b); next
we explain the formalism which overcomes this lack of naturality.
Note that an Fp -algebra A is quasiregular semiperfect if and only if it is quasisyntomic and semiperfect;
so we write QRSP ⊆ QSyn for the subcategory of qrsp algebras. The first assertion of the Lemma 3.16
implies in particular that QRSPop forms a basis for QSynop , as we showed that each object in the former
site can be covered by an object in the latter. It follows that the categories of abelian sheaves on QSynop
and on QRSPop are equivalent via the obvious restriction functor. Using the second assertion of Lemma
3.16, this equivalence extends quite formally from abelian sheaves to ∞-sheaves taking values in any
presentable ∞-category D:
Sh(QSynop , D) → Sh(QRSPop , D)
The key point is that, given an ∞-sheaf F : QRSPop → C, we may extend it to an ∞-sheaf F i :
QSynop → C by setting
F i (A) := lim F(S) // F(S ⊗A S) /// F(S ⊗A S ⊗A S) /// · · ·
,
/
where A ∈ QSyn and A → S is a chosen quasisyntomic cover with S ∈ QRSP. The equivalence of
∞-categories assures us that this definition of F i is independent (up to equivalence) of the chosen cover
A → S, which is how we overcome the lack of the functorial cover R → Rperf which we had in the smooth
case. Following [8], the extension F i will be called the unfolding of F.
In particular, suppose that F is a presheaf of abelian groups on QSynop satisfying the following Cech
condition:
For any quasisyntomic cover A → B between qrsp algebras, the Cech complex 0 → F(A) →
F(B) → F(B ⊗A B) → · · · is exact, i.e., the higher Cech cohomology of F on A with respect
to the cover B vanishes, and the 0th Cech cohomology agrees with F(A). Or, in other words
F|QRSP , viewed as functor to D(Z) is an ∞-sheaf on QRSPop .
Then a general result about sites [Stacks Project, 03AN & 03F9] implies that F|QRSP is an abelian
sheaf on QRSPop and that it has no higher (site-theoretic) cohomology on any object of QRSPop . It
follows also that the above Cech complex defining (F|QRSP )i (A) is nothing other than the cohomology
RΓQSyn (A, F) (we should more correctly write the sheafification F sh of F here instead, but we will
usually lighten the notation by leaving the sheafification implicit when taking cohomology).
The previous paragraph applies in particular to our sheaves of interest:
Lemma 3.20. The following abelian presheaves F on QSynop all satisfy the previous Cech condition:
(i) πn (Li−/Fp ) for any i, n ≥ 0.

(ii) HHn (−/Fp ), HCn (−/Fp ), HCn− (−/Fp ), HPn (−/Fp ) for any n ∈ Z.
Proof. (i): If i 6= n then πn (Li−/Fp ) is identically zero on QRSP by Remark 3.9, so it remains to deal with
the case i = n. But then we have L−/Fp = π1 (L−/Fp )[1], so we must check that for any quasisyntomic
cover A → B between qrsp algebras we have
/ //

∼
LiA/Fp → lim LiB/Fp // LiB⊗ B/Fp // LiB⊗A B⊗A B/Fp // · · ·
A

25
 Matthew Morrow

But this is exactly flat descent for the cotangent complex (Lemma 3.7).
n/2
(ii): As we saw at the start of Lemma 3.8, each HHn (−/Fp ) is either zero or given by πn/2 (L−/Fp ),
to which we apply part (i). The Cech condition for the others then follows from the filtrations of Lemma
3.8.
As a concrete example, the lemma and prior discussion mean that the Cech complex HP0 (Cech(Rperf /R)/k)
which we studied in §3.2 can alternatively be described as

HP0 (Cech(Rperf /R)/k) = RΓQSyn (R, HP2i (−/Fp )),

the quasi-syntomic cohomology of (the sheafification of) HP0 (−/k); and the same holds for HCn− (−/Fp ),
HPn (−/Fp ), and HCn (−/Fp ).

4 S 1 -actions (three points of view) and THH

So far we have studied Hochschild and cyclic homology in terms of relatively explicit, but unmotivated,
double complexes. In this section we will explain that there has been an action of the circle S 1 lurking in
the background; although it may initially appear absurd to discuss S 1 -actions on algebraic objects such
as chain complexes, we will see several ways of making this precise. This S 1 -action will play an essential
role once we pass to topological cyclic homology, where the algebraic manifestations of this action will
no longer be available.
First, in §4.1, we will explain that mixed complexes (Definition 2.19) may be viewed as complexes
equipped with an S 1 -action. Then in §4.2 we we drop any linearity assumptions and return to cyclic
objects (Definition 2.17), which provide a different way of algebraically encoding an S 1 -action. In §4.3 we
will discover an even better (and our final) point of view, namely that of functors out of the classifying
space of S 1 . Once that formalism is available, we explain in §4.4 that the classical definition of Hochschild
homology can be transported to any suitable symmetric monoidal (∞-)category, such as that of spectra;
the result is topological Hochschild homology.

4.1 As k[ε]/ε2 -modules, i.e., mixed complexes

Given a fixed commutative base ring k, here we explain how mixed complexes over k provide a way of
modelling what it means for a complex to be equipped with an S 1 -action.
Note first that in the case of a discrete group G, a module M or complex with G-action is simply
a module or complex over the group algebra k[G]; moreover, the group homology and cohomology are
then given respectively by the derived coinvariants k ⊗Lk[G] M and derived invariants RHomk[G] (k, M ).
Remark 4.1. P The case in which G is a finite group is particularly interesting. Then the norm map
k → k[G], 1 7→ g∈G g is well-defined and k[G]-linear (viewing k as a k[G]-algebra via the augmentation
map), thereby inducing a natural transformation k⊗Lk[G] − → RHomk[G] (k, −). Given a k-module M with
G-action, the cofiber of this norm map k ⊗Lk[G] M → RHomk[G] (k, M ) is known as the Tate cohomology
(the same Tate cohomology which often appears, when G is a cyclic group, in class field theory).
Now we turn to the case of a topological group G. Possibly replacing G by a weakly equivalent
topological group we may assume that G is the geometric realisation of a simplicial group G• (for
example, the Kan complex of simplices Sing(G) is a simplicial group whose geometric realisation is
weakly equivalent to G, though in practice we will be interested in choosing some G• which is much
smaller than Sing(G)). Then, in any category C, we could define “an object with G-action” to be a
simplicial object X• of C which is equipped with an action by G• ; i.e., each Xn is equipped with an
action by Gn in a manner compatible with the face and degeneracy maps. In particular, if X• were
a simplicial set, or simplicial k-module, equipped with a G• -action in this sense, then its geometric
realisation |X• | would be a topological space equipped with a continuous action by G = |G• | (simply
because geometric realisation is functorial and compatible with products). Conversely, if X were any
topological space equipped with a continuous action by G = |G• |, then Sing(X) would be equipped with
an action by Sing(|G• |), which is equivalent to G• . In conclusion, up to issues of homotopy equivalence,

26
 AWS 2019 – Topological Hochschild homology

actions by the simplicial group G• offer a combinatorial model for actions by G. We will return to
this point of view in §4.2, but now we wish to linearise by restricting to complexes of k-modules: then
k-linear actions are modelled by modules over the simplicial k-algebra k[G• ], or equivalently connected
differential graded modules over the dg k-algebra N k[G• ] associated to the simplicial k-algebra k[G• ].
Remark 4.2. We have just used the following fact: given a simplicial ring A• , then its normalised
cochain complex N A• naturally admits the structure of a differential graded ring, with product given
by the so-called “shuffle product”. Note also that if A• is commutative, then N A• is strictly graded
commutative.
We note also that passing from simplicial modules over k[G• ] to connective dg modules over N k[G• ]
does not lose any information: the two model categories are Quillen equivalent via the normalised complex
construction [36, Thm. 1.1(2)].
We specialise now to the circle S 1 , which is equivalent to the geometric realisation of BZ, the
classifying space construction on the infinite cyclic group:
o o
BZ = 0 oo Z oo Z × Z ooo ···

(a2 , . . . , an )
 i=0
di (a1 , . . . , an ) := (a1 , . . . , ai + ai+1 , . . . , an ) 0 < i < n si (a1 , . . . , an ) := (a1 , . . . , ai , 0, ai+1 , . . . , an )

(a1 , . . . , an−1 ) i=n


Lemma 4.3. The dg k-algebra N k[BZ] associated to the simplicial k-algebra k[BZ] is quasi-isomorphic
to the dg k-algebra k[ε]/ε2 , where ε lies in homological degree 1 and the differential kε → k is zero.
Proof. Slick proof: since BZ is equivalent to Sing(S 1 ), the associated simplicial group algebra k[BZ] is
equivalent to Sing(S 1 ; k), which in degree n is by definition the k-module generated by all n-simplices
∆n → S 1 . Passing to dgas, we must compute the usual dga computing H∗ (S 1 , k), which is indeed k[ε]/ε2 .
d d d
Tn Hands-on proof:n the dga N k[BZ] is the normalised complex N0 ←0 N1 ←0 N2 ←0 . . . , where Nn =
±1 ±1
i=1 Ker dn ⊆ k[Z ] = k[t1 , . . . , tn ]. It is a good exercise in understanding definitions to check that
this is indeed quasi-isomorphic to the desired k[ε]/ε2 .
Therefore, once again up to issues of homotopy, we may view a connective complex C of k-modules
as being equipped with an action by S 1 as soon as it is given the structure of a differential graded
k[ε]/ε2 -module. Concretely, this means specifying a k-linear map ε : C → C[1] such that ε2 = 0; but
this is precisely the notion of a mixed complex from Definition 2.19! In fact, when mixed complexes were
first introduced by Burghelea, they were called algebraic S 1 -complexes.
Next we discuss cyclic and negative cyclic homology from this point of view. Observe that k, as a
kε := k[ε]/ε2 -module, has a periodic resolution by free kε -modules
∼ ε ε ε
k ← kper := [kε ← kε [1] ← kε [2] ← · · · ].

Given a mixed complex C over k, i.e., a connective dg k[ε]/ε2 -module, one easily checks the following:
ε ε
- HC(C/k) = [kε ← kε [1] ← · · · ] ⊗k[ε]/ε2 C, which represents k ⊗Lk[ε]/ε2 C, i.e., the “group homology
of the action S 1  C”
- HC − (C/k) = Homk[ε]/ε2 (kper , C), which represents RHomk[ε]/ε2 (k, C), i.e., the “group cohomology
of the action S 1  C”
- Moreover, the inclusion k[1] = kε ,→ k[ε]/ε2 is k[ε]/ε2 -linear and plays the role of the norm map
from Remark 4.1; it induces k ⊗Lk[ε]/ε2 C[1] → Rhomkε (k, C), i.e., HC(C/k)[1] → HC − (C/k).
The reader should check that this map is precisely the boundary map in the “norm sequence” of
Remark 2.15, whence its cofiber “Tate cohomology of the action S 1  C” is HP (C/k).
In conclusion, S 1 -actions on connective chain complexes may be modelled as mixed complexes, in which
case HC, HC − and HP represent respectively group homology, cohomology, and Tate cohomology of the
action.

27
 Matthew Morrow

4.2 As cyclic objects

Before restricting to the k-linear situation in §4.1, we observed that “an object with G-action” could be
modelled as a simplicial object in the category C equipped with an action by the simplicial group G•
(where G ' |G• |). Specialising to the case of S 1 , realised by the simplicial group BZ, we could therefore
study simplicial objects X• equipped with an action by BZ, i.e., each Xn should be equipped with n
commuting automorphisms satisfying certain compatibilities with respect to the face and degeneracies.
Unfortunately it seems that few concrete examples of such BZ-actions exist; in particular, the Hochschild
complex does not admit such an action on the nose (as far as the author is aware).
The problem is that although BZ is a simplicial group realising S 1 , it is enormous when compared
1
to the standard simplicial set Ssimp realising S 1 , whose only non-denerate simplices are one vertex and
one 1-simplex:
o o
1
Ssimp = {v} oo {v, e} oo {v, s1 e, s0 e} ooo ···
1
Unfortunately, Ssimp is not a simplicial group (it is not even a Kan complex) and therefore it does not
make sense to discuss its actions in the above sense. Remarkably, it does however admit the structure of a
so-called crossed simplicial group, a notion introduced in the late 1980s by Krasauskas and Fiedorowicz–
Loday [18] in order to provide much smaller simplicial models of certain topological groups by slightly
weakening the concept of the simplicial group.

Definition 4.4. A crossed simplicial group is a simplicial set G such that each Gn is a group, together
with group homomorphisms Gn → Aut({0, . . . , n}), such that for all 0 ≤ i ≤ n

(i) di (gh) = di (g)dg−1 (i) (h), si (gh) = si (g)sg−1 (i) (h), and

(ii) the following diagrams of sets commute

dg−1 (i) sg−1 (i)

[n − 1] / [n] [n + 1] / [n]

di (g) g si (g) g
   
[n − 1] / [n] [n − 1] / [n]
di si

An action of a crossed simplicial group G• on a simplicial set X• (or many generally on a simplicial
object in an arbitrary category) is the data of actions of Gn on Xn , for each n ≥ 0, such that di (gx) =
di (g)(dg−1 (i) (x)) and si (gx) = si (g)(sg−1 (i) (x)) for all 0 ≤ i ≤ n.

Example 4.5. (i) Any simplicial group G• may be viewed a crossed simplicial group, by declaring
the representations Gn → Aut({0, . . . , n}) to be trivial.

(ii) Connes’ cyclic crossed simplicial group C is defined by Cn = Z/n + 1, with the representation
Z/n + 1 → Aut({0, . . . , n}) being cyclic permutation and with simplicial structure isomorphic to
1
Ssimp given by the bijections

Z/n + 1 = {0, . . . , n} ∼ 1
= Ssimp,n , 0 7→ v, 0 < i 7→ sn−1 · · · si si−2 · · · s0 (e).

It can be shown that the discussion after Remark 4.1 extends to crossed simplicial groups:

Proposition 4.6 ([18, §5]). Let G be a crossed simplicial group. Then:

(i) Its geometric realisation |G| naturally admits the structure of a topological group.

(ii) Given an action of G on a simplicial set X• , there is an induced continuous action of |G| on |X• |.

(iii) Adjoint: if X is a topological space equipped with a continuous action by |G|, then there is an
induced action of G on the simplicial set Sing(X).

28
 AWS 2019 – Topological Hochschild homology

Proof. The key observation is that, given any simpicial set X• , there exists a simplicial set FG (X• ) with
G-action which models the product G × X• . More precisely, define FG (X• ) as follows:
FG• (X• )n := Gn × Xn , di (g, x) = (di (g), dg−1 (i) (x)), si (g, x) = (si (g), sg−1 (i) (x))
It is easy to check that FG (X• ) admits a G action (by the obvious left action of each Gn on Gn × Xn ),
and that there is even a monadic structure FG (FG (X• )) → FG (X• ), (g, h, x) 7→ (gh, x). Importantly, it
can be shown that there is a natural homeomorphism of geometric realisations
'
|FG (X• )| → |G| × |X• |.
So, assuming now that X• is equipped with a G-action, whence there is an action map FG (X• ) →
X• (equivalently, X• is an algebra over the monad FG ), passing to geometric realisations provides a
continuous map |G| × |X• | → |X• |. Consideration of the special case X• = G and the axioms for a
monad easily proves (i) and (ii).
Part (iii) is slightly more involved so we refer the reader to [18, §5].
Remark 4.7. Part (i) of the proposition can be made more precise. For each n ≥ 0, there is an inclusion
1
of Gn into |G| as a discrete subgroup, given by g 7→ [g, ( n+1 1
, . . . , n+1 )]. When n + 1 divides n0 + 1 these
n0 +1
maps are moreover compatible with the degeneracy [n0 ] → [n] sending the first elements to 0, the
n+1
n0 +1
S
next n+1elements to 1, etc., so that Gn ⊆ Gn0 ⊆ |G|. Passage to the limit identifies n≥0 Gn (a filtered
union under the ordering n + 1|n0 + 1) as a dense subgroup of |G|.
Thanks to Proposition 4.6 we see that, for any crossed simplicial group G, a combinatorial way to
model |G|-actions is via G-actions. In particular, we may model S 1 ' |C|-actions by studying simplicial
objects equipped with an action by Connes’ cyclic group C. But, applying Definition 2.19 in the case of
crossed simplicial group C, we see that an object with C-action is precisely a cyclic object in the sense
of 2.17. In conclusion, cyclic objects in a category model S 1 -actions.
Having heuristically claimed that both mixed complexes and cyclic objects model S 1 -actions, we
should compare the two approaches. As we saw in Section 2.2, the un-normalised chain complex con-
struction defines a functor
{cyclic k-modules} −→ {mixed complexes over k}
This functor is indeed an equivalence up to issues of homotopy (more precisely, it can be upgraded to
a Quillen equivalence of model categories, or an equivalence of ∞-categories). We refer to [20, §II.2]
[16, 17] [22, Corol. 2.4] for further details.

4.3 As functors BS 1 → C
Given a discrete group G, the data of an object of a category C equipped with a G-action is the same as
giving a functor G → C, where we follow a standard abuse of notation of denoting by G the category with
a single object ∗ having endomorphisms EndG (∗) = G. Given that the process of replacing a category
by its nerve is fully faithful, this is then the same data as a morphism of simplicial sets (i.e., functor of
∞-categories) BG = N (G) → N (D).
However, let us now replace N (D) by a homotopically richer category D, such as a dg, simplicial,
or ∞-category (in practice it will be D(k) or Sp); we will begin to use some elementary language of
∞-categories (we refer the reader to §4.5 for a short introduction). Then a functor BG → D is now
∼
slightly more subtle: it represents an object X ∈ D, an auto-equivalence g : X → X for each g ∈ G, a
2-cell
X
g
/X h /X,
C

hg

for each g, h ∈ G, 3-cells connecting the 2-cells, and so on. That is, the action of G on X is no longer
associative on the nose, but only up to higher homopoties in a coherent fashion.
This hopefully motives the following definition:

29
 Matthew Morrow

Definition 4.8. Given a topological group G, an object of D with G-action is a functor BG → D, where
BG is a fixed Kan complex whose geometric realisation is a classifying space for G (as a topological
group). The ∞-category of objects of D with G-action is thus DBG := Fun(BG, D).
Remark 4.9. (i) Any two choices of simplicial classifying space BG are homotopy equivalent, hence
categorically equivalent as ∞-categories since they are Kan complexes, and therefore DBG is well-
defined up to categorical equivalence. If G is given to us as the geometric realisation of a simplicial
group G• , then there are two common choices for BG, both built from the bisimplicial set BG• :
either its diagonal or its W -construction (which can be checked directly to be equivalent [?, ?]).
(ii) In the special case that D = N (C) is the nerve of a usual category C, then the adjunction between
nerves and homotopy categories shows that objects of N (C) with G-action are simply objects of C
with an action by the discrete group π0 (G) in the usual sense. To get interesting examples it is
essential that D be homotopically richer.
To compare Definition 4.8 to mixed complexes and cyclic objects we will consider two possible choices
of simplicial classifying space BS 1 .
Firstly, it is well-known that the classifying space of S 1 is CP∞ ' K(Z, 2) ' |BBZ| [Mixed complex
point of view to be added]
Our second model for the classifying space for S 1 comes from Connes’ cyclic category:
Lemma 4.10. The geometric realisation |N (Λ)| of the cyclic category Λ is a classifying space for S 1 .
Proof. We will sketch one method of proof; for further details see [18, 5.9–5.12], for other approaches,
see [32, App. T] [28].
One applies Quillen’s Theorem B to the inclusion of categories i : ∆ ,→ Λ; recall that this is faithful
and bijective on objects. For any object [m] ∈ Λ, the comma category [m]\i has objects given by all
morphisms f in Λ with codomain [m], and has morphisms given by commutative diagrams

[n] VVVVV f
VVVV
VVVV
+
4 [m]
hhhh
λ
 hhhh 0hh h
h f
[n0 ]

where λ ∈ Hom∆ ([n], [m]) ⊆ HomΛ ([n], [m]). Thus its nerve N ([m]\i) is the simplicial set HomΛ (−, [m]).
But any morphism [n] → [m] in Λ may be unique decomposed as an automorphism of [n], i.e., an element
of Z/n + 1, followed by a morphism [n] → [m] in ∆; this is part of the combinatorics which we omitted
when defining the cyclic category. In other words, HomΛ ([n], [m]) = Z/n + 1 × Hom∆ ([n], [m]); one
then checks that the simplicial structure on the right side as n varies is precisely that which appeared
in the construction FC in the proof of Proposition 4.6, i.e., HomΛ (−, [m]) = FC (∆m ), where ∆m :=
Hom∆ (−, [m]) is the m-simplex. [To finish; in the meantime, also for the details of the next paragraph,
see [18]]
Quillen’s Theorem B therefore implies that there is a homotopy fibre sequence

S 1 ' |C| −→ |N (∆)| −→ |N (Λ)|

But the geometric realisation of ∆ is contractible (since, e.g., ∆ has an initial object), and so S 1 '
Ω|N (Λ)|. Taking care that this sufficiently preserves group structures (this is where we are omitting
some details), one completes the proof by recalling that any space X serves as a classifying space for
ΩX.
The canonical unit map of simplicial sets N (Λ) → Sing(|N (Λ)|) may therefore be written as N (Λ) →
BS 1 , which by restriction induces a morphism
1
DBS −→ Fun(N (Λ), D),

i.e., each object X of D with S 1 -action determines a cyclic object X• of D (this is a simplicial object X•
∼
of D equipped with tn : Xn → Xn for each n ≥ 0, such that the axioms of Definition 2.17 are satisfied up

30
 AWS 2019 – Topological Hochschild homology

to higher homotopies in a coherent fashion). Informally, X• is defined by taking the constant simplicial
object Xn := X for all n, with the cyclic operator tn on Xn corresponding to the action of the cyclic
group Z/n + 1 ⊆ S 1 on X.
We can also go the other direction (assuming that D admits geometric realisations of simplicial
objects, which is true in cases like D(k) or Sp): there is also a functor
1
Fun(N (Λ), D) −→ DBS

which takes a cyclic object X• in D and gives |X• | ∈ D (the geometric realisation of the underlying
simplicial object of X• ) equipped with a certain natural S 1 -action in the above sense. This is the
essentially surjective right adjoint to the functor of the previous paragraph.

Remark 4.11. The previous two paragraphs are probably better known to the reader in the simpler
case of ∆ instead of Λ. Then |N (∆)| is contractible and so the canonical unit map N (∆) → Sing(|N (∆)|)
may be written as N (∆) → ∗. This induces the adjoint pair

constant simplicial object

-
D = Fun(∗, D) m Fun(N (∆), D)
geometric realisation

Given an object of D, in general we do not care whether/how it arose as the geometric realisation of
some simplicial diagram, so we work in the left category rather than the right.
1
Analogously to the remark, we do not care whether/how a given object of DBS arose as the geometric
realisation of some cyclic object. In conclusion, although we may pass through various constructions to
1
get there, the final home for our objects with S 1 -actions will be DBS .

4.4 The cyclic bar construction

Having hopefully justified in the section thus far that the correct framework in which to study S 1 -actions
1
is DBS , we should now offer an example. We will abstract the construction of Hochschild homology,
rapidly leading to its topological analogue.
Recall that an (associative) algebra object in a symmetric monoidal category (C, ⊗, 1C ) is the data of
an object A ∈ C together with morphisms µ : A⊗A → A (=multiplication) and 1C → A (= multiplicative
identity) which satisfy the usual commutative diagrams specifying the axioms of an algebra. Given such
an algebra object A, one may write down a simplicial object
o o
A oo A ⊗ A oo A ⊗ A ⊗ A ooo ···

in C by copying the rules for the face and degeneracy maps used in the definition of Hochschild homology
from Remark 2.3. For example, the three face maps d0 , d1 , d2 : A ⊗ A ⊗ A → A ⊗ A are given by µ ⊗ id,
id ⊗µ, µ ⊗ id ◦t, where t is the cyclic endomorphism of A ⊗ A ⊗ A given by a ⊗ b ⊗ c 7→ c ⊗ a ⊗ b. Moreover,
by declaring tn to be the analogous cyclic endomorphism of A⊗n+1 one upgrades this simplicial object
to have the structure of a cyclic object in C, known as the cyclic bar construction of A; it is sometimes
denoted by B•cyc (A).
We now wish to categorise this construction so that it may be applied to associative ring spectra, i.e.,
algebra objects for which the axioms of an algebra are only satisfied up to higher homotopties. The first
step is to observe that algebra objects in C are classified by functors Ass⊗act → C (Lemma 4.13 below),
where the domain category is defined as follows:

Definition 4.12 (Associative algebras). The category Ass⊗ act is defined as follows: objects are finite sets,
and a morphism from T to S is the data of a map of sets f : T → S together with linear orderings ≤s
on the preimages f −1 (s) for all s ∈ S. The composition of f with g : S → R is given by the set-theoretic
composition gf : T → R together with the lexicographic orderings on the preimages; i.e., given r ∈ R
and t, t0 ∈ (gf )−1 (r), then t <r t0 iff “f (t) <r f (t0 )” or “f (t) = f (t0 ) =: s and t <s t0 ”.

31
 Matthew Morrow

The category Ass⊗act is symmetric monoidal under disjoint union, with unit element ∅. Note that
the element {1} ∈ Ass⊗ act admits the structure of an algebra object, with multiplication µ : {1, 2} =
{1} t {1} → {1} corresponding to the ordering 1 < 2. In fact, as promised, {1} ∈ Ass⊗ act represents
algebra objects in general:

Lemma 4.13. The category of algebra objects in C (the notion of a morphism of algebra objects should
be clear) identifies with the category of monoidal functors F : Ass⊗
act → C, via F 7→ F ({1}).

Proof. A monoidal functor obviously sends algebra objects to algebra objects, so the functor F 7→ F ({1})
with induced multiplication is indeed an algebra object in C.
Conversely, from any algebra object (A, µ : A ⊗ A → A, 1C → A), we may define a monoidal functor
A⊗ : Ass⊗ ⊗
act → C by sending a finite set S to A (S) := ⊗s∈S A, and by sending a morphism f : T → S to
Y
⊗t∈T A −→ ⊗s∈S A, ⊗t∈T at 7→ ⊗s∈S at .
t∈f −1 (s)

Here we abusively write the product µ as though A were a set equipped with a law of Q multiplication,
leaving the reader the write a categorical formula; the important point is that the product t∈f −1 (s) at :=
at1 · · · atm is well-defined, since f −1 (s) = {t1 <s · · · <s tm } is equipped with a total order.
cryc
The algebra object {1} ∈ Ass⊗
act has an associated cyclic bar construction B• ({1})
o o
{1} oo {1, 2} oo {1, 2, 3} ooo · · ·.

This is a cyclic object in Ass⊗act , so corresponds to a certain functor Λ

op
→ Ass⊗act , which is denoted
by Cut (and which can be described explicitly should one wish [32, App. T] [28]). Purely formally, it
follows that the cyclic bar construction B•cryc (A) of a general algebra object A ∈ C is now given by the
functor
Cut A⊗
Λop −−→ Ass⊗
act −−→ C,
where A⊗ is the functor corresponding to A in the sense of Lemma 4.13.
We are now prepared to extend the constructions to the ∞-world:

Definition 4.14. Let D be a symmetric monoidal infinity category. An (associative) algebra object in D
is a functor A : N (Ass⊗act ) → D; note that this is the data of objects A1 , A2 , · · · ∈ D, various equivalences
An ⊗ Am ' An+m , a multiplication map A2 → A, an identity map 1D → A1 , homotopies between
them to express “associativity up to homotopy”, etc. Also note that this is equivalent to the operadic
definition of E1 -algebras in D (see [28, Def./Prop. 3.3] and [31, Prop. 2.2.4.9]).
The cyclic bar construction B•cyc (A) of an algebra object is now defined to be the functor

N (Cut) A
B•cyc (A) : N (Λop ) −−−−−→ N (Ass⊗
act ) −→ D,

which is simply making precise the statement there is a cyclic object in D which looks like
o o
A oo A ⊗ A oo A ⊗ A ⊗ A ooo · · ·.

Finally, assuming that D admits geometric realisations of simplical objects, we may define the Hochschild
homology of the algebra object A to be
1
HH(A/D) := |B•cyc (A)| ∈ DBS

(recall from the end of §4.3 the geometric realisation of a cyclic object gives an object with S 1 -action).
1
Thus HH(−/D) is a (well-behaved) functor from the ∞-category of algebra objects in D to DBS .

Example 4.15. (i) Let k be a base ring and A a flat k-algebra. We leave it to the reader to write
down a precise statement that Definition 4.14 recovers HH(A/k).

32
 AWS 2019 – Topological Hochschild homology

(ii) More generally, if A is not necessarily a flat k-algebra, then it induces an algebra object A in the
symmetric monoidal infinity category D(k) such that HH(A/D(k)) is the derived form of HH(A/k)
introduced in §2.4 (since the monoidal structure in D(k) is given by the derived tensor product
⊗Lk , which we modelled earlier by replacing A by a simplicial resolution P• ).

(iii) Let A be an associate ring spectrum, i.e., an algebra object in Sp. Then its topological Hocshchild
homology is
1
THH(A) := HH(A/ Sp) ∈ SpBS .

We will discuss this further in the next subsection.

4.5 Topological Hochschild homology and its variants

A profound idea of Goodwillie and Waldhausen in the 1980s was to consider the formalism of Hochschild
and cyclic homology not only for rings, but more generally for ring spectra. A well behaved theory of ring
spectra was not available at the time (in particular, there was no strictly commutative and associative
smash product, nor was the language of ∞-categories yet developed in order to overcome the higher
coherence issues), but nevertheless Bökstedt succeeded in giving a rigorous definition of topological
Hochschild homology (for “functors with smash product”) in 1985 [10, 11]. Fortunately, the ∞-category
of spectra is now available to us, and Nikolaus–Scholze have redeveloped the subject from this point of
view. (Here we have omitted the entire history of the subject between 1985 and 2017, with apologies to
all those concerned.)
We begin with an informal discussion of ∞-categories and spectra, for readers who have not previously
encountered them. As is well-known, the triangulated category D(k) of complexes of modules over a
ring k up to quasi-isomorphism suffers from many problems, which can be resolved by keeping track
of which quasi-isomorphisms were used to identify complexes. Of course, as soon as one keeps track of
quasi-isomorphisms, one also needs to keep track of which homopoties between quasi-isomorphisms were
used, and so on. The theory of ∞-categories (modelled as Joyal and Lurie’s “quasi-categories”) does
this: one constructs a simplicial set D(k) whose 0-simplices are chain complexes over k and whose higher
simplices precisely encode the aforementioned data. To understand these notes, the reader should not
run into too much trouble by treating ∞-categories as derived categories in which the usual problems of
functoriality/uniqueness (of cones, homotopy limits, etc.) disappear, or rather become contractible in a
precise sense. A good introduction to the foundation of the theory are the notes by Groth.
Next we turn to spectra. In the course of the historical development of mathematics, each new
discovery of a rings of arithmetic interest was larger than the previous, i.e., Z ⊆ Q ⊆ R ⊆ C, but let
us nevertheless consider the hypothetical situation of only knowing the complex numbers. This suffices
to define complex manifolds, but in order to do Hodge theory we must first construct the integers, real
numbers, and the functors

−⊗Z R −⊗R C
, ,
Z -mod k R -mod l C -mod
restriction restriction

The faithful, non-full restriction functor C -mod → R -mod is particular important, as without it we
would never see complex conjugation; in other words, it is essential to have the freedom to view our
complex vector spaces merely as real vector spaces. Once we have done so, we can say that R → C is a
morphism of algebra objects in R -mod and thereby reconstruct C -mod.
In a similar way, the theory of spectra provides a ring (in a generalised sense) S called the sphere
spectrum and a restriction functor from Z-modules to S-modules which corresponds to a morphism of
algebra objects S → Z. Analogously to the naive Hodge theory example, it is now widely accepted that
it is important to allow ourselves the freedom of viewing objects over Z as merely over S: this allows
for maps between them which are no longer Z-linear, but merely S-linear. Of course none of this is true
in a naive set-theoretic fashion: only the derived category of S-modules exists. So to be more precise
there exist a symmetric monoidal ∞-category Sp of spectra, whose unit object is S and whose monidal

33
 Matthew Morrow

structure is written ⊗ = ⊗S , and functors

−⊗S Z
+
Sp j D(Z)
restriction

corresponding to a morphism of algebra objects S → Z in Sp. The restriction functor is known as

the Eilenberg–Maclane construction, but we prefer not to introduce any notion for it and instead allow
ourselves to view any complex as a spectrum. Any spectrum C has abelian “homotopy groups” πn (C),
for n ∈ Z, which are the usual homology groups Hn (C) in case C comes via restriction from D(Z) (or
from D(k) for some other ring k).
By restriction, any algebra object A in D(Z) (e.g., a usual ring, or something more exotic like a
differential graded ring) gives rise to an algebra object in Sp, i.e., a ring spectrum; we may then form its
topological Hochschild homology as in Example 4.15(iii):
o o
THH(A) := HH(A/ Sp) = | A oo A ⊗S A oo A ⊗S A ⊗S A ooo
1
· · · | ∈ SpBS

(Of course, as in Example 4.15(iii), we could also apply this construction to an arbitrary ring spectrum
which does not necessarily come from D(Z), but we are interested mainly in THH of usual rings.)
The homotopy groups of this spectrum are denoted by THHn (A) := πn THH(A) and are known as the
topological Hochschild homology groups of A.
Remark 4.16 (Low degrees, rationalising, and stable homotopy groups of spheres). The homotopy
groups of the sphere spectrum S are given by

0
 n<0
πn (S) = Z n=0.

 th
n stable homotopy group of spheres n > 0

Two concrete consequences for the canonical map THH(A) → HH(A/Z) as are follows.
'
(i) It is an isomorphism in degrees ≤ 2, i.e., THHn (A) → HHn (A/Z) for n ≥ 2 (also both sides vanish
for n < 0)
(ii) It is an isomorphism rationally (since the stable homotopy groups of spheres are known to be
finite), or more precisely the map THHn (A) → HHn (A/Z) has kernel and cokernel killed by some
integer depending only on n. In particular, if A ⊇ Q (whence the two sides are Q-vector spaces), it
is an isomorphism: topological Hochschild homology offers us nothing new in characteristic zero!
1
Let D = D(k) or Sp. Given any object with S 1 -action X ∈ DBS , we may form its group cohomology,
i.e., derived/homotopy S 1 -invariants
1
X hS = limBS 1 X ∈ D

(where, on the right side, we recall that X is a functor BS 1 → D). Similarly we may form its group
homology, i.e., derived/homotopy S 1 -coinvariants

XhS 1 = colimBS 1 X ∈ D.
1
These are respectively the right and left adjoints to the functor D → DBS which equips an object of D
1
with the trivial action. There exists moreover a certain “norm” morphism N : XhS 1 [1] → X hS , whose
1
cofiber is known as the Tate cohomology of the action and denoted by X tS , i.e.,
N 1 1
XhS 1 [1] −→ X hS −→ X tS .

Example 4.17. Suppose that the object with S 1 -action in the preceding discussion is the Hochschild
1
homology HH(A/k) ∈ D(k)BS of some k-algebra A. In §4.1 we informally explained that HC(A/k),

34
 AWS 2019 – Topological Hochschild homology

HC − (A/k), and HP (A/k) may be considered the homology, cohomology, and Tate cohomology of the
S 1 -action. To be precise, one does indeed have natural equivalences in D(k)
HC(A/k) ' HH(A/k)hS1 , HC − (A/k) ' HH(A/k)hS1 , HP (A/k) ' HH(A/k)tS1 .
These folklore identifications are obtained by picking particular models to compute the homotopy invari-
ants and coinvariants, then comparing to the explicit double complexes of §2.2 (similarly to what we did
at the end of §4.1); see [22].
1 1
Furthermore, if we rearrange the above norm sequence into a fibre sequence X hS → X tS → XhS 1 [2],
then the case X = HH(A/k) recovers the sequence of Remark 2.15.
Definition 4.18. Transporting the previous definition to the world of spectra, we define the topological
negative cyclic homology and topological periodic cyclic homology of a ring A to be
1 1
TC − (A) := THH(A)hS ∈ Sp, TP (A) := THH(A)tS ∈ Sp .
We will also be interested in THH(A)hS 1 , which is an analogue of cyclic homology, but we stress that it
is not what is known as topological cyclic homology. The homotopy groups of these spectra are denoted
by TCn− (A), TPn (A), πn THH(A)hS 1 .
In the setting of the previous definition, the general norm sequence looks like
N
THH(A)hS 1 [1] −→ TC − (A) −→ TP (A).
Since THH(A)hS 1 is supported in homotopical degree ≥ 0 (as this is true of A and preserved by colimits),
'
we see that TCn− (A) → TPn (A) for n ≤ 0.
Remark 4.19 (Multiplicative structure). Assuming that A is a commutative ring, which is once again
our only case of interest, then THH(A) is itself a ring spectrum. Moreover, taking the Tate construction
or homotopy fixed points are known to be lax monoidal functors, so that TP (A) and TC − (A) are ring
spectra. Their homotopy groups
M M M
THHn (A), TPn (A), TCn− (A)
n≥0 n∈Z n∈Z

therefore naturally have the structure of graded commutative rings.

5 Topological Hochschild homology of Fp -algebras

In this section we restrict our study of topological Hochschild homology to the case of Fp -algebras. We
begin in §5.1 by studying Fp itself and deriving some analogues for general Fp -algebras A; in particular,
we will see that TP (A) is a 2-periodic lifting of the classical periodic cyclic homology HP (A/Fp ) from
characteristic p to mixed characteristic, and establish an analogue of the HKR filtration of Proposition
2.28. We remark that these results do not have analogues for the topological Hochschild homology of
arbitrary rings (though they do if we replace our base ring Fp by a perfectoid ring).
Then in §5.2 we analyse topological periodic and negative cyclic homologies of smooth algebras by
reduction to qrsp algebras, analogously to §3.

5.1 The case of Fp itself and consequences

The following fundamental highly non-trivial result is sadly beyond the techniques of this course:
Theorem 5.1 (Bökstedt). The homotopy groups of THH(Fp ) are given by
(
Fp n even ≥ 0
THHn (Fp ) ∼
=
0 n otherwise

Regarding multiplicative structure THH∗ (Fp ) = Fp [u] (polynomial algebra on a single variable u) with
u ∈ THH2 (Fp /Zp ) = HH2 (Fp /Z) (same element as in Example 2.32).

35
 Matthew Morrow

Although we have omitted the proof of Bökstedt’s theorem, we will be able to use it as a blackbox to
obtain interesting consequences, starting with a calculation of the homotopy groups of the topological
periodic and negative cyclic homology of Fp :
Proposition 5.2. The homotopy groups of TP (Fp ) are given by
(
∼ Zp n even
TPn (Fp ) =
0 n odd

Regarding multiplicative structure TP∗ (Fp ) = Zp [σ ±1 ], where σ is any generator of the invertible Zp -
module TP2 (Fp ). The map of graded rings TC∗− (Fp ) → TP∗ (Fp ) is injective, with image
(
− pn TP2n (Fp ) n ≥ 0
Im(TC2n (Fp ) → TP2n (Fp )) = ,
TP2n (Fp ) n≤0

i.e., TC∗− (Fp ) = Zp [pσ, σ −1 ], and the canonical morphism TC − (Fp ) → THH(Fp ) sends pσ to u (equiva-
'
lently, without making any choices, TC∗− (Fp )/p → THH∗ (Fp ) for ∗ ≥ 0).
Remark 5.3. It may be helpful to compare to Lemma 3.8 and Remark 3.10, where HP0 inherited an
−
interesting filtration coming from HC2n , n ≥ 0. Here we witness the same phenomenon: the homotopy
−
groups of TC (Fp ) is inducing the p-adic filtration on the homotopy groups of TP (Fp ).
Proof of Proposition 5.2. Recall that there is a class u ∈ THH2 (Fp ) which corresponds to p ∈ pZ/p2 Z
'
under the identifications THH2 (Fp ) → HH2 (Fp /Z) = pZ/p2 Z. We will also use the cohomology class
p
t ∈ H 2 (S 1 , Fp ) which corresponds to the extension 0 → Fp → S 1 −→ S 1 → 0; a standard fact about
group cohomology states that H ∗ (S 1 , Fp ) ∼
= Fp [t].
Using Bökstedt’s theorem and the previous paragraph, the homotopy-fixed points spectral sequence
−
for the action of S 1 on THH(Fp ) looks as follows and converges to TCi+j (Fp ):

0 0 0 0 0

H 4 (S 1 , THH2 (Fp )) = Fp t2 u 0 H 2 (S 1 , THH2 (Fp )) = Fp tu 0 H 0 (S 1 , THH2 (Fp )) = Fp u

0 0 0 0 0

H 4 (S 1 , THH0 (Fp )) = Fp t2 0 H 2 (S 1 , THH0 (Fp )) = Fp t 0 H 0 (S 1 , THH0 (Fp )) = Fp

(the bottom right corner is in bidegree (0, 0)). Since all non-zero terms are in even bidegree, all differen-
tials on all pages are necessarily zero, i.e., the spectral sequence is degenerate.
The key (in fact, only) non-trivial calculation is to check the following: the usual spectral sequence
map
edge
Fil2abutement TC0− (F0 ) = Ker TC0− (Fp ) −−−→ H 0 (S 1 , THH0 (Fp )) −→ H 2 (S 1 , THH2 (Fp ))

sends p to tu. This is proved as follows: the part of the spectral sequence which we have displayed
(i.e., involving THHn (Fp ) for n ≤ 2) is the same as the analogous spectral sequence for the Hochschild
homology HH(Fp /Z), and therefore the calculation is a claim purely about usual Hochschild and negative
cyclic homology. One resolves Fp by a flat dg Z-algebra and proceeds by a direct calculation, for which
we refer to [28, Prop. 2.12].

36
 AWS 2019 – Topological Hochschild homology

−
By multiplicativity of the spectral sequence it follows that pi ∈ Fil2i abut TC0 (Fp ) for all i ≥ 0, and
2i − 2i 1 i i i
that the canonical map Filabut TC0 (Fp ) → H (S , THH2i (Fp )) sends p to t u . That is, the unique
'
ring homomorphism Z → TC0− (Fp ) induces isomorphism pi Z/pi+1 Z → gr2i −
abut TC0 (Fp ) for all i ≥ 0. It
' −
follows that Zp → TC0 (Fp ).
Pick lifts u ∈ TC2− (Fp ) and t ∈ TC−2 −
(Fp ), along the edge maps, of u and t respectively (it would be
more correct to write u e and e t for the lifts, but we follow a standard abuse of notation). The multiplica-
− −
tivity of the spectral sequence forces TC2i (Fp ) = Zp ui for i ≥ 0, and TC2i (Fp ) = Zp ti for i ≤ 0, with the
−
relation ut = p · unit in TC0 (Fp ) = Zp .
Remark 5.4. Theorem 5.1 and Proposition 5.2 remain true if Fp is replaced by an arbitrary perfect
Fp -algebra k and Zp is replaced by the ring of p-typical Witt vectors W (k).
As we promised earlier, from the proposition one obtains some interesting general consequences which
will allow us to control the topological theory in characteristic p. The first is that topological periodic
cyclic homology of Fp -algebras is always 2-periodic; despite its name, this is not always the case, notably
TP (Z). The second is that it lifts classical periodic cyclic homology; since we already know that the
latter is related to de Rham cohomology, this indicates that the former might be related to the standard
lift of de Rham cohomology, namely crystalline cohomology (as we shall see is indeed true).
Corollary 5.5. Let A be a commutative Fp -algebra. Then:
u
(i) There is a natural, S 1 -equivariant fibre sequence THH(A)[2] −
→ THH(A) → HH(A/Fp ).
(This will let us control THH in terms of HH and so serve as a replacement for the classical
periodicity sequence.)
'
(ii) TP (A) is 2-periodic, i.e., the TP0 (A)-module TP2 (A) is invertible and TPn (A) ⊗TP0 (A) TP2 (A) →
TPn+2 (A) for all n ∈ Z.
(iii) There is a natural equivalence TP (A)/p → HP (A/Fp ).
Proof. We need the following general result: given a map of commutative rings k → A, then in the world
of E∞ -ring spectra one has THH(A) ⊗THH(k) k ' HH(A/k). This is proved either by checking that both
sides satisfy an identical universal property similar to that of Example 2.18, or by transferring the proof
of Remark 2.7 to the E∞ context.
To prove (i), it is therefore enough to construct an S 1 -equivariant fibre sequence

THH(Fp )[2] → THH(Fp ) → Fp ,

since we can then base change by THH(A)⊗THH(Fp ) −. The class u := pσ ∈ TC2− (Fp ) = Homh(Sp) (S[2], TC − (Fp ))
corresponds to a morphism u : S[2] → TC − (Fp ) uniquely up to homotopy; composing with TC − (Fp ) →
THH(Fp ) gives us a S 1 -equivariant map S[2] → THH(Fp ), which then linearises to a morphism u :
THH − (Fp )[2] → THH − (Fp ) whose effect on homotopy groups is multiplication by x thanks to the final
sentence of Proposition 5.2. We can also read off that proposition that the cofiber of ×u is Fp [0], as
desired.
(ii) follows formally from the fact that TP∗ (A) is a graded module over TP∗ (Fp ) ∼
= Zp [σ ±1 ].
(iii): Taking Tate constructions in the fibre sequence of (i) gives
u
TP (A)[2] −
→ TP (A) → HP (A/Fp )

But by construction u is given by

TP (A)[2]
u / TP (A)
KKK u:
KKK∼ u uu
=
K u
σ KKK uu p
% uu
TP (A)

37
 Matthew Morrow

Suppose now that R is a smooth algebra over a perfect field k of characteristic p (in fact, we could
suppose that k is any perfect Fp -algebra without any change to the argument). Then we have the anti-
symmstrisation map Ω1R/Fp = Ω1R/Z → HH1 (R/Z) = THH1 (R) which we know induces by multiplication
Ω2R/Fp → HH2 (R/Z) = THH2 (R); since THH∗ (R) is a commutative graded ring, multiplicativity formally
implies that the antisymmstrisation map extends to all degrees Ω∗R/k → THH∗ (R). Combining this with
the map of graded rings THH∗ (Fp ) = Fp [u] → THH∗ (R), we arrive at Ω∗R/k ⊗Fp THH∗ (Fp ) → THH∗ (R),
where ⊗ denotes tensor product of graded rings. One has the following analogue of the classical HKR
theorem (it may initially appear strange that THH is built from multiple copies of de Rham groups, but
when passing to TP these will perfectly stack up on top of each other to form cristalline cohomology):

Proposition 5.6 (Hesselholt’s HKR theorem). Let R be a smooth k-algebra as in the previous paragraph.
Then the map of graded rings Ω∗R/k ⊗Fp THH∗ (Fp ) → THH∗ (R) is an isomorphism. In particular,
THHn (R) is isomorphic as an R-module to i≥0 Ωn−2i
L
R/k .

Proof. By construction the composition

Ω∗R/k ⊆ Ω∗R/k ⊗Fp THH∗ (Fp ) → THH∗ (R) → HH∗ (R)

is the usual antisymmetrisation map, which is an isomorphism by the usual HKR theorem. So THH∗ (A) →
HH∗ (A/Fp ) is surjective on homotopy groups and the fibre sequence of Corollary 5.5(i) breaks into short
'
exact sequence, whence we have an isomorphism of graded rings THH∗ (A)/u → HH∗ (A/Fp ).
In other words, the map of graded Fp [x]-modules

Ω∗R/k ⊗Fp THH∗ (Fp ) = Ω∗R/k ⊗Fp Fp [x] −→ THH∗ (R)

is an isomorphism modulo x. It follows formally (by induction up the degree) that it is an isomorphism.

Hesselholt’s HKR theorem leads to an analogue of the HKR filtration from Proposition 2.28, which
gives us a technique to control THH via the cotangent complex.

Corollary 5.7 (HKR filtration on THH). Let A be an Fp -algebra. Then THH(A) admits a descending,
Lbn/2c
N-indexed, complete filtration whose nth -graded piece is i=0 Ln−2i
A/Fp [n].

Proof. Just left Kan extend the previous proposition; see [8, Corol. 6.10] for some details.

In order to extend the techniques of §3.1–3.2 to topological Hochschild homology and its variants,
we of course need to know that they satisfy flat descent; although there is a technique to deduce this in
general from the case of HH (see [8, Corol. 3.3], the following special case is enough for us:

Corollary 5.8. The Sp-valued functors

THH(−), THH(−)hS 1 , TC − (−), TP (−)

on Fp -algs satisfy flat descent.

Proof. The argument is similar to Lemma 3.7. The previous corollary and flat descent for all the Li−/Fp
yield flat descent for THH(−). The other cases then follow by suitably taking limits and colimits, taking
care of the connectivity issues which arise for THH(−)hS 1 .

5.2 Topological periodic & negative cyclic homology of smooth and qrsp
algebras
The main theorem we wish to discuss is the following analogue of Theorem 3.1 in which HP is replaced
by TP and de Rham cohomology is replaced by crystalline cohomology:

38
 AWS 2019 – Topological Hochschild homology

Theorem 5.9. If R is a smooth k-algebra, where k is a perfect field of characteristic p > 0, then TP (R)
and TC − (R) admit natural, complete, descending Z-indexed filtrations whose nth graded pieces are given
respectively by
RΓcrys (R/W (k))[2n], N ≥i RΓcrys (R/W (k))[2n],
where N ≥i refers to a certain filtration on crystalline cohomology known as the Nygaard filtration.
The filtrations of the theorem are precisely those arising from the general technique presented in
Corollary 3.6, which is valid since we verified flat descent for TP and TC − in Corollary 5.8. Therefore,
following the same method of proof of §3.1–3.2, we must attempt to identify TP and TC − of the qrsp
rings Rperf ⊗R · · · ⊗R Rperf . Let we begin with some coarse information analogous to Lemma 3.8
Lemma 5.10. Let A be a qrsp algebra and set N := π1 (LA/Fp ). Then THH∗ (A), TC∗− (A), TP∗ (A),
and π∗ THH(A)hS 1 are all supported in even degrees. Moreover, THH2i (A) admits a finite increasing
filtration with graded pieces A, N, Γ2A (N ), . . . , ΓiA (N ), and TP0 (A) admits a complete, descending, N-
indexed filtration by ideals such that
Fili ∼ −
= TC2i (A), TP0 (A)/ Fili ∼
= π2i−2 THH(A)hS 1 , gri ∼
= THH2i (A) ∼
= πi (LiA/Fp )
for all i ≥ 0.
Proof. The proof is entirely analogous to that of Lemma 3.8; firstly, as already argued there, each wedge
powers LiA/Fp = ΓiA (N )[i] is supported in degree i. The HKR filtration of Corollary 5.8 therefore proves
that THH∗ (A) is supported in even degree and gives the desired finite increasing filtration (we leave it
to the reader to carefully check that the degrees are ok).
The rest of the structure can be formally read off Corollary 5.5 and the norm sequence THH(A)hS 1 →
TC − (A) → T P (A), in a very similar way to the arguments of Lemma 3.8.
Continuing to following the same line of argument we present in §3.2, we would now like to identify
explicitly the filtered ring TP0 (A). In the case of classical periodic cyclic homology we could do this in
two ways, either in terms of a divided power envelope or in terms of derived de Rham cohomology. Now
we lift these two constructions from characteristic p to mixed characteristic.

5.2.1 First construction: derived de Rham(–Witt) and crystalline cohomology

Let R be a smooth algebra over a perfect field k of characteristic p, and let R
e be a p-adically complete,
•
formally smooth W (k)-algebra lifting R (i.e., R/p = R). Let ΩR/W
e
e (k)
be the p-adically complete de
Rham complex (we should really add a hat to indicate the p-adic completion, but we prefer to keep
notation light and will never consider the non-complete version).
Assuming that Re is equipped with a lift of the absolute Frobenius (i.e., there exists a ring homomor-
phism ϕ e : R → R compatible with the usual Frobenius on W (k) and such that ϕ(f
e e e ) ≡ f p mod pR e for
•
all f ∈ R),
e then we define a filtration on Ω
R/W
e (k)
by setting, for i ≥ 0,

d d d d d d
pmax(i−•,0) Ω•R/W
e (k)
:= pi R
e−→ pi−1 Ω1R/W
e (k)
− → pΩi−1
→ ··· − R/W
e (k)
→ ΩiR/W
− e (k)
→ Ωi+1
− R/W
e (k)
−
→ ···

(although the definition of this filtration does not depend on ϕ,

e it is only reasonable to define it when ϕ
e
exists; this is partly because of the next lemma).
Lemma 5.11. The graded pieces of the above filtration on Ω•R/W
e (k)
are given (up to quasi-isomorphism)
by
∼
e i : griN RΓcrys (R/W (k)) → τ ≤i ΩR/k
ϕ/p
(where τ ≤i denotes canonical – not naive – truncation).
e induces an endomorphism of ΩnR/W
Proof. Note that ϕ e (k)
which is divisible by pn , so there is indeed an
0 0 0
→ Ω1R/k −
induced map as indicated. The graded pieces of the filtration are given by R − → ··· −→ ΩiR/k and
the induced map on cohomology looks like ΩnR/k → HdR n
(R/k) (for n ≤ i); one checks that it is precisely
the Cartier isomorphism.

39
 Matthew Morrow

As an object of the derived category D(W (k)), the complex Ω•R/W e (k)
depends only on R. Indeed,
the theory of crystalline cohomology implies that Ω•R/W
e (k)
is equivalent to the crystalline cohomology
RΓcrys (R/W (k)), or equivalently to the de Rham–Witt complex W Ω•R/k . This independence remains
true for the filtration we have just defined:

Lemma 5.12. The above filtration on Ω•R/W

e (k)
also depends only on R. (More precisely Ω•R/W
e (k)
, as
an object of the filtered derived category over W (k), depends naturally on R.)

Proof. There are two ways to check this, either via crystalline cohomology or via de Rham–Witt theory.
For the crystalline approach one uses a result of Berthelot–Ogus, stating that the Frobenius ϕ induces
∼
an equivalence ϕ : RΓcrys (R/W (k)) → Lηp RΓcrys (R/W (k)), where Lηp refers to the décalage functor
of [8]. There is a natural filtration on the right side arising from the décalage functor (see ...), which
therefore induces a filtration on the left side. Under the equivalence Ω•R/W
e (k)
' Γcrys (R/W (k)), this is
•
precisely the above filtration on ΩR/W
e (k)
.
For the de Rham–Witt approach, recall that Ω•R/We (k)
identifies up to equivalence with W Ω•R/k ; then
the above filtration can be shown to identify with the so-called Nygaard filtration on W Ω•R/k given by

d d d d d d d
N ≥i W Ω•R/k := pi−1 V W (R) −
→ pi−2 V W Ω1R − → pV W Ωi−2
→ ··· − R − → V W Ωi−1
R − → W Ωi+1
→ W ΩiR − R − → ··· .

In conclusion, we have naturally associated to each smooth k-algebra R a filtered complex RΓcrys (R/W (k))
(even an E∞ -W (k)-algebra). If we left Kan extend and complete with respect to the resulting filtra-
tion then the result, in the case of a qrsp algebra A, is a complete filtered ring LW \ ΩA/Fp such that
\
LW ΩA/F /p = LΩ
p
c A/F . More details to be added.
p

5.2.2 Second construction: a divided power envelope

Let A be qrsp. We denote by A◦crys (A) the divided power envelope of the composition W (A[ )  A[  A,
and by Acrys (A) its p-adic completion. For example, Acrys (OQp /p) is the period ring Acrys of p-adic Hodge
theory.
The usual Witt vector Frobenius ϕ induces a Frobenius endomorphism ϕ of Acrys (A), using which
we define its Nygaard filtration as follows:

N ≥i Acrys (A) := {f ∈ Acrys (A) : ϕ(f ) ∈ pi Acrys (A)}.

Finally, write A
b crys (A) := lim A (A)/N ≥i Acrys (A) for the completion with respect to the Nygaard
←−i→∞ crys
filtration, and A b crys (A)  Acrys (A)/N ≥i Acrys (A)) for the induced filtration.
b crys (A) := Ker(A
The advantage of Acrys (A) is that it is reasonably explicit.
b

5.2.3 Back to TP
The following is the analogue for the topological theories of Proposition 3.13:

Proposition 5.13. Let A be a qrsp algebra. Then there are natural isomorphisms of filtered rings

TP0 (A) ∼
=Ab crys (A) ∼ \
= LW ΩA/Fp

which modulo p recover the filtered isomorphisms of Corollary 3.13.

Proof. The proof of this is relatively self-contained so we refer to [8, Thm. 8.15].

The proposition implies Theorem 5.9, similarly to the second proof of Theorem 3.1.

40
 AWS 2019 – Topological Hochschild homology

6 (Topological) Hochschild and cyclic homologies of

Zp -algebras – a glimpse
It is beyond the scope of this course to discuss in detail the topological Hochschild and cyclic homologies
of p-adic algebras, but we nevertheless indicate some of the main results without proof. The goal is
essentially to overview the analogous results of §3 and §5 in the case of p-adic algebras.

Remark 6.1 (p-completeness convention). When the input ring A is p-adically complete (but not killed
by a power of p), then HH(A/Zp ), THH(A), etc., as we have already defined them, will contain large
amounts of undesirable, junk data. (For example, if S is any perfect Fp -algebra, then we would like
HH(W (S)/Zp ) to be supported in degree 0, since LW (S)/Fp should vanish; but these statements are only
true after p-completing the complexes.) In this section we therefore adopt the following convention: if A
is a p-adically complete ring, then we write HH(A/Zp ), THH(A), etc. to mean HH(A/Zp )bp , THH(A)bp ,
where the hat denotes derived p-adic completion.

6.1 The case of perfectoid rings and consequences

We begin with the analogue of Bökstedt’s calculation (Theorem 5.1) and the consequences for topological
periodic and topological negative cyclic homologies (analogue of Proposition 5.2). Although we stated
these only for Fp we remarked that they held for arbitrary perfect Fp -algebras; we therefore begin by
recalling the p-adic analogue of a perfect Fp -algebra:

Definition 6.2. A ring A is perfectoid if it satisfies the following conditions:

(i) it is p-adically complete and separated;

(ii) A/pA is semiperfect;

(iii) there exist π ∈ A and u ∈ A× such that π p = pu;

(iv) the kernel of Fontaine’s map θA : Ainf (A) → A is principal.

It is not our intention here to review the theory of perfectoid rings, in particular do not discuss Fontaine’s
ring Ainf (A) := W (A[ ) or the associated map θA : Ainf (A) → A. We refer to [7] or to [Lecture IV of
Bhatt’s notes on prisms, available on his webpage].

Example 6.3. (i) An Fp -algebra is perfect if and only if it is perfectoid.

(ii) Let C be a perfectoid field (e.g., an algebraically closed field which is complete under a rank one
valuation) of mixed characteristic containing all p-power roots of unity, and let OC ⊆ C denote its
ring of integers. Then OC is a perfectoid ring, which often serves as the base ring in the theory.

Theorem 6.4. Let A be a perfectoid ring. Then the homotopy groups of THH(A) are given by
(
A n even ≥ 0
THHn (A) ∼
=
0 n otherwise

Regarding multiplicative structure, THH∗ (A) = A[u] (polynomial on a single variable u), with u any
generator of the invertible A-module THHn (A) = HH2 (A/Zp ) = Ker θA /(Ker θA )2 .

Sketch of proof. This is the analogue of Bökstedt’s theorem. It is proved by performing various base
changes to capture all the data either from characteristic p (i.e., Bökstedt) or rationally (i.e., usual
Hochschild homology). See [8, Thm. 6.1].

Let ξ ∈ Ainf (A) be a generator of Ker θA .

41
 Matthew Morrow

Theorem 6.5. The homotopy groups of TP (A) are given by

(
∼ Ainf (A) n even ≥ 0
TPn (A) =
0 n otherwise

Regarding multiplicative structure, TP∗ (A) = Ainf [σ ±1 ], where σ is any generator of the invertible
Ainf (A)-module TP2 (A). The map of graded rings TC∗− (A) → TP∗ (A) is injective, with image
(
− ξ n TP2n (A) n ≥ 0
Im(TC2n (A) → TP2n (A) = ,
TP2n (A) n≤0

i.e., TC∗− (A) = Ainf (A)[ξσ, σ −1 ], and the canonical morphism TC − (A) → THH(A) sends ξσ to u (equiv-
'
alently, without making any choices, TC∗− (A)/ξ → THH∗ (A) for ∗ ≥ 0).
Proof. This is proved similarly to Proposition 5.2, noting that the universal property of Fontaine’s ring
Ainf (A) provides a map Ainf (A) → TP0 (A) which serves as a replacement for the map Zp → TP0 (Fp )
which we used in the earlier proof.
The previous theorems lead to analogues of results 5.5–5.8; see [8, §6.3].

6.2 Quasiregular semiperfectoids

The mixed characterisation of a quasiregular semiperfect Fp -algebra is as follows:
Definition 6.6. A ring A is called quasiregular semiperfectoid if it satisfies the following conditions:
(i) it is p-adically complete and A/pA is semiperfect;
(ii) it has bounded p∞ -torsion;
(iii) there exists a perfectoid ring S and a map S → A such that the p-completion of LA/S is supported
in homological degree 1 where it is given by a module N := H1 (L b A/S ) such that N/pr is a flat
r
A/p -module for all r ≥ 0.
When A is qrsp in this sense, we remark that the condition on the cotangent complex is then satisfied
for any map from a perfectoid ring to A (which may in fact be chosen to be surjective).
Example 6.7. (i) Any quasiregular semiperfect Fp -algebra is of course quasiregular semiperfectoid.
Nevertheless, we obtain some new presentations: for example, the quasiregular semiperfect Fp -
[
algebra OC /p[ is the same as OC /p.
∞
(ii) The prototypical example of a quasiregular semiperfectoid not of characteristic p is OC hT ±1/p i/(T − 1).
Arguing as in Lemmas 3.8 and 5.10, one again sees that TP0 of a quasiregular semiperfectoid ring
admits a filtration coming from negative cyclic homology:
Lemma 6.8. Let A be a quasiregular semiperfectoid ring, fix an integral perfectoid ring S and map
S → A, and set N := π1 (L b A/S ). Then THH∗ (A), TC − (A), TP∗ (A), and π∗ THH(A)hS 1 are all supported
∗
in even degrees. Moreover, THH2i (A) admits a finite increasing filtration with graded pieces given by
the p-completions of A, N, Γ2A (N ), . . . , ΓiA (N ), and TP0 (A/Fp ) admits a complete, descending, N-indexed
filtration by ideals such that
Fili ∼ −
= TC2i (A), TP0 (A)/ Fili ∼
= π2i−2 THH(A)hS 1 , gri ∼
= THH2i (A) ∼ bi )
= πi (LA/S

for all i ≥ 0.
In the case of classical periodic cyclic homology we saw two descriptions of HP0 of qusasiregular
semiperfect rings: either in terms of derived de Rham cohomology or an explicit divided power envelope.
We had two similar descriptions of TP0 , either in terms of derived crystalline cohomology or again via
a divided power envelope construction. We would like a similar description of TP0 of any quasiregular
semiperfectoid:

42
 AWS 2019 – Topological Hochschild homology

Theorem 6.9. Let A be a qrsp OC -algebra. Then there are natural isomorphisms of filtered rings

TP0 (A) ∼
=∆b A/A ∼
inf
= AΩ
dA

(The middle and right filtered rings are defined below.)

Proof. This is proved by a rather intricate set of arguments bouncing back and forth between qrsp and
formally smooth algebras, and also requires Bhatt–Scholze’s forthcoming work on prismatic cohomology.
We aren’t going to say anything about it.

6.2.1 Derived AΩ cohomology

Let R be the p-adic completion of a smooth OC -algebra; assume that R is small in the sense that
there exists a formally étale morphism OhT1±1 , . . . , Td±1 i → R for some d ≥ 0 (but do not fix any such
morphism). In [7] the authors introduced an E∞ -Ainf -algebra AΩR ∈ D(Ainf ) which may be described
in any of the following ways:

(i) (Faltings-style Galois cohomology) Assuming that Spf R is connected, let R[ p1 ] be a “universal
cover” of R, i.e., the filtered union of all finite étale extensions of R[ p1 ] inside a fixed algebraic
closure of Frac R. In particular, the geometric Galois group ∆ := π1ét (Spec R) acts naturally on
R[ p1 ]. Next let R be the p-adic completion of the integral closure of R inside R[ p1 ] and observe that
R inherits a continuous action by ∆. It turns out (here the smallness assumption is used) that R is
integral perfectoid, so we may form its associated Fontaine ring Ainf (R), which is again equipped
with a continuous action by ∆, and the resulting Galois cohomology RΓcont (∆, Ainf (R)). The first
definition of AΩR is
AΩR := Lηµ RΓcont (∆, Ainf (R)),
where Lηµ is the décalge functor from [7].
(ii) (q-de Rham complex) The previous definition of AΩR is a priori uncomputable since the geo-
metric Galois group ∆ is inaccessible. However, the almost purity theorem and a curious prop-
erty of the décalage functor overcome this difficulty as follows. Fix a formally étale morphism
∞
OhT1±1 , . . . , Td±1 i → R and define R∞ to be the p-adic completion of R ⊗OhT ±1 i OhT ±1/p i. Then
R∞ ⊆ R, and the ∆-action on R restrictions to R∞ : in fact, on R∞ the action factors through
∆ → Γ := Zp (1)d , where Zp (1)d acts on R∞ by rescaling the p-power roots of the variables by
suitable p-power roots of unity. A consequence of the almost purity theorem for perfectoid rings is
that the induced morphism RΓcont (Γ, Ainf (R∞ )) → RΓcont (∆, Ainf (R)) is in fact an almost equiv-
alence; moreover, the almost difference remarkably disappears after applying the décalage functor,
i.e.,
∼
Lηµ RΓcont (Γ, Ainf (R∞ )) → Lηµ RΓcont (∆, Ainf (R)) = AΩR
But the left side is defined in terms of an explicit action of a pro-cyclic group and can be computed
in terms of Koszul complexes; it turns out to be a “q-de Rham complex”, for which we refer to [7]
for further details.
(iii) (pro-étale cohomology) The problem with both of the previous definitions is that they do not
enjoy strong functoriality properties: the q-de Rham description is completely non-functorial as
it depends on the chosen coordinates, while the Galois cohomology descriptions depends at least
on the choice of an algebraic closure of Frac R. The solution to this problem adopted in [7] is
to replace the Galois cohomology RΓcont (∆, Ainf (R)) by the site-theoretic pro-étale cohomology
RΓproét (Spa(R[ p1 ], R), Ainf ) of a period sheaf Ainf on the adic spectrum Spa(R[ p1 ], R). Similarly
to (ii), these two cohomologies are almost equivalent and become equivalent after décalge, i.e.,
AΩR ' RΓproét (Spa(R[ p1 ], R), Ainf ). This yields a definition of AΩR which is functorial and can be
easily extended to arbitrary smooth, formal O-schemes.
Definition 6.10. For an arbitrary p-complete OC -algebra A, let AΩA denote the (p, ξ)-adic completion
of the left Kan extension of AΩ− from p-adic completions of smooth O-algebras.

43
 Matthew Morrow

When R is a formally smooth OC -algebra, then AΩR is moreover equipped with a complete filtration
whose graded pieces are given by truncations of the de Rham complex of R. By left Kan extending this
filtration and mimicking the arguments of §5.2.1, one checks that if A is a qrsp OC -algebra then AΩA is
a supported in degree 0, where it is given by a complete filtered ring whose graded pieces are related to
derived de Rham cohomology; in that case we write AΩ dA for the completion of A with respect to this
filtration.

6.2.2 Prisms and δ-rings (following Bhatt–Scholze)

Let A be a qrsp OC -algebra. The new theory of prismatic cohomology, in which (very roughly) divided
powers are replaced by p-derivations, associates to A its prismatic cohomology ∆ A/Ainf . Since A is qrsp
this cohomology is supported in degree zero, where it is given by a sort of period ring constructed in
terms of p-derivations. We now explain this construction.
Fix an integral perfectoid OC -algebra S equipped with a surjection S → A. Let Ainf (S)[ 1ξ ]b be the
p-adic completion of Ainf (S)[ 1ξ ] and observe that the usual Witt vector Frobenius ϕ on Ainf (S) = W (S [ )
extends uniquely to a Frobenius lift on this completion (indeed, one just needs to check that ϕ(ξ) is a unit
mod p in Ainf (S)[ 1ξ ], but this is clear since ϕ(ξ) ≡ ξ p mod p). We denote by δ : Ainf (S)[ 1ξ ]b → Ainf (S)[ 1ξ ]b
the associated p-derivation, i.e., δ(f ) := p1 (ϕ(f ) − f p ).
Let ∆ A/Ainf be the (p, ξ)-adic completion of the Ainf (S)-subalgebra of Ainf (S)[ 1ξ ]b generated by terms
θ
of the form δ n ( fξ ) for all f ∈ Ker(Ainf (S) −
→ S → A) and n ≥ 0. We equip ∆ A/Ainf with a Nygaard-style
≥i
filtration N ∆ A/Ainf := {g ∈ ∆ A/Ainf : ϕ(g) ∈ ξ i∆ A/S } for i ≥ 0, and let ∆
b A/A be its completion for
inf
this filtration.

6.3 Smooth algebras

It remains only to say something about the topological cyclic homologies of a smooth OC -algebra; for
these we have the following analogue of Theorems 3.1 and 5.9:
Theorem 6.11. Let R be a p-adic formally smooth OC -algebra. Then TP (R) and TC − (R) admit natural,
complete, descending Z-indexed filtrations whose nth graded pieces are given respectively by

N ≥n AΩR , [2n] AΩR [2n],

where we use the AΩR complex and its filtration from §6.2.1.
Sketch of proof. Similar to the proof of the aforementioned theorems: in a p-adic version of the quasisyn-
tomic topology, R may be covered by quasiregular semiperfectoid OC -algebras. On the TP and TC − of
these latter algebras we impose the two-speed Postnikov filtration, whose graded pieces are calculated
by Theorem 6.9. Descending back down to TP (R) and TC − (R) then implies the theorem.

44
AWS 2019 – Topological Hochschild homology

Projects

45
 Matthew Morrow

A Motivic cohomology with modulus in characteristic p

(Themes: derived de Rham–Witt complexes)

A.1 Motivation: classical motivic cohomology

To any finite type scheme X over a field k, Bloch associated his cycle complex

z n (X) = [z n (X, 0) ← z n (X, 1) ← · · · ]

This is constructed more naturally as a simplicial abelian group, in which z n (X, i) consists of formal sums
of irreducible, codimension-n subvarieties of X ×k ∆ik (where ∆ik is the spectrum of k[t1 , . . . , ti ]/(t1 +
· · · + ti − 1)) which intersect all faces correctly. The homology of this chain complex is Bloch’s higher
Chow groups CH n (X, i) := Hi (z n (X)); up to reindexing these are the same as Voevodsky’s motivic
i
cohomology HM (X, Z(q)) := CH q (X, 2q − p) = H p (z q (X)[−2q]).
We recall also that each presheaf X ⊇ U 7→ z n (U, i) turns out to be a Zariski sheaf, and that this
complex of Zariski sheaves U 7→ z n (U ) even satisfies Zariski descent; therefore the above definitions can
be rephrased in terms of the hypercohomology of this complex of sheaves. In particular, introducing
the complex of Zariski sheaves Z(i)X : U 7→ z n (U )[−2n], we have HM i
(X, Z(q)) = HiZar (Z, Z(q)X ). (In
an attempt to avoid mistakes when shifting and passing between (c)ohomological indexing, it might
be helpful to note the following: Z(q)X cohomologically vanishes in degrees > 2q, since each z q (U ) is
homologically supported in degrees ≥ 0; also, Z(q)X should be cohomologically supported in degrees
≥ 0, but this is precisely the Beilinson–Soulé vanishing conjecture.)
Here is the statement we really care about:
Theorem A.1 (Bloch–Lichtenbaum, Levine, Voevodsky). Assuming X is smooth over k (otherwise one
should replace K(X) by K 0 (X) in what follows), then the K-theory spectrum K(X) admits a complete,
descending N-indexed filtration whose ith graded pieces is

RΓM (X, Z(i))[2i] := HZar (Z, Z(q)X )[2q] = z q (X)

Therefore there is an associated “Atiyah–Hirzebruch” spectral sequence converging to the K-groups

K∗ (X).

A.2 Motivic cohomology beyond the smooth case

An open problem is to find an analogous motivic filtration on the K-theory of arbitrary schemes (and
to relate it to existing objects, such as algebraic cycles). The particular case which has attracted most
attention in recent years had aimed at constructing such a motivic filtration on the relative K-group
K(X, D), where X is smooth and D ,→ X is a simple normal crossing divisor (which is not necessarily
reduced), such that the graded pieces are related to so-called “Chow groups with modulus”. Sc
Let X be a smooth variety over a perfect field k of characteristic p, and let D = i=1 Di ,→ X
be a reduced normal crossing divisor. For a tuple of positive integers r = (r1 , . . . , rc ), we write rD :=
∪ci=1 ri Di , where ri Di means the ri -infinitesimal thickening of D. In other words, if D is defined locally
by the equation t1 · · · tc = 0, then rD is defined by tr11 · · · trcc = 0.
As explained above, the relative K-theory K(X, rD) should be equipped with a motivic filtration
whose graded pieces are related to higher Chow groups with modulus, which themselves are known to
be close to de Rham–Witt groups and similar objects.
The K-theory K(X, rD) is built from K(X, D) and K(rD, D), which one may refer to respectively
as the tame and wild parts. Since D ,→ rD is an infinitesimal thickening in characteristic p, the trace
map induces an equivalence K(rD, D) ' TC(rD, D); the latter is built from TC(rD) and TC(D), which
carry filtrations from [8] with ith graded piece given by the Zariski hypercohomology of
ϕ
pi −1
hofib N ≥i LW ΩrD/k −−−→ LW ΩrD/k

(resp. replacing rD by D), where LW Ω−/k is the derived de Rham–Witt complex equipped with its
Nygaard filtration.

46
 AWS 2019 – Topological Hochschild homology

Question A.2. Can this be used to build the desired motivic filtration on K(X, rD)? In other words:
(i) Firstly, can the above Frobenius fixed points be related to other objects appearing in the theory
of motivic cohomology with modulus? How does the derived de Rham-Witt complex of a simple
normal crossing divisor look? Is Bhatt–Lurie–Mathew’s saturated de Rham–Witt complex better
adapted?

(ii) Secondly, once we have the right filtration on the wild part, can it be glued to a filtration on the
tame part to get the desired filtration on K(X, rD)?
Question A.3. A closely related but more immediately accessible problem is simply to compute the
THH, etc. (including K-theory) of rings B such as k[t]/tr and k[t1 , . . . , tc ]/(tr11 , . . . , trcc ) (warning: the
latter ring is not the one defining the ncd rD above). Many such results are already available (see
particularly Hesselholt–Madsen and Angeltveit–Gerhardt–Hill–Lindenstrauss respectively), but it would
be nice to have a more natural presentation in terms of (derived) de Rham–Witt theory. The point is
that the relative K-groups can be computed in terms of the topological cyclic homology TC(B), which
fits into a fibre sequence
ϕ−1
TC(B) −→ TC − (B) −→ TP (B),
all of which carry filtrations similar to the discussion above.
More ambitiously, what about rings like Z/pr or O/pr , where O is the ring of integers of a perfectoid
field of characteristic zero? (This seems hard without the forthcoming theory of prismatic cohomology
of Bhatt–Scholze, which will provide an interpretation of the objects to be calculated in terms of p-
derivations.)

47
 Matthew Morrow

B Quasisyntomic coefficients
(Themes: quasisyntomic site, crystals)

Let R be a smooth algebra over a perfect field of characteristic p (more generally one could allow R to
be any quasisyntomic Fp -algebra, or replace Spec R by a non-affine variety).
On the quasisyntomic site we have a sheaf TP0sh which associates to each quasiregular semiperfect A
the (filtered) ring
TP0 (A) ∼=Ab crys (A) ∼ \
= LW ΩA/Fp .

Question B.1. How do locally free sheaves of TP0sh -modules on the quasisyntomic site over R look?
Are they related to crystals on the crystalline site of R? It might be helpful to first consider the mod
p-situation HP0sh = TP0sh /p = LΩ
c −/F . It might also be better to study sheaves of filtered modules, or
p
perhaps modules with a Frobenius.
How are such objects related to modules of cyclotomic spectra over THH(R)?
In theory the analogue of the previous question should already have been answered if R is a smooth
algebra over a field k of characteristic 0: then one expects modules with S 1 -action over HH(R/k) to be
related to R-modules equipped with a filtration and connection satisfying Griffiths transversality. But I
do not know of a reference.

48
 AWS 2019 – Topological Hochschild homology

C Smoothness conditions and Cartier isomorphisms

(Theme: cotangent complex)

C.1 Smoothness conditions

There are three possible nilpotence condition which one can impose on an ideal I:

- nilpotent means that there exists n ≥ 1 such that I n = 0;

- nil means that there exists n ≥ 1 such that xn = 0 for all x ∈ I.

- locally nilpotent means that for each x ∈ I there exists n ≥ 1 such that xn = 0.

Obviously nilpotent ⇒ nil ⇒ locally nilpotent.

A ring homomorphism A → B is called formally smooth if for each A-algebra C and nilpotent ideal
I ⊆ C, any A-algebra homomorphism B → C/I lifts to an A-algebra homomorphism B → C; when
the lifting is moreover always unique, one says that A → B is formally étale. By replacing “nilpotent”
in this definition by “nil” or “locally nilpotent”, we may similarly define n-formally smooth/étale and
ln-formally smooth/étale. (This terminology is totally non-standard; the concept of n-formally smooth
appears elsewhere in the literature, notably in the theory of crystalline cohomology and especially in the
thesis of Berthelot, where it is called “quasi-smooth”.)
One can also ask whether B is a filtered colimit of smooth étale A-algebras, in which case we say it
is ind-smooth; one defines ind-étale similarly. In summary we arrive at the following implications:

ind-smooth ⇒ ln-formally smooth ⇒ n-formally smooth ⇒ formally smooth

and similarly for étale.

There is in fact also a class which is between ind-étale and and ln-formally étale (and strictly larger
than ind-étale), namely weakly étale. Recall here Bhatt–Scholze’s notion of weakly étale, which means
that both A → B and µ : B ⊗A B → B are flat. According to [9, Thm. 1.3], if A → B is weakly étale
then there exists a faithfully flat map B → C such that A → B is ind-étale (i.e., weakly étale maps
are locally ind-étale in the flat topology), whence flat descent implies that A → B is ln-formally étale.
It seems difficult to construct an ln-formally étale map which is not weakly eétale, and so Nikolaus has
raised the following question:

Question C.1. Is ln-formally étale actually equivalent to weakly étale? Consider first the case of an
extension of fields k → k 0 (in which case weakly étale is equivalent to ind-étale, i.e., that k 0 is an algebraic
separable extension of k).

We can also consider conditions coming from the cotangent complex: say that A → B is L-smooth if
LB/A is supported in degree 0 and Ω1B/A is a projective B-module; say it is L-étale if LB/A ' 0. It is
known that

L-smooth ⇒ formally śmooth

whence (since formally étale is equivalent to formally smooth and vanishing of Ω1 ) we get

L-étale ⇒ formally étale

but the converse is false by the next example:

Example C.2. See [Stacks project, Lem. 102.37.1] for an example of a formally étale morphism with
non-vanishing cotangent complex.
Gabber has given an example of a non-reduced (hence not ind-smooth) Fp -algebra B such that
LB/Fp ' 0, i.e., L-étale does not imply ind-smooth. You can find the construction of B on Bhatt’s
webpage. According to the remark between 1.9 and 1.10 of [12], Gabber claims that the construction
can be modified (by replacing the perfections of the Bi which occur in Bhatt’s write-up by the extensions
obtained by adjoining all squares of elements) to produce a Q-algebra with similar properties (see the
next question).

49
 Matthew Morrow

Question C.3. (i) Carry out Gabber’s suggestion to carefully construct a Q-algebra B which is qua-
sismooth (see below) but not ind-smooth.
(ii) Bhatt asks the following: Does there exist an L-étale Q-algebra which is not ind-étale?
(iii) Remaining on the subject of special base rings, is every formally étale Fp -algebra automatically
L-étale?
(iv) Does L-smooth imply ln-formally smooth? Similarly for étale?
Recall from Remark 3.18 that we say A → B is quasismooth if LB/A is supported in degree 0 and
Ω1B/A is a flat B-module. So
L-smooth ⇒ quasismooth
It turns out that if A and B are Noetherian then A → B is quasismooth if and only if it is ind-smooth,
and so all our smoothness conditions are then equivalent. This is a consequence of Néron–Popescu
desingularisation and a hard theorem of M. André [1, Thm. 30, pg. 331]. André’s theorem is actually
more general (for example, he can say something even if A is not Noetherian, as long as A → B is flat),
and it might be useful.

C.2 Incorporating the Cartier isomorphism

Let A be an Fp -algebra. Then the inverse Cartier map
C −1 : ΩnA/Fp → H n (Ω•A/Fp )
is the linear map characterised by
C −1 (f dg1 ∧ · · · ∧ dgn ) = f p g1p−1 · · · gnp−1 dg1 ∧ · · · ∧ dgn
for all f, g1 , . . . , gn ∈ A. In [26], we say that A is Cartier smooth if the following conditions are satisfied:
(Sm1) Fp → A is quasismooth;
(Sm2) the inverse Cartier maps are isomorphisms for all n ≥ 0.
The above criteria are of course satisfied if A is a smooth Fp -algebra, or more generally if A is a regular
Noetherian Fp -algebra by Néron–Popescu desingularisation (an alternative proof avoiding Néron–Popescu
may be found in the recent work of Bhatt–Lurie–Mathew [6, Thm. 9.5.1]), or simply an ind-smooth Fp -
algebra, or even a smooth algebra over a perfect Fp -algebra. But They are also satisfied if A is a valuation
ring of characteristic p: criterion (Sm1) is due to Gabber–Ramero [19, Thm. 6.5.8(ii) & Corol. 6.5.21],
while the Cartier isomorphism (Sm2) is a result of Gabber obtained by refining his earlier work with
Ramero, the proof of which may be found in the appendix of [27].
Note that condition (Sm2) is not a consequence of condition (Sm1), as explained in [6, Warning 9.6.3]:
Bhatt–Gabber’s Example C.2 is a semiperfect, non-perfect Fp -algebra B such that LB/Fp ' 0; then the
inverse Cartier map in degree n = 0 identifies with ϕ : A → A, which has non-zero kernel. Condition
(Sm2) is designed to overcome this sort of pathological behaviour: for example, Cartier smooth + L-étale
is equivalent to perfect, which implies formally étale (even n-formally étale).
Question C.4. (i) Assuming that A is quasismooth over Fp , observe that condition (Sm2) is equiv-
alent to the adjunction LΩ•A/Fp → Ω•A/Fp being an equivalence (c.f., [6, 9.6.2]). Take care of the
difference between LΩ• and its Hodge completion LΩ
d• .

(ii) Is there a good notion (say, closed under composition) of what it means for a morphism of Fp -
algebras to be Cartier smooth? If so, can you prove that OCp /p → O/p is Cartier smooth? (For
the notation see Project C.3. Maybe one can replace O by O and then note that OCp /p → O/p
[ [
can be rewritten as OC p
/p[ → O /p[ , to which one might be able to apply existing results in
characteristic p?)
(iii) How does Cartier smooth compare to Bhatt–Lurie–Mathew’s “universal Cartier isomorphism”?

50
 AWS 2019 – Topological Hochschild homology

C.3 Extending to mixed characteristic

The relation of Cartier smoothness to the material of §5 is as follows: for any Fp -algebra A, we
write LW ΩA/Fp := Rlimr LWr ΩA/Fp , or equivalently the p-adic completion of the left Kan extension
of W Ω−/Fp on A, and we let LW
\ ΩA/Fp be its Nygaard completion; these are connected by natural
morphisms
LW ΩA/Fp −→ LW\ ΩA/Fp −→ W ΩA/Fp
The left object is perhaps the most classical, the middle object is the right one to describe TP (A), and
the right one is the most computable. The key point is the following: if A is Cartier smooth over Fp
(e.g., a valuation ring of characteristic p) then the morphisms are equivalences (this can be extracted
from the arguments of [26]).
The goal of the rest of this project is to establish a similar result in mixed characteristic; some
familiarity with [7] is required.
Let Cp be an algebraically closed, non-archimedean field of mixed characteristic, e.g., Cp , and let
OCp denote its ring of integers. Let C be a complete valued field extension of Cp (possibly highly
transcendental and not necessarily algebraically closed) with corresponding ring of integers O ⊇ OCp
(if C has rank > 1, then we could even let O be a non-maximal valuation subring). The project
concerns various cohomology theories attached to O. Note that Gabber–Ramero [19, Thm. 6.5.8(ii) &
Corol. 6.5.21] again tells us that OC → O is quasismooth. (In contrast, although Zp → OC also has
cotangent complex supported in degree 0, the OC -module Ω1OC /Zp is far from being flat: its p-torsion
Ω1OC /Zp [p] is a rank one OC /p-module.)
Let C be the completed algebraic closure of C, and O ⊆ C its ring of integers; let ∆ be the absolute
[
Galois group of C. Let O be the tilt of O. The first cohomology we are interested in is the Galois
cohomology RΓ(∆, O) and its décalage Ω e O := Lηζ −1 RΓ(∆, O) (Note: similarly to the smooth case
p
b+ ) where X := Spa(O, C)).
e O as Lηζ −1 RΓproét (X, O
studied in [7], it should be equivalent to define Ω p X
The following Hodge–Tate comparison plays the role of the Cartier isomorphism in mixed characteristic:
Question C.5. Do there exist natural (but ignoring Breuil–Kisin–Fargues twists) isomorphisms

ΩiO/OCp ∼
= H i (Ω
eO)

for i ≥ 0? Compare with [7, §8] for the case of a smooth OCp -algebra in place of O. By mimicking the
argument in the smooth case (and taking advantage of the aforementioned result of Gabber–Ramero), it
should be possible to construct a comparison map in the direction →. To check that it is an isomorphism,
maybe reduce modulo p and relate it to the Cartier map for O/p; I suspect one needs to know that
OCp /p → O/p is Cartier smooth, which is why I asked it in Question C.4.
Update: After some discussions with Bhatt, we have decided that the case of an arbitrary complete
valued field extension C is probably too ambitious; one should perhaps assume that the Fp -algebra O/p
admits a finite set of generators over its subring of pth -powers. Lifting these to generators to units
t1 , . . . , td ∈ O should provide a basis dt1 , . . . , dtd of the module Ω1O/OC ; meanwhile, the resulting map
OhT1±1 , . . . , Td±1 i → O provides the analogue of the framing which usually appears in p-adic Hodge
theory. Without this map it seems hard to attack the question. Note that such valuation rings O are
reasonably abundant: take any smooth OC -algebra R and define O to be the p-adic completion of RmR ,
where m is the maximal ideal of OC .
[
Next let Ainf (O) := W (O ) be Fontaine’s infinitesimal period ring construction of O, and
AΩO := Lηµ RΓ(∆, Ainf (O)).
Assuming that the previous question can be answered in the affirmative, then I am confident that it can
be shown that AΩO is exactly the prismatic cohomology of O, that it is already Nygaard complete, and
that the motivic filtration on TP (O) has graded pieces given by AΩO [2i], for i ∈ Z.
Question C.6. Is there any remotely explicit description of the complexes Ω
e O or AΩO , comparable to
the q-de Rham complex in the case of a smooth OCp -algebra in place of O?

51
 Matthew Morrow

D Computing THH(Z)
Theme: spectral sequence calculation

Here are two useful results for analysing the topological Hochschild homology of a ring A.
(i) M. Bökstedt’s [15, Thm. 4.1.0.1] calculation of the groups THHn (Z); they are = Z/mZ if n = 2m ≥
0, and otherwise they vanish.
(ii) Pirashvili–Waldhausen’s [33, Thm. 4.1] first quadrant spectral sequence
2
Eij = HHi (A, THHj (Z, A)) =⇒ THHi+j (A).

Here THH(Z, A) refers to THH of the integers with coefficients in A, whose homotopy groups will
be given by (
THHn (Z, A) =∼ A/mA n = 2m − 1
A[m] n = 2m.
Similarly HH(A, M ) denotes HH of A with coefficients in the A-module M .
For example, if one combines these two results with Bökstedt’s calculation of THH(Fp ), then carefully
chasing through the spectral sequence can be used to compute THH∗ of the ring of integers O of a mixed
characteristic perfectoid field. However, it is also possible to calculate THH∗ (O) without using (i), as
was done in [8, Thm. 6.1].
Question D.1. By using the description of THH∗ (O), as we vary the residue characteristic of O, can
the P-W spectral sequence be run backwards to reprove Bökstedt’s description of THH∗ (Z)?

52
 AWS 2019 – Topological Hochschild homology

E A filtration on HH etc. depending on coordinates

(Theme: classical cyclic homology)

Let k be a base ring and S := k[x1 , . . . , xn ] a polynomial ring over k; let V be the free k-module on
x1 , . . . , xn , so that S = Symk V ; more generally the following works for any free V -module k, or for S
being étale over Symk V .
A different proof of the HKR isomorphism of Theorem 2.8 uses an argument in the case of polynomial
algebras which is a quasi-isomorphism of mixed complexes

mixed complex with zero B-differential corresponding to Ω•S/k

∼
→ (HH(S/k), B)
Vn
See [29, 3.2.2&3.2.3]. The maps defining this depend on writing ΩnS/k = S ⊗k k V , which of course
depends on the choice of co-ordinates (more precisely, only on choice of the free module V they span).
But from this quasi-isomorphism, we see that the statement of Theorem 2.22 remains true for S, even
though we did not assume that k ⊇ Q; the problem is that the decompositions appears a priori to depend
on the coordinates.
Question E.1. When k is a perfect field of characteristic p, are the resulting filtrations on HC − (S/k),
HP (S/k), HC(S/k) the same as the filtrations we constructed in Theorem 3.1? Note that if this is true,
then there will be two interesting consequences: (1) The filtration arising from the above construction
does not depend on choice of coordinates; (2) The filtrations of Theorem 3.1 are split (though the splitting
depends on choice of coordinates).

53
 Matthew Morrow

F An elementary description of HP0 (A/Fp )

Let A be a quasiregular semiperfect Fp -algebra. Logically the following question is redundant, but it is
nevertheless tantalising:
Question F.1. Is it possible to give a direct proof of the isomorphisms of Proposition 3.13 without
using topological periodic cyclic homology or Kan extending from the smooth case?

More precisely, as we explained in the second paragraph of the proof of Proposition 3.13, it would
be nice to directly construct a comparison map DA[ (I) → HP0 (A/Fp ). Since the target HP0 (A/Fp )
is an algebra over HP0 (A[ /Fp ) = A[ (to prove this equality, use the vanishing of LA[ /Fp to see that
'
HP (A[ /Fp ) → HP (A[ /A[ )), it remains to show that HP0 (A/Fp ) admits divided powers along its ideal
Fil1 = Ker(HP0 (A/Fp ) → A) (where we use the filtration of Lemma 3.8).
By using some formal tricks (as in [8, Props. 8.12 & 8.15]) one can try to reduce the existence of
∞
divided powers to the universal case A = Fp [t1/p ]/t − 1; this ring is the group algebra of G := Qp /Zp
over Fp . Since G is p-divisible, a standard group homology calculation (e.g., Brown, Homology of
groups §V.6) identifies the graded Fp -algebra H∗ (G, Fp ) with the divided power algebra Γ∗Fp (V ) on the
p-torsion V := G[p] = H2 (G, Fp ). Proposition 2.24 therefore identifies HP0 (G/Fp ) with Γ b ∗ (V ), the
Fp
completion of Γ∗Fp (V ) with respect to the divided power filtration. In conclusion, the natural map
HP0 (G/Fp ) → HP0 (Fp [G]/Fp ) = HP0 (A/Fp ) shows that the target does contain some divided powers.
One of the problems is showing that this production of divided powers in HP0 (A/Fp ) is independent
∞
of the chosen identification A = Fp [t1/p ]/t − 1. To do this it might be helpful to instead work with the
×
surjection Fp [A ] → A, having reduced to the case that A is local and its units are p-divisible. Or try
to lift to the p-torsion-free case by using Kaledin’s Hochschild–Witt construction?

54
 AWS 2019 – Topological Hochschild homology

References
[1] André, M. Homologie des algèbres commutatives. Springer-Verlag, Berlin, 1974. Die Grundlehren
der mathematischen Wissenschaften, Band 206.

[2] Antieau, B. Periodic cyclic homology and derived de Rham cohomology. arXiv:1808.05246
(2018).

[3] Avramov, L. L., and Vigué-Poirrier, M. Hochschild homology criteria for smoothness. Inter-
nat. Math. Res. Notices, 1 (1992), 17–25.

[4] Bhatt, B. Completions and derived de Rham cohomology. arXiv:1207.6193 (2012).

[5] Bhatt, B. p-adic derived de rham cohomology. arXiv:1204.6560 (2012).

[6] Bhatt, B., Lurie, J., and Mathew, A. Revisiting the de Rham–Witt complex.
arXiv:1805.05501 (2018).

[7] Bhatt, B., Morrow, M., and Scholze, P. Integral p-adic Hodge theory. arXiv:1602.03148
(2016).

[8] Bhatt, B., Morrow, M., and Scholze, P. Topological Hochschild homology and integral p-adic
Hodge theory. arXiv:1802.03261 (2018).

[9] Bhatt, B., and Scholze, P. The pro-étale topology for schemes. Astérisque, 369 (2015), 99–201.

[10] Bökstedt, M. Topological hochschild homology. Bielefeld preprint (1985).

[11] Bökstedt, M. Topological hochschild homology of Z and Z/p. Bielefeld preprint (1985).

[12] Bouthier, A., and Kazhdan, D. Faisceaux pervers sur les espaces d’arcs. arXiv:1509.02203
(2015).

[13] Burghelea, D. Cyclic homology and the algebraic K-theory of spaces. I. In Applications of
algebraic K-theory to algebraic geometry and number theory, Part I, II (Boulder, Colo., 1983),
vol. 55 of Contemp. Math. Amer. Math. Soc., Providence, RI, 1986, pp. 89–115.

[14] Connes, A. Cohomologie cyclique et foncteurs Extn . C. R. Acad. Sci. Paris Sér. I Math. 296, 23
(1983), 953–958.

[15] Dundas, B. I., Goodwillie, T. G., and McCarthy, R. The local structure of algebraic K-
theory, vol. 18 of Algebra and Applications. Springer-Verlag London Ltd., London, 2013.

[16] Dwyer, W. G., and Kan, D. M. Normalizing the cyclic modules of Connes. Comment. Math.
Helv. 60, 4 (1985), 582–600.

[17] Dwyer, W. G., and Kan, D. M. Three homotopy theories for cyclic modules. In Proceedings
of the Northwestern conference on cohomology of groups (Evanston, Ill., 1985) (1987), vol. 44,
pp. 165–175.

[18] Fiedorowicz, Z., and Loday, J.-L. Crossed simplicial groups and their associated homology.
Trans. Amer. Math. Soc. 326, 1 (1991), 57–87.

[19] Gabber, O., and Ramero, L. Almost ring theory, vol. 1800 of Lecture Notes in Mathematics.
Springer-Verlag, Berlin, 2003.

[20] Goodwillie, T. G. Cyclic homology, derivations, and the free loopspace. Topology 24, 2 (1985),
187–215.

[21] Hesselholt, L., and Scholze, P. MFO Arbeitsgemeinschaft: Topological cyclic homology.
https: // www. mfo. de/ document/ 1814/ preliminary_ OWR_ 2018_ 15. pdf (2018).

55
 Matthew Morrow

[22] Hoyois, M. The homotopy fixed points of the circle action on Hochschild homology.
arXiv:1506.07123 (2015).
[23] Illusie, L. Complexe cotangent et déformations. I. Lecture Notes in Mathematics, Vol. 239.
Springer-Verlag, Berlin-New York, 1971.

[24] Illusie, L. Complexe cotangent et déformations. II. Lecture Notes in Mathematics, Vol. 283.
Springer-Verlag, Berlin-New York, 1972.
[25] Kassel, C. Cyclic homology, comodules, and mixed complexes. J. Algebra 107, 1 (1987), 195–216.
[26] Kelly, S., and Morrow, M. K-theory of valuation rings. arXiv:1810.12203 (2018).

[27] Kerz, M., Strunk, F., and Tamme, G. Towards Vorst’s conjecture in positive characteristic.
arXiv:1812.05342 (2018).
[28] Krause, A., and Nikolaus, T. Lectures on topological cyclic homology and cyclotomic spectra.
https: // www. uni-muenster. de/ IVV5WS/ WebHop/ user/ nikolaus/ papers. html (2017).
[29] Loday, J.-L. Cyclic homology, vol. 301 of Grundlehren der Mathematischen Wissenschaften [Fun-
damental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, 1992. Appendix E by Marı́a
O. Ronco.
[30] Loday, J.-L., and Quillen, D. Cyclic homology and the Lie algebra homology of matrices.
Comment. Math. Helv. 59, 4 (1984), 569–591.

[31] Lurie, J. Higher topos theory, vol. 170 of Annals of Mathematics Studies. Princeton University
Press, Princeton, NJ, 2009.
[32] Nikolaus, T., and Scholze, P. On topological cyclic homology. arXiv:1707.01799 (2017).
[33] Pirashvili, T., and Waldhausen, F. Mac Lane homology and topological Hochschild homology.
J. Pure Appl. Algebra 82, 1 (1992), 81–98.

[34] Quillen, D. On the (co-) homology of commutative rings. In Applications of Categorical Algebra
(Proc. Sympos. Pure Math., Vol. XVII, New York, 1968). Amer. Math. Soc., Providence, R.I., 1970,
pp. 65–87.
[35] Quillen, D. G. Homology of commutative rings. Unpublished MIT notes (1968).

[36] Schwede, S., and Shipley, B. Equivalences of monoidal model categories. Algebr. Geom. Topol.
3 (2003), 287–334.
[37] Shukla, U. Cohomologie des algèbres associatives. Ann. Sci. École Norm. Sup. (3) 78 (1961),
163–209.

[38] Shukla, U. A relative cohomology for associative algebras. Proc. Amer. Math. Soc. 15 (1964),
461–469.
[39] Weibel, C. A. An introduction to homological algebra, vol. 38 of Cambridge Studies in Advanced
Mathematics. Cambridge University Press, Cambridge, 1994.
[40] Weibel, C. A., and Geller, S. C. Étale descent for Hochschild and cyclic homology. Comment.
Math. Helv. 66, 3 (1991), 368–388.

56