ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN

LECTURES, BONN 2015)

CHARLES REZK

Abstract. Lecture notes for a series of talks given in Bonn, June 2015. Most of the topics
covered touched in one way or another on the role of power operations in elliptic cohomology.

In June of 2015 I gave a series of six lectures (the Felix Klein lectures) in Bonn. These are
some of my lecture notes for those talks. I had hoped to polish them more carefully, but that
hasn’t happend yet, and at this point probably will not. I am making them available more or
less as-is.
I include only the notes for the first five lectures. Note that some bits in these notes never
made it into the spoken lectures. The notes for the final lecture are too disjointed to be very
useful, so I have omitted them. I hope to soon have preprints on some aspects of what I
spoke about in that lecture.
I’d like to thank the Hausdorff Institute for their hospitality, and for the opportunity to
give these talks, which were a great challenge, but also great fun.
1. What is elliptic cohomology?
I’ll start with a brief ”pseudo-historical” account of elliptic cohomology. This is meant to
be an imprecise overview. The idea is to introduce the basic questions and objects we’re
interested in, and to highlight the main themes of these lectures, which could be summarized
as “power operations” and “isogenies”.
1.1. Genera. A genus is a function which assigns to each closed manifold M of some type
an element Φ(M ) ∈ R of a commutative ring R, satisfying
• Φ(M1 q M2 ) = Φ(M1 ) + Φ(M2 ).
• Φ(M1 × M2 ) = Φ(M1 )Φ(M2 ).
• Φ(∂N ) = 0.
This is the same as giving a ring homomorphism from a suitable cobordism ring, e.g.,
Φ : M SO∗ → R or Φ : M U∗ → R.
Genera with values in R with Q ⊂ R can be described entirely in terms of characteristic
classes, by a formalism due to Hirzebruch.1 For instance, associated to a genus Φ : M U∗ →
R ⊗ Q is a characteristic class for complex vector bundles
KΦ (V → X) ∈ H ∗ (X; R ⊗ Q),
which is completly determined by its characteristic series, i.e., its value on the universal
line bundle
KΦ (x) = KΦ (O(1) → BU (1)) ∈ H ∗ (BU (1); R ⊗ Q) = R ⊗ Q[[x]]
Date: May 8, 2018.
This work was partially supported by the National Science Foundation, DMS-1406121.
1See for instance, [HBJ92].

1
2 CHARLES REZK

where x = c1 (O(1)) ∈ H ∗ (BU (1)) is the usual first chern class, together with a sum formula
Kφ (V ⊕ W ) = Kφ (V )Kφ (W ). Then for a stably almost-complex M ,
φ(M ) = hKφ (T M ), [M ]i.
Conversely, any such characteristic series K(x) determines a genus M U∗ → R ⊗ Q. A series
with K(x) = K(−x) determines a genus M SO∗ → R ⊗ Q.
Genera are not created equal. The most interesting ones (1) have a geometric (or analytic)
interpretation and (2) lift to integral invariants.
1.2. Example (Todd genus). Characteristic series KTd (x) = x/(1 − e−x ). On stably almost
complex manifolds, Td(M ) ∈ Z, and for complex manifolds
X
Td(M ) = (−1)k dim HCoh
k
(M, OM ).

1.3. Example (A-genus).

b Characteristic series KAb(x) = (x/2)/ sinh(x/2). On spin manifolds
A(M ) ∈ Z, and for such computes the index of the Dirac operator on the spinor bundle over
b
M.
These genera come with “families” versions. In either case, the integer value of the genus
can be constructed as an “index”. If we have a family M → S of spin manifolds Ms over a
space S, the family genus defines a class φ(M ) ∈ E ∗ (X) in a generalized cohomology theory
of S. Thus, for a family of n-dimensional spin manifolds, we have A(Mb ) ∈ KO−n (S).
In homotopy theory, all of this can be encoded in a single map: a map of ring spectra
φ: MG → E
where M G is an appropriate bordism spectrum and E a cohomology theory.
1.4. Modular forms. Modular forms are sections of certain line bundles over the moduli of
elliptic curves. Here is how this works in the complex analytic category.
Let
X := {Im(τ ) 6= 0} ⊂ C.
Over this we have a Γ := GL(2, Z)-equivariant bundle of tori.
Eo ? _ Eτ C/(τ Z + Z)
Γy π
 
X o ? _ {τ }

aτ +b
Here A = ( ac db ) ∈ Γ y X by Aτ := cτ +d
, and Γ y E by
A(τ, z) := (Aτ, (det A)(cτ + d)−1 z).
Each fiber is an elliptic curve, i.e., a complex analytic curve of genus 1 with distinguished
point at the equivalence class of z = 0. In fact, the object (orbifold) (E → X )//Γ is the
universal family of such curves.
Let ω → X be the Γ-equivariant line bundle with ωτ = T0∗ Eτ . A (complex analytic)
modular form of weight k is an equivariant holomorphic section
f ∈ H 0 (X //Γ, ω ⊗k ) =: M Fk ⊗ C
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 3

which satisfies a certain growth condition; namely, that for Im(τ )  0, we have
X
f (q) = an q n , q = e2πi τ ,
n≥0

for some an ∈ C, converging near q = 0. Note: we can regard this as sections of line bundles
over a certain compactification X //Γ of this orbifold, obtained by “putting in the cusp”.
The Eisenstein series for k ≥ 1 are described by the q-expansions
∞
B2k X X 2k−1 n
G2k (q) = − + d q .
4k n=1 d|n
2πi τ 1
P
With q = e , they extend to functions of τ ; in fact G2k (τ ) ∼ (a,b)6=(0,0) (aτ +b)2k . For k ≥ 2

they are modular forms. (The series G2 (q) is merely a “quasimodular form”.)
The ring of modular forms (of “level 1”) over C is given by M F∗ ⊗ C ≈ C[G4 , G6 ].
Variants involve sections which are invariant for the action of certain subgroups of Γ.
Every elliptic curve is, canonically, an algebraic curve over C. More generally, one may
consider the algebraic analogue of the above moduli stack, classifying maps E → S which are
proper smooth curves of relative dimension 1 equipped with a section. The picture is this:
X //Γ / MEll
The object MEll is a compactification of the moduli stack MEll of (algebraic) elliptic
curves. Integral modular forms are M Fk := H 0 (MEll , ω ⊗k ). We have
M F∗ ≈ Z[c4 , c6 , (c34 − c26 )/(12)3 ]
where c4 = 240 G4 and c6 = 504 G6 . This ring consists precisely of analytic modular forms
with q-expansions in Z[[q]].
1.5. Elliptic genera: the Witten genus. Elliptic genera were first constructed by Ocha-
nine, via a particular characteristic series with coefficients in modular forms. Another genus
was defined by Witten. The Witten genus, with characteristic series
X x2k

KW (x) = KW (x, τ ) = exp 2 G2k (τ ) .
k≥2
(2k)!
(Note that the non-modular form G2 is excluded in this expression.) This is KW (x) = x/σ(x),
where σ(x, τ ) is the Weierstrass σ-function, and the associated genus is a homomorphism
W : M SO∗ → M F∗ ⊗ C.
The Witten genus has the following remarkable property: applied to spin-manifolds M
such that p21 (M ) = 0, the Witten genus gives a modular form with integral q-expansion:
KW (x) ∈ M F∗ . The idea is that in terms of q = e2πi τ and u = ex , the characteristic series
has the product expansion
x/2 Y (1 − q n )2 2
KW (u, q) = n n −1
e−G2 (τ )x .
sinh(x/2) n≥1 (1 − q u)(1 − q u )
If we remove the last term with G2 (τ )x2 , this can be calculated in terms of the characteristic
series of a twisted version of the A-genus,
b which takes values in Z[[q]] for spin manifolds by
the index theorem. The last term contributes nothing exactly when (p1 /2)(M ) = 0.
4 CHARLES REZK

Thus, remarkably, the Witten genus takes values in integral modular forms, which are
invariants of algebraic elliptic curves.
This is a remarkable fact, and is one reason for the interest in this: somehow, we are led
directly to arithmetic algebraic geometry.
1.6. Elliptic genera and quantum field theory. Even more remarkable is Witten’s
explanation for this, in terms of field theory, which has been elaborated and clarified by many
others, including Segal [Seg88], [Seg07], and Stolz-Teichner [ST11]. Very loosely, we look at a
2-dimensional (extended) field theory, which is a symmetric monoidal functor
F : Bord2 → C
where Bord2 is the bordism 2-category, and C is a 2-category in the world of linear algebra.
The example of topological field theories has become very familiar, due to work of Lurie.
In this case, the field theory should be a supersymmetric conformal field theory: 2-morphisms
of Bord2 should be some kind of conformal (or Euclidean) manifolds (Riemann surfaces), and
in fact should be 2|1-dimensional supermanifolds.
The 2-category C should have 1-morphisms be some kind of hilbert space, and 2-morphisms
as operators. One might additionally allow some kind of twists, obtaining 2-categories C k for
k ∈ Z; and we might actually be looking at sections of a 2-functor C ek → Bord2 .
A 2-morphism between empty 1-morphisms between empty 0-morphisms is a closed 2-
manifold Σ. If we apply F to a torus R2|1 /Λτ , we obtain the “partition function”
ZF (τ ) := F (R2|1 /Λτ ) ∈ C,
a function depending only on the modular parameter. Thus, ZF should be a modular function
(or form).
Given a suitable manifold M , there is a 2-category Bord2 (M ), in which elements U of
Bord2 are equipped with maps U → M . The idea is that there should be a map
{F : Bord2 (M ) → C} → Ell(M ).
In fact, we might hope these are closely related, i.e., that Ell(M ) is the homotopy invariant
approximation to field theories.
The Witten genus of M is supposed2 to be associated to a particular field theory
FW : Bord2 (M ) → C defined naturally on manifolds with string-structure, evaluated on
suitable maps Σ2 → ∗ → M .
This is an example of “dimensional reduction”: because the maps factor through a map from
a 0-manifold, we are supposed to reinterpret this as a 0-dimensional field theory (classified
by H ∗ (−, C)) which is decorated by some additional data coming from the R2|1 /Λτ .
Similarly, we might restrict FW to circle bundles over 1-dimensional manifolds E → Σ1 →
M , obtaining dimensional reduction to a 1-dimensional field theory (classified by K-theory).
(Because F associates to these operators on Hilbert space.) The action of the circle on E
means we live in K[[q]].
1.7. Rigidity. Another property of elliptic genera discovered by Ochanine and Witten is
rigidity. Here is one formulation. Given a manifold M with a U (1)-action, we obtain a
bundle
M ×U (1) EU (1) → BU (1).
2Much of my understanding of this comes from [BE13].
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 5

Using the characteristic series KW (z) one defines a families elliptic genus
W (M x U (1))(x) ∈ H ∗ (BU (1), M F∗ ⊗ C) ≈ M F∗ ⊗ C[[x]].
The rigidity theorem (proved by Bott and Taubes) says that for a U (1)-equivariant spin
manifold with p1 /2(M ) = 0, we have W (M x U (1))(x) = W (M x U (1))(0).
1.8. How is this explained in topology? The modularity and integrality of the Witten
genus is explained by a map of ring spectra M String → tmf [AHR06]. Here M String is the
bordism for associated to the fiber of BString → BSpin → K(Z, 4). The object tmf is the
elliptic cohomology theory associated to the universal algebraic elliptic curve.
Rigidity is explained by the existence of an equivariant version of elliptic cohomology,
taking values in sheaves on the universal elliptic curve itself.
1.9. Multiplicative structure in generalized cohomology theories and power oper-
ations. A generalized cohomology theory h consists of a collection of functors {hn : Top →
Ab}n∈Z together with some additional data, e.g., suspension isomorphisms e hn (X) →
hn+1 (S 1 ∧ X). Such an theory is represented by a spectrum E, which in the original
e
(and simplest) formulation is a sequence of spaces {En }n∈Z and weak homotopy equivalences
∼
{En −→ ΩEn+1 }, so that
hn (X) = hn (X, E) ≈ HomhTop (X, En ).
There is an associated homotopy category of spectra hSp.
The cohomology theories we are usually interested in are multiplicative: h∗ (X) ⊗ h∗ (X) →
h∗ (X). The product pairing in cohomology is encoded using the smash product of spectra:
E ∧ E → E. Thus, if E is an associative and commutative monoid in the homotopy category
of spectra, then h∗ (−, E) takes values in graded commutative rings.
In particular, such a cohomology theory has a natural m-th power map h∗ (X, E) →
h∗ (X m , E) by x 7→ x×m ; restriction along the diagonal embedding X ⊂ X m gives the internal
m-th power.
If E is a structured commutative ring spectrum (e.g., a commutative S-algebra or equivalent
notion), then the associated cohomology theory admits a refinement to a power operation Pm .
diag∗
h∗ (X m ×Σm EΣm ) / h∗ (X × BΣm )
6
Pm
fib∗ incl∗
 
h∗ (X) / h∗ (X m ) / h∗ (X)
x7→x×m diag∗

We are using X × BΣm ⊂ X m ×Σm EΣm → BΣm .

1.10. Power operations and mulitplicative transfer. A theory with power operations
(such as one coming from a commutative S-algebra) is endowed with multiplicative transfers.
This associates to any finite n-fold covering map f : Y → X a multiplicative function
Nf : h∗ (Y ) → h∗ (X).
Here’s a definition. Let π : P → X be the associated principal Σn -bundle of f . Thus,
P ≈ φ : n → Y which identify n with a fiber of f ⊂ Y m .

6 CHARLES REZK

Passage to homotopy quotient by the Σm -action gives

∼
− P ×Σm EΣm → Y m ×Σm EΣm ,
X = P/Σm ←
i.e., up to homotopy a map tf : X → Y m ×Σm EΣm . The mulitpliicative transfer is the
composite
P t∗f
Nf : h∗ (Y ) −−→
m
h∗ (Y m ×Σm EΣm ) −
→ h∗ (X).
Conversely, the power operations can be recovered from the mulitplicative transfers. They
are equivalent formalisms.
A few properties: Ngf = Ng ◦ Nf , and a “push-pull” identity g ∗ ◦ Nf = Nf 0 ◦ g 0∗ associated
to a pullback of covering maps.
1.11. Example (Steenrod operations). Let h∗ = H ∗ (−, Z/2). Then Steenrod defined an
operation
P2 : h∗ (X) → h∗ (X × BΣ2 ) ≈ h∗ (X) ⊗ Z/2[t],
which is in fact the power operation associated to the mod 2 Eilenberg MacLane spectrum.
For x ∈ hn (X) this has value
P2 (x) = Sqn (x) + Sqn−1 (x) t + Sqn−2 (x) t2 + · · · + Sq0 (x) tn ∈ h2n (X × BΣ2 ).
The Steenrod operations Sqk are defined by reading off the coefficients of powers of t.
Note that P2 (x)|t=0 = x2 , which is Sqn (x) = x2 .
You can also use the language of multiplicative transfer to describe these, though it is not
so common: in group cohomology, the multiplicative transfer is called the Evens norm.
1.12. Power operations in K-theory. K-theory was invented by Grothendieck for the
sake of its power operations, which is the theory of lambda rings. It’s easiest to describe
this for topological K-theory using equivariant K-theory, following Atiyah [Ati66]. Thus, for
G y X, we write KG0 (X) = K 0 (X//G) for the Grothendieck group of G-equivariant complex
vector bundles on X.
In this case, the power operation is defined by tensor power of vector bundles, with evident
symmetry.
P diag∗
P m : K 0 (X) −−→
m
K 0 (X m //Σm ) −−−→ K 0 (X//Σm ) = K 0 (X) ⊗ RΣm ,
V 7→ Σm y (V m → X m ) 7→ (Σm y V ⊗m ) → X .

It is not immediately obvious that this construction defined on Vect(X//G) extends to

its Grothendieck group, since Pm is not additive, but this can be shown (for instance, by
extending to complexes of bundles).
P
We can write P m (x) = π φπ (x) ⊗ Vπ , sum over irreducible representations of Σm . In
particular, φsgn (x) = λm (x), the operation defined by the mth exterior power.
All φπ can be written as polynomials in λk s. These operations give K 0 (X) the structure of
a λ-ring: a ring equipped with functions λm : K 0 (X) → K 0 (X) satsifying some properties
of the type satisfied by exterior powers.
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 7

1.13. θp -rings and λ-rings. Suppose p is prime, and restrict to the cyclic subgroup Cp ⊂ Σp .
Write
P Cp
K 0 (X) −−→ K 0 (X//Cp ) ≈ K 0 (X) ⊗ RCp
x 7→ ψ p (x) ⊗ 1 − θp (x) ⊗ (1 + T + · · · + T p−1 ).
Here T is the standard 1-dimensional complex rep of Cp , so RCp = Z[T ]/(T p − 1). The
coefficients are natural functions ψ p , θp : K 0 (X) → K 0 (X).
Properties.
• Multiplicativity: P Cp (xy) = P Cp (x)P Cp (y) and P Cp (1) = 1.
• Setting T 7→ 1 (i.e., forgetting the Cp action) identifies P Cp (x)|T =1 = xp , and thus
ψ p (x) = xp + p θp (x).
• Setting T 7→ e2πi/p (evaluation of character of Cp -representation at generator) sends
N = 1 + T + · · · + T p−1 7→ 0, so
P Cp (x)|T =e2πi/p = ψ p (x).
• The “binomial formula” for tensor product gives
 
⊗k ⊗p−k 1p
P Cp (V + W ) = P Cp (V ) + · · · + V ⊗W ⊗ p
N + · · · + P Cp (W )
k
where N = 1 + T + · · · + T p−1 . In particular, evaluation at T 7→ e2πi/p shows that
ψ p : K 0 (X) → K 0 (X) is a ring homomorphism.
Investigating the behavior of θp in the same way gives identities
θp (1) = 0,
θp (xy) = θp (x)y p + xp θp (y) + p θp (x)θp (y)
X p
p p p
θ (x + y) = θ (x) + θ (y) − 1
p
xk y p−k .
0<k<p
k

Thus, the ring K 0 (X) admits the structure of a θp -ring (Bousfield [Bou96]) (also called a
δ-ring (Joyal [Joy85a]) or a ring with p-derivation (Buium [Bui05])). Note that the identity
ψ p (x) = xp + pθp (x) recovers the Adams operation from θp , and the above identities imply
that ψ p is a ring endomorphism which lifts Frobenius.
We think of θp as a “witness” to the fact that ψ p is a lift of Frobenius.
If p and q are distinct primes, then it’s not hard to show that P Cp P Cq = P Cpq = P Cq P Cp ,
from which it is not hard to see that ψ p ψ q = ψ q ψ p , and in fact ψ p θq = θq ψ p .
1.14. Theorem (Joyal [Joy85b]; see also [Rez14]). There is an equivalence of categories
λ-rings  (R, {θp }primes p ) s.t. ψ p θq = θq ψ p .
 

So these are equivalent descriptions of this theory.

We note the following consequence (which was in fact proved first).
1.15. Theorem (Wilkerson [Wil82]). Torsion free λ-rings are the same as rings equipped
with a family {ψ p } of pairwise commuting lifts of Frobenius.
In some sense, the category of λ-rings can be completly reconstructed from this observation.
8 CHARLES REZK

1.16. Application: Hopf invariant one. Having set all this up, it seems too good not to
give a famous application: a proof of the Hopf invariant 1 theorem due to Adams-Atiyah
[AA66].
1.17. Theorem (Adams). There exists a two-cell CW-complex X = S 2k ∪ D4k with H ∗ (X) ≈
Z[x]/(x3 ) iff k ∈ {1, 2, 4}. (E.g., CP2 , HP2 , OP2 .)
Proof. Suppose we have such an X. A standard argument lets you replace the statement
about H ∗ (X) with the same one for K 0 (X). We know how Adams operations act on K 0 (S 2k ),
so we know that for a prime p,
ψ p (x) = pk x + ap x2 , for some ap ∈ Z.
For p = 2,
ψ 2 (x) = 2k x + a2 x2 , a2 ≡ 1 mod 2,
because ψ is a lift of Frobenius. Apply the identity ψ 2 ψ p = ψ p ψ 2 to x with p odd; equating
2

the coefficients of x2 gives

(2k − 1)2k ap = (pk − 1)pk a2 .
Since p and a2 are odd, we must have 2k |(pk − 1), which is only possible if k ∈ {1, 2, 4} (as
can be seen by checking for p = 3.) 
1.18. Power operations and the multiplicative group. The above constructions make
sense for equivariant K-theory generally. Consider
K 0 (∗//U (1)) = RU (1) = Z[T, T −1 ].
Since µ : U (1) × U (1) → U (1) is a group homomorphism, this is an abelian Hopf algebra,
with µ∗ (T ) = T ⊗ T . It is natural to regard RU (1) = OGm , the ring of functions on the
multiplicative group scheme.
We have ψ p (T ) = T p . By naturality, ψ p is a map of Hopf algebras, corresponding to the
pth power map [p] : Gm → Gm .
1.19. Refinements of power operations. I pulled a switch: I said that power operations
come from a commutative S-algebra, then I produced power operations for K-theory using
equivariant K-theory, which doesn’t come from having the commutative ring spectrum.
K theory is given by a commutative S-algebra. However, the operations I describe are a
refinement:
K 0 (X//Σm ) = K 0 (X) ⊗ RΣm
5
completion

K 0 (X) / K 0 (X × BΣm )
In this case, the difference is not huge, but it is there. In the “classical” case, you can
only give a power operation construction of ψ p , θp as functions K 0 (X) → K 0 (X; Zp ). The
equivariant setting lets you constuct these operations integrally.
The equivariant K-theory power operations come from the fact that equivariant K-theory
is represented by a globally equivariant, ultracommutative ring spectrum (see [Sch16]).
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 9

We can go in the other direction, and replace K-theory by p-adic K-theory:

K 0 (X) / K 0 (X × BΣm )

 
K 0 (X; Zp ) / K 0 (X × BΣm ; Zp )
Something strange happens: the operations ψ q (for primes other than p) do not arise as
power operations on K 0 (−; Zp ). However, they still exist: they come from automorphisms
ψ q : Kp → Kp of the representing spectrum for p-adic K-theory. In fact, they extend to
an action Z× p p
p y Kp . The operations ψ and θ persist as power operations. In fact: the
p-completed K theory of commutative S-algebras is naturally a θp -ring with a compatible
action by Z× p . This is a key feature in the original construction of tmf, and has been used
by Laures, Lawson-Naumann and others to carry out constructions in K(1)-local homotopy
theory.
The moral here is that power operations for K-theory relate to isogenies of the multiplicative
group. One therefore expect that power operations for elliptic cohomology relate to isogenies
of elliptic curves. This is one of the main themes of these lectures.
1.20. Elliptic spectra. Let me jump ahead and describe the current state of the art about
elliptic cohomology.
An even periodic ring spectrum is a homotopy commutative ring spectrum E such that
the groups E e n (S 0 ) = π−n E are 0 for n-odd, and such that the map π2 E ⊗π0 E π−2 E → π0 E
is an isomorphism.
For such a spectrum, we have that E 0 BU (1) ≈ E 0 (∗)[[x]] (non-canonically). The natural
map induced by multiplication µ : U (1) × U (1) → U (1) gives rise to the structure of a formal
group (commutative of dimension 1). We write
GE := Spf(E 0 BU (1)).
The restriction along S 2 = CP1 ⊂ CP∞ = BU (1) gives a canonical identification of ωE =
π2 E = E e 0 (S 2 ) with the cotangent space of GE at the identity.
Thus, every even periodic ring is complex orientable.
An elliptic spectrum is (E, C, α), consisting of an even periodic ring E, an elliptic curve
C/Specπ0 E, and an isomorphism α : GE → C b to the formal completion of C at the identity
section.
Many elliptic spectra can be constructed by the following observation.
1.21. Theorem. If C/SpecA is an elliptic curve whose representing morphism C : SpecA →
MEll is flat, then there exists an elliptic spectrum (EllC , C, α) with π0 EllC = A.
The flatness condition in this theorem turns out to depend only on the formal completion
C of the curve, and the theorem itself amounts to a special case of the Landweber exact
b
functor theorem. In particular, if the line bundle ωC admits a trivialization, we can describe
the resulting homology theory by
(EllC )∗ (X) = M U∗ (X) ⊗M U∗ A[u± ].
1.22. Example (Landweber-Ravenel-Stong). Let A = Z[ 16 , c4 , c6 , (c34 − c26 )/(12)3 ], and let C be
c4 1
the projective curve with affine eqaution y 2 = x3 − 48 x− 864 c6 . There is a corresponding elliptic
10 CHARLES REZK

spectrum, which is a periodic version of the Landweber-Ravenel-Stong elliptic cohomology

theory.
One would like to regard this construction as some kind of “sheaf of spectra” on the stack
of elliptic curves with the flat topology (say). However, the assignment C/SpecA 7→ EllC is
only a functor to the homotopy category of spectra.
1.23. Theorem (Goerss-Hopkins-Miller; see the “Talbot book” [DFHH14]). There is a
Top
presheaf OEll of commutative S-algebras over the étale site of MEll , so that for each étale
Top
map C : SpecA → M` the ring spectrum OEll (C) is an elliptic spectrum with curve C/SpecA.
Top
We can think of OEll as a homotopy theoretic lift of the “classical” structure sheaf OEll
over MEll , whose value OEll (C/SpecA) = A. Furthermore, we have isomorphisms of sheaves
Top
π2n OEll ≈ ω ⊗n .
There is an extension of this result to the compactification MEll .
1.24. Theorem (Goerss-Hopkins-Miller, Hill-Lawson [HL16]). The above construction extends
Top
to a presheaf OEll of commutative S-algebras over the log-étale site of MEll , with an analogous
correspondence to the “classical” structure sheaf.
We can evaluate this sheaf over any log-étale map N → M, obtaining a commutative
Top
S-algebra OEll (N ). Furthermore, there is a spectral sequence
Top
H s (N , ω t/2 ) =⇒ πt−s OEll (N ).
Top
1.25. Example. Define TMF := OEll (MEll ). This is the ring of periodic topological
modular forms. The edge map of the spectral sequence
π∗ TMF → H 0 (MEll , ω ∗ ) = M F∗ [∆−1 ]
is not an isomorphism, though it becomes one after 6 is inverted. The discriminant ∆ ∈ M F12
is not in the image; however, the spectrum TMF is periodic on ∆24 ∈ π576 TMF.
Top
1.26. Example. Define Tmf := OEll (MEll ). This is the ring of topological modular forms.
The edge map of the spectral sequence
π∗ Tmf → H 0 (MEll , ω ∗ ) = M F∗
is not an isomorphism. Though M F∗ is concentrated in non-negative degree, Tmf has much
homotopy in negative degrees, much of it contributed by H 1 (MEll , ω ∗ ).
1.27. Example. Define tmf := Tmf ≥0 , the (−1)-connected cover of Tmf. This is the ring of
connective topological modular forms.
1.28. Example. There is a map M(n) → M which over M carries the universal example
of an elliptic curve C equipped with a (naive) full level n structure (i.e., an isomorphism
C[n] ≈ (Z/n)2 of group schemes). The resulting spectrum of global sections is called Tmf(n).
Top
The object (MEll , OEll ) is a derived Deligne-Mumford stack. Lurie has given an interpre-
tation of this object using the notion of a derived elliptic curve [Lur09].
Let A be a commutative S-algebra. A derived elliptic curve over A is a derived abelian
group scheme C → SpecA which is an abelian group object in derived schemes over SpecA,
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 11

and such that the underlying map of schemes is an elliptic curve over π0 A (in the ordinary
sense).
An oriented derived elliptic curve is data (A, C, α) consisting of a derived elliptic curve
C → SpecA together with an isomorphism
∼
α: Cb−→ Spf ABU (1)+
of formal derived group schemes.
1.29. Theorem (Lurie).
Top
• The object (MEll , OEll ) is the moduli stack of oriented elliptic curves. In particular,
there is a universal oriented derived elliptic curve C → MEll .
• Given a map C : SpecA → MEll classifying an oriented derived elliptic curve, there
exists a globally equivariant cohomology theory associated to it: i.e., for each compact
Lie group, an equivariant cohomology theory EllC (−//G) : GTopop → Sp, with change
of group isomorphisms Ell∗C ((X ×H G)//G) ≈ Ell∗C (X//H) whenever H ⊆ G and
H y X.
• When G is an abelian compact Lie group and X is a finite G-CW-complex, the value
Ell∗C (X) is naturally the global sections of a coherent sheaf F(X) on the derived group
scheme C ⊗ G. b

In particular, for G = U (1), we obtain for each C → SpecA a U (1)-equivariant cohomology

theory taking values in coherent sheaves on C.
Some remarks.
• This is a great theorem. However, it is not so easy to use, since in general it is hard
Top
to construct maps SpecA → (MEll , OEll ). Note that there is one case where such
maps exist for free: those whose underlying map is étale.
• There are many examples of elliptic cohomology theories which don’t fit into this set, or
at least aren’t known to. For instance, there is an elliptic cohomology theory associated
to the general Weierstrass equation, defined over the ring A = Z[a1 , a2 , a3 , a4 , a6 , ∆−1 ].
The spectrum is
EllWeier = TMF ∧ (ΩSU (4))γ ,
where the last term is the Thom spectrum on the Bott map ΩSU (4) → BU . This is
an E2 -ring; I don’t know if it can be realized as a commutative S-algebra.
Another such example is the ring of topological quasimodular forms. There is a stack
Q → MEll , which classifies the data (C, s) consisting of an elliptic curve C/SpecA
together with a splitting s of the “Hodge extension”
0 → H 0 (C, Ω1C ) → HdR
1
(C/S) → H 1 (C, OC ) → 0
(an exact sequence of A-modules). You can extend this over the cusp, and the resulting
ring of global sections is
QMF∗ = H 0 (Q, ω ∗ ) ≈ Z[b2 , b4 , b6 , 14 (b2 b6 − b24 )].
The element b2 ∈ QMF2 can be related to the Eisenstein series 12 G2 (τ ). There is a
candidate topological version:
tqmf := tmf ∧ (ΩS 5 )γ ,
12 CHARLES REZK

with a map π∗ tqmf → QM F∗ (the class b2 isn’t in the image, but 2b2 = 24 G2 (τ ) is).
This is a strictly associative ring spectrum, and is also homotopy commutative. Can
it be made strictly commutative, or even just E2 ?
• Lurie’s technology is functorial with respect to isomorphisms of (oriented derived)
elliptic curves. One might ask whether one obtains interesting maps between equi-
variant elliptic cohomology theories coming from homomorphisms between elliptic
curves.
There is an obvious constraint. Equivariant cohomology theories are naturally
associated to the pair (C/S, α), where α is an orientation. Any map C → C 0 of oriented
derived elliptic curves over A is necessarily étale, since preserving the orientation
implies that Cb → Cb0 is an isomorphism. Thus, we can only expect to construct maps
of equivariant cohomology theories associated to separable isogenies.
I’m not going to explain the proofs of the above theorems. Nowadays there are decent
references for the Goerss-Hopkins-Miller construction, e.g., in the Talbot book.
1.30. The σ-orientation. There are several results relating to the Witten genus. Work of
Ando-Hopkins-me, building on work of Ando-Hopkins-Strickland [AHS01], [AHS04], proves
1.31. Theorem (Ando-Hopkins-Rezk). The Witten genus refines to a map M String → tmf
of commutative S-algebras.
I will discuss some aspects of this in later talks. The proof involves commutative S-algebra
models for elliptic cohomology, and depends in the end on power operations for various
theories.
Rigidity can be explained using models of equivariant elliptic cohomology: in the complex
analytic case by Rosu [Ros01] and Ando-Basterra [AB02]. Lurie has a more general formulation
of this.
2. Descent for isogenies and deformations of formal groups
There is one piece of equivariant elliptic cohomology which is accessible from non-equivariant
algebraic topology. This is the one associated to the universal deformation of a supersingular
elliptic curve, for groups G which are finite p-groups. This cohomology theory can be identified
with Borel equivariant Morava E-theory.
2.1. Formal groups in finite characteristic. The theory of power operations for Morava
E-theory is due to Ando, Hopkins, and Strickland. It is a template for power operations in
equivariant elliptic cohomology. Morava E-theory is the theory associated to the universal
deformation of some formal group of finite height.
We consider formal groups over some ring R, and homomomorphisms between such. A
particularly important example in characteristic p is the
2.2. Example (Frobenius homomorphism). If Fp ⊆ R, write σ : R → R for the absolute
Frobenius. The relative Frobenius is a homomorphism
F r : G → (σ r )∗ G.
defined on functions by
σr
id ⊗σ : R ⊗R OG = Oσr ∗ G → OG .
r
In terms of any local coordinate on G, it is given by x 7→ xp .
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 13

If G/k where k is a field of characteristic p, then any non-zero homomorphism f : G → G0

of formal groups can be factored uniquely as f = g ◦ F r , where g : (σ r )∗ G → G0 is an
isomorphism. We say f has degree pr .
Suppose k has characteristic p, and consider the homomorphism [p] : G → G. We say that
G has height h if deg[p] = ph , or infinite height if [p] = 0. We have h ∈ {1, 2, . . . , ∞}.
The category of formal groups over a field k of char p is well understood.
2.3. Theorem. If k is separably closed, then two formal groups over k are isomorphic if and
only if they have the same height. For a given G of finite height h, the endomorphisms End(G)
are a finite pro-etale ring scheme, non-canonically isomorphic to Dh ≈ WFph hSi/(S h − p)
(where Sa = aσ S), a maximal order in a central division algebra of rank h2 and invariant
1/h over Qp .
The automorphism group Sh = Aut(G0 ) is called the Morava stabilizer group.
2.4. Deformations of formal groups. Let R be a complete Noetherian local ring, with
p ∈ m. Say that a formal group G/R is a deformation of a height h formal group if GR/m
is height h.
Morphisms between such are completely determined by their restriction to the special fiber.
2.5. Proposition. Suppose f, f 0 : G → G0 are homomorphisms between formal groups over R
which are deformations of height h formal groups. Then f = f 0 if and only if fR/m = fR/m
0
.
Proof. Because we can add homomorphisms, it suffices to consider the case with f 0 = 0.
Because R is complete and noetherian, the result follows by proving it inductively for R with
mn = 0. In fact, suppose R ⊃ m ⊃ I with mI = 0 (e.g., I = mn−1 ), and suppose fR/I = 0.
We want to show f = 0.
We use the identity f ◦ [p]G = [p]G0 ◦ f . In choices of coordinates for G and G0 , write
f (x) = axk + higher degree ∈ I[[x]],
h
[p]G (x) ≡ cxp + higher degree mod m,
ph
[p]G0 (y) ≡ c0 x + higher degree mod m,
h h
where c, c0 are units. Comparing leading coefficients (of xkp ) in the identity, we get ack = c0 ap .
Since c and c0 are units and a ∈ I squares to 0, we must have a = 0. 
Fix a formal group G0 of height h < ∞, over a perfect field k.
A deformation of G0 to G is data (G, i, α), where
• G is a formal group over R,
• i : k → R/m is an inclusion of fields,
∼
• α : GR/m −→ i∗ G0 is an isomorphism of formal groups over R/m.
(If k = Fp , we can omit i.)
The data (i, α) will also be called a G0 -deformation structure on G. Write DG0 (G/R)
for the set of such. Any isomorphism f : G → G0 of formal groups over R induces a map
f! : DG0 (G/R) → DG0 (G0 /R), sending
−1
f! : (i, α) 7→ (i, fR/m ◦ α).
14 CHARLES REZK

f
G / G0
∼

fR/m
GR/m / G0R/m
∼

α ∼ ∼ α0
 
∼ / i∗ G0
i∗ G0
id
By the above proposition, if (G, i, α) and (G0 , i0 , α0 ) are G0 -deformations, there exists at
most one isomorphism f : G → G0 compatible with the deformation structure (in the sense
that i = i0 and α0 ◦ fR/m = α; these are sometimes called ?-isomorphisms). In particular,
deformations have no non-trivial automorphisms.
2.6. Theorem (Lubin-Tate). The functor
R 7→ iso. classes of G0 -deformations over R
is representable by a complete local ring A = AG0 . Furthermore, there is a non-canonical
isomorphism
A ≈ Wk[[a1 , . . . , ah−1 ]].
The tautological example is the universal deformation Guniv of G0 over A.
We can also think of AG0 as classifying deformation structures on a given formal group.
Given a formal group G/R, there is a bijection
DG0 (G/R) = {G0 def. str. on G/R}  {φ : AG0 → R s.t. φ∗ Guniv ≈ G as f.g. /R}
There’s an obvious action of Aut(G0 ) y DG0 (G/R), (by γ · (i, α) = (i, i∗ (γ) ◦ α)) and thus
an action Aut(G0 ) y AG0 .
2.7. Deformation structures and isogenies. We say that a homomorphism f : G → G0
of formal groups is an isogeny if the induced map on functions f ∗ : OG0 → OG is finite and
locally free. Any homomorphism f between deformations is an isogeny iff fR/m is a non-zero
homomorphism.
Isogenies have well-defined kernels: Ker f = SpecOG ⊗OG0 R is a finite abelian group scheme
over R.
We can extend the construction DG0 to a functor
{f.g. over R and isogenies} → {Sets}
as follows. Given an isogeny f : G → G0 of degree pr over R and (i, α) ∈ DG0 (G/R), define
f! (i, α) = (i0 , α0 ) ∈ DG0 (G0 /R) using the diagram
f
G / G0

fR/m
GR/m / G0R/m
α ∼ ∼ α0
 
i∗ G0 / i∗ (σ r )∗ G0
Fr
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 15

so that i0 := i ◦ σ r and α0 is the unique isomorphism making the diagram commute. This
works because all isogenies of degree pr between formal groups over k factor through F r .
In other words, we can push-forward a deformation structure along isogenies, by factoring
it through the appropriate power of Frobenius.
Conversely, given deformations (G, i, α) and (G0 , i0 , α0 ) and an r ≥ 0, there exists at most
one isogeny f : G → G0 which is compatible with F r , in the sense that the above diagram
commutes. In fact, the data “pair of deformations related by an isogeny compatible with F r ”
(up to ?-isomorphism) is representable by a ring
Ar = AG0 ,F r ≈ (AG0 ⊗A
b G0 )/Jr .
This ring carries the universal example of an isogeny s∗ Guniv → t∗ Guniv which deforms
F r : G0 → (σ r )∗ G0 .
Observation: if we fix the source deformation (G, i, α), then an isogeny f : (G, i, α) →
(G0 , i0 , α0 ) which deforms F r is determined, up to ?-isomorphism, by the kernel H := Ker f ,
which is a finite subgroup scheme of G. Thus,
{Ar → R}  {(G0 -def. (G, i, α), H ≤ G subgp. of rank pr )}
2.8. Morava E-theory. For any G0 /k (finite height), there is an even periodic ring spectrum
K = KG0 with π0 K = k and GK = G0 . (With the caveat that if p = 2, it is not actually
homotopy commutative.) This admits a strictly associative ring structure (Robinson, Baker),
but it does not admit the structure of a commutative S-algebra. Such K are sometimes
called periodic Morava K-theories.
(The actual Morava K-theory spectrum K(h) is an “indecomposable summand” of KG0 .)
For any G0 /k, there is an even periodic ring spectrum E = EG0 associated to the universal
deformation of G0 , with π0 E = AG0 , and GE = Guniv . It can be constructed as the spectrum
representing a Landweber exact cohomology theory.
2.9. Theorem (Goerss-Hopkins-Miller). Every EG0 admits (essentially uniquely) the structure
of a commutative S-algebra. Furthermore, the assignment G0 /k 7→ EG0 refines to a functor
{finite ht. f.g. over perfect fields and isos.} → {comm S-alg.}.
In particular, the automorphism group of G0 acts on EG0 though maps of commutative
S-algebras.
2.10. Power operations for Morava E-theory. Consider the power operation associated
to the commutative S-algebra E. This has the form
Pm : E 0 X → E 0 (X × BΣm ).
Explicitly, this is defined by the construction which takes a map a : Σ∞
+ X → E of spectra,
and produces the map
∞ ×m f ∧m
Σ∞ ∞ ∧m ∧m
+ X → Σ+ XhΣm ≈ (Σ+ X)hΣm −−→ EhΣm → E.

This is multiplicative Pm (ab) = Pm (a)Pm (b) but not additive. However, the failure to be
additive is encoded in a formula
X
Pm (a + b) = TrΣΣi ×Σj Pi (a) × Pj (b).
m

i+j=m
16 CHARLES REZK

Here Pi (a) × Pj (b) ∈ E 0 (X × BΣi × BΣj ) is a partial product, and TrΣ

Σi ×Σj is the transfer
m

map. This is a consequence of the “binomial formula” for symmetric products

a
(X q Y )×m
hΣm ≈ (X ×i × Y ×j × Σm /(Σi × Σj ))hΣm .
i+j=m
0
The transfer ideal I ⊂ E X × BΣm is the sum of the images of all transfer maps
Tr : E 0 X × BΣi × BΣj → E 0 X × BΣm
associated to proper subgroups of the form Σi × Σm−i ⊂ Σm . The resulting map to the
quotient
E 0 X → E 0 (X × BΣm ) → E 0 (X × BΣm )/I
is a ring homomorphism.
It is a fact (HKR) that E 0 BΣm is finite and free as a π0 E-module. So we can rewrite this
as
P m : E 0 X → E 0 (X) ⊗π0 E E 0 BΣm → E 0 X ⊗π0 E E 0 BΣm /I.
Because E is a p-local theory, I = E 0 BΣm unless m = pr .
2.11. Theorem (Strickland [Str97], [Str98]). There is an isomorphism of rings Ar ≈
E 0 BΣpr /I. Furthermore, the universal deformation of F r is obtained by evaluating the
map at X = BU (1), in the sense that
• s : A0 → Ar , classifying the source, is E 0 (∗) → E 0 BΣpr /I induced by BΣm → ∗;
• t : A0 → Ar , classifying the target, is P pr : E 0 (∗) → E 0 BΣpr /I;
• the universal lift s∗ Guniv → t∗ Guniv defined over Ar is represented by the ring map
t s
E 0 BU (1) ⊗A0 Ar → E 0 BU (1) ⊗A0 Ar
produced by evaluating the commutative square
P pr
E 0 (∗) / E 0 (∗) ⊗E 0 (∗) E 0 BΣpr /I

 
E 0X / E 0 X ⊗E0 (∗) E 0 BΣpr /I
P pr

Remarks.
• We have E 0 (∗)/m ⊗E 0 (∗) E 0 BΣpk /I ≈ E 0 (∗)/m. Thus,
E 0X / E 0 X ⊗E 0 (∗) E 0 BΣpr / E 0 X ⊗E 0 E 0 BΣpr

r
x7→xp '  
E 0 X ⊗E 0 E 0 (∗) / E 0 X ⊗E 0 E 0 /m
showing that the isogeny defined by topology really is a deformation of F r .
• An important part of proving this involves showing that E 0 BΣpr /I is itself a free
E 0 -module. This is a consequence of results of Kashiwabara [Kas98] that K(n)∗ (QS 0 )
is a polynomial algebra.
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 17

• The rest of Strickland’s argument is quite complicated. (1) He shows that s : A0 → Ar

is finite and free of the expected degree (same as the rank of E 0 BΣpr /I as an E0 -
module). (2) The map Ar → E 0 BΣpr /I is shown to be injective after tensoring with
A0 /m by a delicate calculation using a description of certain elements in terms of
euler classes of vector bundles.
Schlank and Stapleton [SS15] have an alternate proof of the second part of the
argument, using a “transchromatic character map” to reduce to the case of height 1.
2.12. Operations on the homotopy of K(n)-local commutative S-algebras. I’ve de-
scribed power operations as operations on E 0 X.
In fact, the proper context is K(n)-local commutative E-algebras. Given a commutative
E-algebra R, the power construction defines a function
+
Pm : π0 R → π0 RBΣm ,
where RY+ = F(Σ∞ + Y, R). The case of R = E
X+
is what is discussed above.
A spectrum F is K(n)-local if [X, F ] ≈ 0 for any spectrum X with K(n)∗ X = 0. The
map
+ +
R ∧E E BΣm → RBΣm
is not a weak equivalence in general, but is if R is assumed to be K(n)-local as well. Thus
we obtain a function
+
Pm : π0 R → π0 (R ∧E E BΣm ) ≈ π0 R ⊗π0 E E 0 BΣm ,
and we proceed as above.
Let B = π0 R. The data of power operations provides for each r ≥ 0 a map of A0 -algebras
s t
ψr : B → B ⊗A0 Ar .
Furthermore, for each r, r0 the diagram
ψr
B / B ⊗A0 Ar
s t

0 0
ψ r+r ψ r ⊗id
 
B ⊗A0 Ar+r0
s t / s
B ⊗A0 Ar0 ⊗A0 Ar
t s
id ⊗µ∗
t s
commutes, where µ∗ : Ar0 +r → Ar0 ⊗ Ar classifies the composition of a deformation of F r
0
with a deformation of F r .
We say that the A0 -algebra B is equipped with “descent for isogenies”. That is, any
G0 -deformation to R determines a ring
G,(i,α)
B R (G, (i, α)) := B ⊗A0 R,
0
while any isogeny f : G → G determines a map of rings
B R (G, (i, α)) → B(G0 , f! (i, α)),
defined using
t ψr ⊗1 s
B ⊗A0 Ar −−−→ B ⊗A0 Ar
and the map Ar → R classifying f . Recall that f! (i, α) is the pushforward of the deformation
structure along the isogeny described earlier.
18 CHARLES REZK

The data B R is a functor

{G0 -defs. to R and isogenies} → {comm. R-algebra}.
2.13. Frobenius congruence. There is a pullback square of rings
E 0 BΣp / E 0 BΣp /I
(∗→BΣp )∗
 
E 0 (∗) / E 0 (∗)/p
The right-hand column is a map A1 → A0 /p. The ring A0 /p supports the universal deforma-
tion of G0 to any ring R of characteristic p. There is a distinguished (but not unique) example
of an isogeny between deformations in characteristic p, namely the (relative) Frobenius
isogeny
F : G → σ ∗ G.
s
The fact that ψ1 : π0 R → π0 R ⊗A0 A1 lifts to a map π0 R → π0 R ⊗E0 (∗) E 0 BΣp means that
ψ1
π0 R / π 0 R ⊗A 0 A 1

 
π0 R/p / π0 R/p = π0 R ⊗A0 A0 /p
x7→xp

commutes. We call this the Frobenius congruence [Rez09]. The operation Pp : π0 R →

π0 R ⊗ E 0 BΣp is an explicit lift of ψ1 , which we think of as a witness for the Frobenius
congruence.
Frobenius congruence amounts to saying that, for any R ⊃ Fp the functor B R carries the
Frobenius isogeny F : G → σ ∗ G to the relative pth power map between R-algebras.
2.14. Example: multiplicative group. Let G0 /Fp be the multiplicative group. Then
A0 = Zp , and Guniv can also be taken to be the multiplicative group. The Morava E-theory
can be identified with E = Kp , p-complete K-theory.
We have that Ar = Zp for all r.
Thus, the data of descent for isogenies on an A0 -algebra B amounts to giving a single ring
homomorphism
ψ : B → B,
which must satisfy ψ(x) ≡ x mod pB. The lift to a function B → B ⊗ E 0 BΣp gives an
p

explicit witness to this congruence.

Thus, π0 of a K(1)-local Kp -algebra is a θp -ring.
2.15. Supersingular elliptic curves. Let C0 /k be an elliptic curve over a field of charac-
teristic p, and let C
b0 be its formal completion. Then either
• Cb0 has height 1, in which case we say C is ordinary, or
• Cb0 has height 2, in which case we say C is supersingular.
There are only finitely many supersingular curves (up to isomorphism) at any prime p.
Suppose C0 /k is supersingular. Then C0 /k has a universal deformation, which is an elliptic
curve Cuniv /A0 with A0 ≈ Wk[[a]]. The underlying formal group C buniv is in fact the universal
deformation of the height 2-formal group C b0 .
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 19

The associated Morava E-theory is an elliptic spectrum.

2.16. Example ([Rez08]). Let C0 /F2 be the curve with (affine) Weierstrass equation y 2 +y = x3 .
Write E = ECb0 for the associated Morava E-theory, with π0 E = A0 = Wk[[a]].
We have A1 ≈ A0 [d]/(d3 − ad − 2), with
s t
A0 →
− A1 ←
− A0 , a 7→ a, a0 = a2 + 3d − ad2 ←[ a.
The map A1 → A0 /p classifying Frobenius is given by a 7→ a, d 7→ 0.
If B = π0 R, then the above structure is entirely determined by the single map Pp : B →
B ⊗ E 0 BΣp . The map Pp produces a ring a homomorphism
ψ1 s
B −→ B ⊗A0 A1 ,
together with a witness for the Frobenius congruence (which says ψ1 (x) ≡ x2 mod d). The
homomorphism ψ1 is subject to a single relation, namely that there exists a ψ2 -factoring
ψ1
B / B ⊗A0 A1
s t

ψ2 ψ1 ⊗id
 
B ⊗A0 A2 / /
s t s t s
B ⊗A0 A1 ⊗A0 A1
id ⊗µ∗

(The inclusion A2 ⊂ A1 ⊗ A1 is split as a map of left A0 -modules.) All higher ψr are uniquely
determined once this is known. ThisQis a generic feature for descent for isogenies associated
to any formal group: the bialgebra Ar is quadratic in a precise sense [Rez12a].
For a formal group of height 2, the subring A2 ⊂ A1 ⊗ A1 is a pullback
A2 / t
A1 ⊗A0 A1
s

id ⊗w
 
A0 / A1
s

where This map w classifies the “dual isogeny”. Every degree p-isogeny f factors the
multiplication by p map, so there exists a unique fb such that fbf = p. When the formal
group has height 2, fb is also a degree p-isogeny, and f 7→ fb is represented by w. In our p = 2
example, w acts by a 7→ a0 and d 7→ d0 = a − d2 .
Analogous calculations are known for s.s. curves over F3 and F5 , by work of Yifei Zhu
[Zhu14], [Zhu15]. For more about calculation of power operations in the height 2 case, see
[Rez13].
3. Modular isogeny complexes, and the Koszul property
Last time, I described how power operations for Morava E-theory are basically equivalent
to giving “descent for isogenies” to deformations of formal groups.
I want to put this is a wider context, by thinking about “descent for isogenies” for elliptic
curves.
20 CHARLES REZK

3.1. Descent for isogenies in the elliptic moduli stack. Recall M = MEll , the moduli
stack of (smooth) elliptic curves.
{S → M} ↔ {(groupoid of) ell. curves C → S}
For each N ≥ 1, there is an object MIsog N , classifying N -isogenies of elliptic curves.
{S → MIsog N } ↔ {gpd. of N -isogenes C → C 0 def. over S}.
s t
There are maps M ← − MIsog N →− M encoding the source and target. We can think of the
map s as carrying the “universal subgroup of rank N ”. That is, every N -isogeny f : C → C 0
over S deterimines a family G := Ker f ≤ C of subgroup schemes (finite flat over S of rank
N ). Conversely, given such G ≤ C we can form the quotient map g : C → C/G, and any
N -isogeny f : C → C 0 with ker f = G factors uniquely through g.
The map s is representable. This means that for every C : SpecA → M, there is a
pullback square
SpecAIsog N / MIsog N

s
 
SpecA / M
C
of stacks.
3.2. Theorem (Katz-Mazur [KM85]). Each s : A → AIsog N is flat and locally free, of rank
= #{subgps. of (R/Z)2 of order N }.
`
Let MIsog := N ≥1 MIsog N . There is a simplicial object M• of the form
o
o
o o
MIsog o
t s
Mo MIsog ×M MIsog o ···
o
o
which encodes the fact that “elliptic curves and isogenies” forms a category. We can think
of Md as the thing that represents sequences {C0 → · · · → Cd } of isogenies over S, or
equivalently as representing chains G1 ≤ G2 ≤ · · · ≤ Gd ≤ C0 of finite subgroups, where
Gi = Ker[C0 → Ci ].
3.3. Curves with descent for isogenies. Fix a map of stacks Y → M. We can think
of this as characterizing an elliptic curve over Y. Alternately, Given C : S → M, we can
consider the collection (set or groupoid or ∞-groupoid) of lifts
 

 >Y 
Y (C/S) :=
  
/M
 
S
The element α ∈ Y (C/S) is a “Y -structure on C/S”, and we can regard Y as the moduil
stack of elliptic curves equipped with a chosen Y -structure.
3.4. Example. The universal curve Y = E → M. Here Y (C/S) = C(S) = Γ C → S) the set
of sections.
∼
3.5. Example. A “full level n-structure” on C/S is a choice C/S of isomorphism λ : (Z/n)2S −
→
C[n] of group schemes. There is a corresponding stack Y = M(n) → M. Note that
M(n) = SpecM (n) for n ≥ 4, and the map is étale.
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 21

Say Y → M has descent for isogenies if it is equipped with a map Y• → M• of

simplicial stacks such that
• Y0 → M0 is the given map,
• for each k, the square
Yk / Mk

s s
 
Y0 / M0

is a pullback of stacks. (Here s is the operator induced by h0i : [0] → [k].)

s
3.6. Remark. Here is how to think about this. Note that Y1 ≈ Y ×M MIsog . We have
{S → Y1 } ↔ {(f : C → C 0 , α ∈ Y (C/S))}.
The simplicial structure in Y• induces
(C 0 ,α0 )

(C,α,f ) %
S / Y1 / Y
t
s
$ 
(C,α)
Y
Thus we obtain f! : Y (C/S) → Y (C 0 /S), by f! (α) = α0 . Thus, descent for isogenies provides
a way to “push Y -structures forward along isogenies”. The properties imposed on Y• imply
that this pushforward is natural and functorial.
3.7. Remark. Suppose Y = SpecA → M is representable by a ring A, and has descent
` s t
for isogenies. Then Y1 = SpecAIsog N . Thus the maps Y ←
− Y1 →
− Y are represented a
collection of ring maps
s t
A→ − AIsog N ←
− A.
t s
There are also maps AIsog N N 0 → AIsog N ⊗A AIsog N 0 representing composition.
Associated to this is a category Mod(Y) of isogeny modules on Y. The objects are
A-modules M equipped with A-module maps
s t
ψIsog N : M → M ⊗A AIsog N
satisfying the identities of a comodule structure. (You can define a category Mod(Y) whenever
Y → M is representable and has descent for isogenies.)
3.8. Examples.
3.9. Example. The universal curve E → M has descent for isogenies tautologically. For
f : C → C 0 over S, the map f! : E(C/S) → E(C 0 /S) is just f : C(S) → C 0 (S).
3.10. Example. The stack Y = M(n) → M of level n-structures. For any isogeny f : C → C 0
of degree prime to n, we have an induced isomorphism C[n] → C 0 [n] of group schemes, and
hence a map f! : Y(C/S) → Y(C 0 /S).
Thus, M(n) → M has descent for isogenies of degree prime to n.
22 CHARLES REZK

3.11. Example. Fix a supersingular curve C0 /k at some prime p. There is a formal stack Y =
M∧C0 → M, which classifies deformations of C0 to complete local rings with p topologically
nilpotent.
This object has descent for pth power isogenies, by the same recipe I described last time
for deformations of formal groups. Thus Ar = AIsog pr .
3.12. Example. The Tate curve Tate : SpecZ((q)) → M is the “deleted formal neighborhood
of infinity” in M. This object has descent for isogenies. This data turns out to encode the
power operations for equivariant EllTate constructed by Ganter [Gan07], [Gan13].
3.13. Example. Consider the analytic moduli space of elliptic curves. This is Man = X //Γ,
where X = { τ ∈ C | Imτ 6= 0 } and Γ = GL(2, Z). The universal curve is E//Γ, where
Eτ = C/(τ Z + Z).
Let M = GL(2, Q) ∩ M2×2 (Z), the monoid of integer matrices with non-zero determinant.
We can extend the action to M y X , by the same formula. This is covered by an action
M y E defined by A(τ, z) = (Aτ, (det A)(cτ + d)−1 z). The induced map A : Eτ → EAτ is an
isogeny of degree det A.
Then
Man
Isog = (M × X )//(Γ × Γ),
by the action (B, C)(A, τ ) = (BAC −1 , Cτ ).
I’ll focus on SpecA = Y → M which have descent for pth power isogenies.
3.14. Koszul duality. Koszul duality is a particularly nice case of bar-cobar duality.
Consider a coalgebra A over k. For a comodule M , the cobar construction is a cochain
complex
C(M, A, k) ≈ M ⊗hA k.
Note this complex is naturally a module over B := C(k, A, k), which is a dga. We can soup
this up to dg-comodules M , or even assume C is a codga. Thus we have a functor of derived
categories
D(ComodA ) → D(ModB ),
Bar-cobar duality is the observation that this wants to be an equivalence, at least if you
apply some adjectives and/or replace source and target by certain core full subcategories.
Koszul dualityLis a situation where the above correspondence is particularly computable.
Suppose A = Ar is a graded coaugmented coalgebra, with A0 = k. Then the dga
B := C(k, A, k) inherits a grading B = ⊕B[r] .
B[0] : A0

B[1] : A1

B[2] : A2 / A1 ⊗ A1

A2 ⊗ A1
B[3] : A3 / + / A1 ⊗ A1 ⊗ A1
A1 ⊗ A2
We say that A is Koszul if H∗ (B[r] ) ≈ 0 for ∗ =
6 r. This implies that the dga B is equivalent
to a smaller, formal dga C with Cr = Hr (B[r] ).
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 23

3.15. Example. A = (SV )∗ , the coalgebra dual to the symmetric algebra on a finite dimensional
vector space V . Then C = Λ(V ∗ [−1]), an exterior algebra with d = 0.
An immediate payoff of the Koszul property is a the Koszul complex, a quasi-isomorphism
 
M ≈ C(M, C, A) ≈ M ⊗ C0 ⊗ A → M ⊗ C1 ⊗ A → M ⊗ C2 ⊗ A → · · ·

of dg-comodules.
Another consequence is that A is quadratic: it is “cogenerated” by A1 , the map A2 →
A1 ⊗ A1 is injective, and all relations are determined by this one inclusion. Also, the dual
algebra C is formally determined by A: thus C ≈ T (A)/hA2 i.
3.16. Descent coalgebroids are Koszul. Fix Y = SpecA0 → M with descent for pth
powerLisogenies. The coalgebra {Ar = AIsog pr } over A0 has an associated cobar complex
B= B[r] , which we call the modular isogeny complex.
3.17. Theorem ([Rez12b]). A is Koszul, i.e., H∗ (B[r] ) ≈ 0 for ∗ = 6 r. Furthermore,
• Cr = Hr (B[r] ) = 0 if r ≥ 3.
• C0 , C1 , C2 are locally free A0 -modules of ranks 1, p + 1, p respectively.
This immediately implies the claims I made last time about power operations for Morava
E-theory, in the case of height 2. `
I give an idea of the proof. Let MIsog p∗ = MIsog pr , for the associated simplicial object
M• , and consider Y• → M• with descent for pth power isogenies for Y = SpecA0 . The first
observation is that forming the modular isogeny complex only involves part of the simplicial
structure o
o
o o
Y0 Y1 o Y2 o ···

namely that which always preserves the “source”. In particular, B[∗] is a complex of A0 -
modules. This means we can basechange by any ring map A0 → R. In particular, we can
form the modular isogeny complex for SpecR → M, even if it does not admit descent by pth
power isogenies.
Each Ar is a flat and locally free A0 -module. Thus, by standard commutative algebra, it
suffices to show that
H∗ (k ⊗A0 B[r] ) ≈ 0 for ∗ =
6 r
for A0 → k where k is an algebraically closed field.
• If p−1 ∈ k, then the statement is purely combinatorial. In this case, C[pr ] ≈ (Z/pr )2 ,
so Y
(B[r] )d ≈ k.
G1 ≤···≤Gd ≤C[p∞ ]
|Gd |=pr

In fact, B[r] ≈
L
G≤C[p∞ ]
e∗−2 (PG ; k), the cochains on the “order complex” of sub-
C
|G|=pr
groups of G. The complex PG is contractible unless G is elementary abelian.
• When p = 0 in k, there are two cases: C is ordinary (C b is height 1), and C is
supersingular (C
b is height 2). I’ll describe the height 2 case, which is what is
important for Morava E-theory.
24 CHARLES REZK

Fix a supersingular curve C0 /k, and consider

SpecA0 → M
where A0 = k[[a]] is the ring which classifies deformations of C0 to rings of characteristic
p. (Thus, A0 = π0 ECb0 /p, the mod p-reduction of the object of the last talk.) This
has descent for pth power isogenies, so we obtain an associated modular complex B[r] .
We prove the Koszul property for this, then specialize to B[r] ⊗A0 k to get the result
for C.
In fact, Katz-Mazur describe these rings explicitly. We have
A1 ≈ k[[a, a0 ]]/((ap − a0 )(a − a0p )), a0 = ψ1 (a).
This is an avatar of the “classical” Kronecker congruence for the modular equation of
degree p isogenies. More generally,
i j
Y
Ar ≈ k[[a, a0 ]]/( (ap − a0p )).
i+j=r

Thus, we have complete explicit control of B[r] ; proving the vanishing of H ∗ is a

calculation.
3.18. The building picture. A version of the above was conjectured by Ando-Hopkins-
Strickland. Let me explain their picture.
Let M• be the simplicial object encoding pth power isogenies. The idea is that the
category Mod(M• ) of pth power isogeny modules is entirely determined by a subobject of
M• associated to isogenies whose kernel is killed by p.
That is, M• contains a sub-semisimplicial object
MIsog p o
o
 
id
N• := Mo
(s,t)
q o s MIsog p
(id,Ψ) o w
M
Here MIsog p q M ⊂ MIsog p q MIsog p2 ⊂ M1 , where the second term classifies isogenies
f : C → C 0 with ker f = C[p]. Note that any such f is canonically isomorphic to [p] : C → C.
f g
− C0 →
Likewise MIsog p ⊂ MIsog p ×M MIsog p ⊂ M2 classifies composites C → − C 00 with
f g f g0
− C0 →
ker gf = C[p]. Given f , the sequence C → − C 00 is canonically isomorphic to C →
− C0 −→
C, by the same argument as above.
The content of the Koszul theorem is roughly that this semi-simplicial object “generates”
M• . If SpecA → M has descent for pth power isogenies, then an isogeny module M for this
is exactly: an A-module, together with A-module maps
s t Ψ
ψ1 : M → M ⊗A A1 , φ : M → M ⊗A A
such that
ψ1
M / M ⊗A A1
s t

φ ((id ×w)⊗id)(ψ1 ⊗id)

 
M ⊗A A
Ψ / M ⊗A A1
s sΨ
id ⊗s

Here w : A1 → A1 is the map classifying dual isogenies, as in the previous lecture.

 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 25

3.19. Koszul property for power operations for Morava E-theory. The general case
requires a different argument, which relies on topology, not algebraic geometry.
Let Σm y X be a finite Σm -set. We can consider E 0 (XhΣm ), the E-cohomology of the
homotopy orbit space.
Let (2m − 2) ⊂ 2m be the subset of the power set of an m-element set with ∅ and m
removed. Set
 
0 m Transfer 0
Qm (X) := Cok E (X × (2 − 2))hΣm −−−−→ E (XhΣm )

In fact,
Qpr (∗) ≈ E 0 (BΣpr )/Itransfer = Ar .
Observe that Qm is actually a Mackey functor for Σm .
The Koszul result follows from the following two observations.
• The modular isogeny complex B[r] is isomorphic to
B[r] = Q(P pr ).
Here P pr is the reduced partition complex. Thus
H ∗ B[r] = HBre
∗
(P pr ; Q).
• Arone-Dwyer-Lesh [ADL16] show that for suitable Mackey functors Q (including this
∗
one), HBre (P pr ; Q) ≈ 0 for ∗ =
6 r.
3.20. Example: The Behrens Q(`) spectrum. Let ` be a prime, and consider M[ 1` ], the
stack which classfies elliptic curves C → S over base schemes S which lie over SpecZ[ 1` ].
In this case, each MIsog `r [ 1` ] → M is étale, and in fact all maps between stacks in M• [ 1` ]
are étale. Therefore, the general machinery of Goerss-Hopkins-Miller applies. We obtain a
cosimplicial commutative S-algebra
[n] 7→ Γ(Mn [ 1` ], OTop ),
whose inverse limit is the spectrum Q(`) constructed by Mark Behrens [Beh06], [Beh07],
[BL06].
The building picture says that you can construct Q(`) as the inverse limit of a semi-
cosimplicial ring
TMFIsog p [ 1` ] /
/
1
TMF[ ` ] × / TMFIsog p [ 1 ]
/
/ `
TMF[ 1` ]
(In fact, you can build this with TMF replaced by Tmf, etc., by Hill-Lawson.) We can
K(2)-localize at a prime other than p. Behrens’ conjecture is that for p odd and ` chosen
suitably, there is a cofiber sequence
DQ(`)K(2),p → SK(2),p → Q(`)K(2),p ,
and this is proved for p = 3 and ` = 2.
26 CHARLES REZK

3.21. Example: Φh S 2d−1 . There is no construction of Q(p) at the prime p, i.e., without
inverting p.
However, there turns out to be something that seems to play its role. This is Φh S 2d+1 , the
Bousfield-Kuhn functor applied to an odd sphere.
Fix a Morava E-theory of height h associated to G0 /k. Let C denote the category of
E∗ -modules equipped with power operations. E.g., if G0 is the formal completion of a s.s.
curve, then C is the category of pth power isogeny modules (except that things are allowed
to have odd grading).
Work of Behrens and me (in progress), shows that there is a spectral sequence
E2s,t = ExtsC (ω d−1 , ω (t−1)/2 ⊗ nul) ⇒ Et−s
∧
Φh S 2d−1 .
e0S 2.
Here nul = E0 with coaction map ψ1 ≡ 0, while ω = E
4. Multiplicative orientation
Let’s calculate power operations in an important example.
4.1. BV. Let E be an even periodic ring theory. Thus E 0 BU (1) ≈ E 0 [[x]]. The choice of a
generator x is called a coordinate. A choice of class determines a notion of Euler class for
line bundles:
(L → X) x(L) ∈ E 0 X
defined by
E 0 (L)
X−
L
→ BU (1) E 0 BU (1) −−−→ E 0 X
x 7→ x(L)
Note that x = x(Luniv ) where Luniv → BU (1) is the universal line bundle.
The associated formal group law x1 +F x2 ∈ E 0 [[x1 , x2 ]] is defined so that x(L1 ⊗ L2 ) =
x(L1 ) +F x(L2 ); the group law depends on x.
Let a
BV = BU (n),
the classifying space for complex vector bundles. This space has a multiplication BV × BV →
BV encoding direct sum of bundles. We can identify the ring E∗ BV by
E∗ BV
O
E∗ [bk , k ≥ 0]
O
E∗ (Luniv )

E∗ BU (1) E∗ {bk , k ≥ 0}
where Luniv : BU (1) → BV classifies the universal line bundle. The E∗ -module basis {bk } of
E∗ BU (1) is defined to be that which is dual to the “monomial basis” {xk } of E ∗ BU (1) = E∗ [ x]].
(More precisely, the basis b0 , . . . , bn of E∗ CPn is dual to the monomial basis of E ∗ CPn for
each n.)
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 27

4.2. The spectrum F = E ∧Σ∞ ∞

+ BV. Let F := E ∧Σ+ BV. This is a homotopy commutative
ring spectrum.
Given a vector bundle V → X, there is a characteristic class
hV i ∈ F 0 X,
defined tautologously: if V : X → BV is the map representing the bundle, then hV i is the
composite
Σ∞
+V 1∧id
Σ∞+X − −−→ Σ∞ ∞
+ BV −−→ E ∧ Σ+ BV = F.
This class satisfies the whitney sum formula
hV ⊕ W i = hV ihW i.
4.3. Example. Consider Luniv → BU (1). You can calculate
X
hLuniv i = bk xk ∈ F 0 BU (1)
k≥0

Here x is the image of x ∈ E BU (1) → F 0 BU (1). Note that the right-hand side appears to
0

depend on the choice of x (which determines the bk as well), but the left-hand side does not.
The element γuniv := hLuniv i is the universal example of a function on the formal group
φ
GE of E. That is, given φ : E0 → R and an element γ ∈ Oφ∗ GE = E 0 BU (1) ⊗E0 R, there
exists a unique ring homomorphism F0 → R exending φ so that
F0 BU (1) = E 0 BU (1) ⊗E0 F0 → E 0 BU (1) ⊗E0 R sends γuniv 7→ γ.
In terms of a coordinate x, if γ(x) = ck xk , then F0 → R sends bk 7→ ck .
P

4.4. Remark. The diagonal map ∆ : BV → BV × BV makes E0 BV into a Hopf algebra. What
does the comultiplication represent? The calculation of ∆∗ happens on E0 BU (1), and we
have ∆∗ (bk ) = i+j=k bi ⊗ bj (it is dual to the cup product on E 0 BU (1)). This means that
P
∆∗ represents multiplication of functions
(γ1 , γ2 ) 7→ γ1 γ2 : Oφ∗ G × Oφ∗ G → Oφ∗ G .
Note that if L → X is any line bundle, then
hLi = γuniv (x(L)),
and more generally
hL1 ⊕ · · · ⊕ Lk i = γuniv (x(L1 )) · · · γuniv (x(Lk ))
for a sum of line bundles. You can use the “splitting principle” to compute hV i for any
bundle. The class hV i is a kind of universal characteristic class taking sums to products.
4.5. Power operations for F . The space BV is actually an E∞ -space. This implies that
Σ∞+ BV is an E∞ -ring spectrum, and so can be realized as a commutative S-algebra.
Suppose E is also a commutative S-algebra. Then so is F . I would like to compute power
operations for F , e.g., the map
Pm : F 0 X → F 0 (X × BΣm ).
This is too difficult to make sense of for arbitrary elements of F 0 X. However, there is a
formula for classes of the form hV i.
28 CHARLES REZK

4.6. Proposition. We have

Pm (hV i) = hV  ρm i.
m
Here we write ρm = (C x Σm ) for the permutation representation of Σm , and also for the
induced bundle Cm ×Σm EΣm → BΣm .
Proof. The point is that the class hV i is produced by a map of spaces. Since BV is an
E∞ -space, we have a power construction
diag
X × BΣm −−→ X m ×Σm EΣm → BV m ×Σm EΣm → BV
The induced bundle is V ⊕m ×Σm EΣm → BΣm , which is what we have called V  ρm . Now
use commutativity of
Σ∞ / Σ∞ m / Σ∞ m / ∧m
+ (X × BΣm ) + (XhΣm ) + (BVhΣm ) FhΣm

V ρm
+  
Σ∞ /
+ BV F

To understand what this formula says, it is useful to restrict from the symmetric group
to certain abelian subgroups. Consider A ≤ Σm which is abelian and transitive, i.e., the
induced action A y m is transitive; this implies |A| = m. One such example is a cyclic group
Cm ≤ Σm .
The restriction of the permutation representation along A ⊆ Σm is a regular representation:
M
ρm |A ≈ λ.
λ∈A
b

b = Hom(A, C× ). Thus, the restriction of Pm (hV i) to F 0 (X × BA) is

Here A
X
Y
hV  λ i= hV  λi.
λ∈A
b λ∈A
b

I’m using λ as notation for the associated line bundle over BA. If we let V = Luniv → BU (1),
and choose a coordinate x ∈ E 0 BU (1), this becomes
Y Y Y
(4.7) hLuniv  λi = γuniv (x(Luniv  λ)) = γuniv (x +F x(λ)).
λ∈A
b λ∈A
b λ∈A
b

4.8. The case of Morava E-theory. Now assume that E = EG0 /k is a Morava E-theory
for a height h formal group. To plug in the theory of power operations described in lecture 2,
we need to consider
Fb := LK(h) F, Fb0 ≈ (F0 )∧mE .
This ring still represents functions on the formal group, as long as we use ring homomorphisms
which are continuous wrt the mE -adic topology.
We want to describe the induced descent-for-isogenies structure on SpecF0 . This ring
represents the functor
(G/R, i, α) 7→ OG .
Descent for isogenies means that for every isogeny f : G → G0 between deformations of G0 to
R, there is an induced pushforward map
f! : OG → OG0 .
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 29

One thing we can say about this is that f! is multiplicative: f! (γ1 γ2 ) = f! (γ1 )f! (γ2 ), because
the product of functions is represented by the diagonal map, which is a map of E∞ -spaces.
Also, the “Frobenius congruence” implies that for a pth power Frobenius F : G → σ ∗ G, we
must have F! (γ) = γ p .
An isogeny f : G → G0 induces a map of function rings
f ∗ : OG0 → OG .
Because f is an isogeny of degree pr , this map presents OG as a free OG0 -module of rank pr .
4.9. Theorem ([AHS04]). The function f! : OG → OG0 is the multiplicative norm. I.e., f! (γ)
is determinant of γ· : OG → OG as a map of OG0 -modules.
(1) It is enough to compute this for the universal example f : s∗ Guniv → t∗ Guniv of
an isogeny of degree pr . The universal example s∗ γuniv of a function on s∗ Guniv is
s
represented by the identity map of Fb0 ⊗E0 Ar , and the desired function f! (s∗ γuniv ) on
t∗ Guniv is represented by the power operation
ψr s
Fb0 −→ Fb0 ⊗E0 Ar .
(2) There is a tautological commutative diagram
represents f! (γuniv )
Fb0 / s
Fb0 ⊗E0 A1
(4.10)
 
Fb0 BU (1) / s
Fb0 BU (1) ⊗E0 Ar
γuniv 7→ψr (γuniv )

s
Note that Fb0 BU (1) ⊗E0 Ar ≈ Os∗ Guniv . We claim that
ψr (γuniv ) = f ∗ f! (γuniv ).
This is sufficient to read off f! (γuniv ), since f ∗ : Ot∗ Guniv → Os∗ Guniv is injective.
Here’s a proof. Given a fixed E0 -algebra R which is complete with respect to some
ideal, classifying a deformation G, write
G(R) := Hom(E 0 BU (1), R)
for the set of all continuous homomorphisms (using the mE +(x)-topology on E 0 BU (1).)
We think of these as R-valued points of G. If γ ∈ E 0 BU (1) is a function on GE ,
we write “γ(p)” for the image of γ under p (i.e., γ(p) := p(γ) (!)). Note that if
x ∈ E 0 BU (1) is a coordinate, then
x(p1 + p2 ) = x(p1 ) +F x(p2 ).
More generally, p ∈ G(R), then the map
p
E 0 BU (1)⊗
b E0 Fb0 = Fb0 BU (1) →
− R
classifying data (G, γ ∈ OG , p ∈ G(R)) sends γuniv 7→ γ(p).
s
Fix a homomorphism φ : Fb0 BU (1) ⊗E0 Ar → R, i.e., a map
b Fb0 ⊗E0 s Ar → R.
E 0 BU (1)⊗
This classifies data
(f : G → G0 , γ ∈ OG , p ∈ G(R)),
30 CHARLES REZK

where f is an isogeny between deformations, γ a function on G, and p is a R-point in

G.
Now consider the composite
ψr φ
φ0 : E 0 BU (1)⊗ b Fb0 ⊗ s Ar →
→ E 0 BU (1)⊗
b Fb0 − − R.
This classifies data
(γ 0 ∈ OG0 , f (p) ∈ G0 (R)),
and in fact γ 0 = f! (γ) by commutativity of (4.10).
Now let’s follow the universal function γuniv through this map:
γuniv 7→ ψr (γuniv ) 7→ φ(ψr (γuniv )) = γ 0 (f (p)) = f! (γ)(f (p)).
Taking φ to be the identity map, we see that ψr (γuniv ) = f ∗ f! (γuniv ), as desired.
(3) Now we need to relate ψr (γuniv ) to the multiplicative norm of γ along f . Let A ⊆ Σpr
be a transitive abelian subgroup. There is a commutative diagram
E 0 BΣpr / E 0 BA

 
E 0 BΣpr /I / E 0 BA/I 0
where I 0 is the ideal generated by transfers from proper subgroups of A. It turns out
(see [Str98]) that the restriction map
Y
E 0 BΣpr /I → E 0 BA/I 0
is injective, where the product is over transitive abelian subgroups A. Thus, to compute
s
ψr (γuniv ), it it is enough to compute its projections to Fb0 BU (1) ⊗E0 E 0 BA/I 0 .
s
The element ψr (γuniv ) is itself the image of Ppr (hLuniv i) ∈ Fb0 BU (1) ⊗E0 E 0 BΣpr .
By (4.7), we know that
Y
Pm (hLuniv i)|BA = γuniv (x +F x(λ))
λ∈A
b

in terms of a coordinate x ∈ E 0 BU (1).

λ f
On the other hand, given an exact sequence 0 → A b→− G→ − G0 → 0 of “physical”
groups, where A is finite, the multiplicative norm can be computed by
Y Y
f ∗ Nf (γ(g)) = (Trans∗λ(a) γ)(g) = γ(g +G g(a)).
a∈A a∈A
Q
I.e., if you use the basis of OG over OG0 = A OG obtained by picking a set theoretic
section of G → G0 , diagonal matrix.
Such a “physical” exact sequence exists over a fraction field of the quotient ring
E 0 BA/I 0 . The tautological function
b → G(E 0 BA), E 0 Bλ
`: A λ 7→ (E 0 BU (1) −−−→ E 0 BA),
is a homomorphism of groups, which is universal for homomorphisms from A:
{A
b → G(R)} ↔ {defs. (G, i, α) to R, hom. ` : Ab → G(R) }.
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 31

Over E 0 BA/I 0 , the function ` is an example of a Drinfel’d level structure on

GE 0 BA/I 0 .
Pulled back to the ring S −1 (E 0 BA/I 0 ) obtained by inverting euler classes of all
non-trivial line bundles, ` gives an isomorphism between A and the kernel of the
universal isogeny (which now must be considered not as a map of formal groups,
but as a map of p-divisible groups). It turns out that E 0 BA/I 0 → S −1 (E 0 BA/I 0 ) is
injective, so you can read off the desired formula for the multiplicative norm in this
case. (See [AHS04] for the details about this argument.)
4.11. Some friends of BV. This norm interpretation of power operations on BV comes
from Ando’s thesis [And95]. It leads to similar interpretations for friends of BV. The following
diagram shows (i) some E∞ -spaces or spectra R, (ii) the corresponding functor represented
by E0 R, and (iii) the “universal element” for the representable functor.
BU MU

 
/ /
a _
BV = BU (m) Z × BU MV = M U (m) MUP

×
OG /R× Trivrigid (I)
O O

OG o ×
OG Io Triv(I)

1−L
(BU (1) −−→ BU ) (BU (1)1−L → M U )

L L
(BU (1) −
→ BV) (BU (1) −
→ Z × BU ) (BU (1)L → M V) (BU (1)L → M U P )
Here I ⊂ OG (e) is the augmentation ideal, i.e., the functions on G which vanish at the
identity element. A trivializaton of the ideal is a choice of generator; a rigid trivialization is
D
a section of the projection I −→ ωG (e).
Let x ∈ E 0 BU (1) be a coordinate for Guniv . This gives rise, for any deformation (G, i, α)
of G0 to R, a coordinate xG,i,α ∈ OG (by E 0 BU (1) → E 0 BU (1) ⊗E0 R).
4.12. Theorem (Ando [And95]). A necessary condition for the coordinate x to come from a
map M U P → E of commutative S-algebras, is that
Nf (xG,i,α ) = xG0 ,i0 ,α0
for any isogeny f : G → G0 compatible with deformation structures.
(To check this condition, it suffices to show it for the universal p-isogeny over A1 .)
Matt Ando proved that such coordinates exist, at least for EG0 where G0 satisfies F h = p.
32 CHARLES REZK

There is an extension to connected covers of BU [AHS04].

(1−L1 )(1−L2 )(1−L3 )
BU h6i Θ3 (G, OG ) (BU (1)∧3 −−−−−−−−−−−−→ BU h6i)
O

 (1−L1 )(1−L2 )
BSU Θ2 (G, OG ) (BU (1)∧2 −−−−−−−−→ BSU )
O

 1−L
0 × BU Θ1 (G, OG ) (BU (1) −−→ BU )
The set Θk (G, OG ) is the set of functions f on Gk such that
• f (0, . . . , 0) = 1,
• f is symmetric,
• f (a + b, c, . . . )f (a, b, . . . ) = f (a, b + c, . . . )f (a, b, . . . ).
The sequence runs out here, since these are the only covers of BU with even cohomology.
Similar results hold for M SU and M U h6i, with OG replaced by I. The action of power
operations on these is described by a norm formula.
An element of Θ3 (G, I) is called a cubical structure. Any elliptic curve C has a unique
cubical structure, which therefore prescribes a preferred cubical structure on its formal
completion C. b Therefore, the above correspondence picks out, for any elliptic spectrum, a
unique map M U h6i → E of ring spectra. Furthermore, the norm of a cubical structure is
another one, so if E is an elliptic spectrum, then the above map is H∞ .
Finally, the unique cubical structure can be described using the Weierstrass σ-function,
which is a particular choice of section of Θ1 (C, I) when C is the Tate curve: the cubical
structure is
σ(0)σ(a + b)σ(a + c)σ(b + c)
s(a, b, c) = .
σ(a)σ(b)σ(c)σ(a + b + c)
This is partial progress towards constructing
M U h6i / EO


M String / tmf
as a map of commutative S-algebras.
4.13. The string orientation. The eventual construction of the S-algebra map M String →
tmf is very different in detail.
Given a commutative S-algebra, there is an associated units spectrum gl1 (R), whose
underlying space is GL1 (R). Its delooping BGL1 (R) classifies stable spherical fibrations.
There is an adjoint pair
Σ∞ ∞
+ Ω : ((−1)-connected spectra)  (comm S-algebras) : gl1 ,
analogous to Z[−] : (ab gps)  (comm rings) : (−)× .
Let o → gl1 (S) be the J-homomorphism. There is a correspondence between null-
homotopies of the composite
g → o → gl1 (S) → gl1 (R)
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 33

and the space of commutative S-algebra maps M G → R, as torsors over Map(g, gl1 (R)) ≈
Map(Σ∞ + BG, R).
Thus, to construct orienations, we need to understand the homotopy type of gl1 (R).
For complex oriented E, we have
{(maps of ring spectra M U → E)} ↔ {coordinates x ∈ E 0 BU (1)}.
To describe this set, we can use the following formalism. In the rationalization EQ of E, there
is a standard coordinate xHQ coming from HQ → EQ. Thus, associated to f : M U → E
is a coordinate xf ∈ E 0 BU (1), and we can write Kf (xHQ ) = xHQ /xf . This is precisely the
Hirzebruch characteristic series, corresponding to a map κf : BU → GL1 (EQ). This map
measures the “difference” between the two sides of the non-commuting square
f
MU / E
g
 
HQ / EQ
Write
X xk

Kf (x) = exp tk .
k≥1
k!

4.14. Proposition. In degree 2k, κf : π2k BU → π2k (EQ) = π2k E⊗Q sends the Bott generator
to (−1)k tk .
This describes a function
Y
Hom(M U, E) → π2k E ⊗ Q.
We ask the question: which elements in the image come from commutative S-algebra maps?
Likewise, there is a function
Y
Hom(Σ∞ + BU, E) → π2k E,
which associates f : Σ∞ 0
+ BU → E, corresponding to f ∈ E BU (1) with f (0) = 1, to the
sequence tk defined by
X xk
f (xHQ ) = exp( tk ).
k!
4.15. Example. The Todd genus is given by the characteristic series
x X Bk xk
KTd (x) = = exp(− ).
1 − e−x k≥1
k k!
The only bernoulli number with k odd is B1 = 1/2. Thus we can modify this to a formula
x X Bk xk
KAb(x) = x/2 = exp(− ).
e − ex/2 k≥2
k k!

4.16. Example. The Witten genus corresponds to

X xk
KW (x) = exp(−2 Gk ),
k≥4
k!
34 CHARLES REZK

where
B2k X X 2k−1 n
G2k (q) = −+ d q .
4k
The appearance of bernoulli numbers is not accidental. We can consider the universal case
of the orientation 1 : M U → M U .
A good toy example for this kind of calculation is maps from Σ∞+ BU to K.

4.17. Theorem ([Wal09]). The image of

Y
HomS−alg (Σ∞
+ BU, K) ≈ HomΩ∞ (BU, gl1 K) → π2k K
is the collection of sequences (tk )k≥1 of integers such that for each prime p, the sequence
t∗k,p := tk (1 − pk−1 ) is a set of moments for a p-adic measure µp on Z×
p . That is,
Z
t∗k,p = xk dµp (x).
Z×
p

This turns out to be the same as the set of H∞ -maps BU → K, which corresponds to
×
elements f ∈ (1 + xZ[[x]]) ⊂ OG m
which are compatible with norms after completing at all
primes p.
5. Logarithm
5.1. Units spectrum. Given a commutative S-algebra R, let GL1 (R) be the pullback
GL1 (R) / Ω∞ R

 
(π0 R)× / π0 R
It represents the functor
(R0 (X))× = [X, GL1 (R)]
The space GL1 (R) is an infinite loop space. Its corresponding connective spectrum is denoted
gl1 R.
Problem: say something about the homotopy type of the spectrum gl1 R.
5.2. The idea. Let E and F be spectra. Consider
[F, E]Sp → [Ω∞ F, Ω∞ E]H ⊆ [Ω∞ F, Ω∞ E]Top∗ ,
induced by the functor Ω∞ : Sp → Top∗ . The image lands in the set of H-space maps.
In general, there is no expectation that this map be either surjective or injective.
5.3. Rational linearization. Let us suppose
F is 0-conn., E = EQ , i.e., π∗ E ≈ π∗ E ⊗ Q.
5.4. Proposition. In this case, the map Ω∞ : [F, E]Sp → [Ω∞ F, Ω∞ E]Top∗ admits a retraction.
Here is the construction of the retraction. Recall that since E = EQ ,
∼
[F, E]Sp −
→ Hom(π∗ F, π∗ E),
Σn H(πn E).
Q
because E ≈
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 35

Define r : [Ω∞ F, Ω∞ E]Top∗ → [F, E]Sp by

∼
[Ω∞ F, Ω∞ E]Top∗ → Hom(π∗ Ω∞ F, π∗ Ω∞ E) ≈ Hom(π∗ F, π∗ E) ←
− [F, E]Sp .
This works because Ω∞ F is connected, so the induced map π∗ Ω∞ F → π∗ Ω∞ E is a homo-
morphism of groups.
You can also show that all H-space maps are infinite loop maps:
∼
→ [Ω∞ F, Ω∞ E]H .
[F, E]Sp −
We can apply this to GL1 (R) when R is rational.
5.5. Rational linearization, again. Let me do this again, but more complicated.
Write LQ : [Ω∞ F, Ω∞ E] → [Ω∞ F, Ω∞ E] for the idempotent implicitly defined by the
retraction above. Given f ∈ [Ω∞ F, Ω∞ E], we can regard it as defining a cohomology
operation
f : F 0 (X) → E 0 (X).
The problem is to calculate the cohomology operation Lf given f .
5.6. Proposition. For X a finite dimensional connected CW-complex,
X (−1)n−1
(5.7) (LQ f )(x) = Crn f (x, . . . , x).
n≥1
n

Here Crn is the nth cross-effect. Given any function f : A → B between abelian groups,
define functions Crn f : A×n → B by
Cr1 f (x) := f (x) − f (0),
Cr2 f (x1 , x2 ) := f (x1 + x2 ) − f (x1 ) − f (x2 ) + f (0),
···
X X
Crn f (x1 , . . . , xn ) := (−1)n−|I| f ( xi ).
I⊆n i∈I

Note that for any f : Ω∞ F → Ω∞ E, the operation Crn f : (Ω∞ F )×n → Ω∞ E factors
through the quotient map (Ω∞ F )×n → (Ω∞ F )∧n . This means that for n > dim X, we have
Crn f (x, . . . , x) = 0, so the sum is finite.
Proof. If B = B ⊗ Q, then
Map(A, B) ≈ Hom(Q[A], B)
by the correspondence f ↔ fe given by
f
A /
=B
δ : a7→[a]
 fe
Q[A]

If Crn f = 0, then we can factor fe through Q[A]/I n , where I = Ker(Q[A] → Q).

If we formally define L by the right-hand side of (5.7), then Lf = fe(Lδ). We compute
X (−1)n−1 X (−1)n−1
(Lδ)(a) = Crn δ(a, . . . , a) = ([a] − 1)n ,
n n
36 CHARLES REZK

and it is clear that Lδ : A → B is a group homomorphism. Applied to A = F 0 (−) and

B = E 0 (−), we get an additive cohomology operation. If X = S k , then the terms n ≥ 2
vanish in (5.7), so we see that Lf has the same effect on homotopy groups as f .

Symm
5.8. Exercise. Consider BU −−−→ Ω∞ K → Ω∞ KQ , where Symm is the operation correspond-
ing toPmth symmetric power of vector bundles. Then LQ Symm = m1 ψ m . (Hint: consider
St = Symm tm : BU → K[[t]] first.)
5.9. Rational logarithm. We can apply the above if R = RQ, to the map
x7→x−1
s : GL1 (R)≥1 −−−−→ Ω∞ R,
where GL1 (R)≥1 is the basepoint component of GL1 (R). For a connected X, we get a natural
homomorphism
LQ s : (R0 X)× → R0 X,
and we compute
X (−1)n−1
(LQ s)(x) = (x − 1)n = log x.
n
5.10. K-theory. Let’s consider the case of E = F = Kp , following Madsen-Snaith-Tornehave
[MST77].
Ω∞
[Kp , Kp ] −−→ [Ω∞ Kp , Ω∞ Kp ]H ⊂ [Ω∞ Kp , Ω∞ Kp ].
This map Ω∞ is injective. In fact,
[Ω∞ Kp , Ω∞ Kp ]H ≈ Zp [[Zp ]] ≈ lim Zp [Zp /pn ],
with generators [λ] corresponding to the Adams operation ψ λ . And
[Kp , Kp ] ≈ Zp [[Z×
p ]],

the map induced by the evident inclusion Z×

p → Zp .

5.11. The idempotent. In fact, we can identify the image of Ω∞ in terms of an idempotent
E on the set [Ω∞ Kp , Ω∞ Kp ]H . Here are several recipes.
∼
• Recall the Bott periodicity map β : Kp − → Ω2 Kp . Given f : Kp → Kp , define ωf by
ωf
Kp / Kp
β ∼ ∼ β
 
Ω2 Kp / Ω2 Kp
Ω2 f

Compute that
ω(ψ λ ) = λ ψ λ .
Now consider
k
Ef := lim ω (p−1)p (f ).
k→∞
That this converges and gives the right result amounts to the fact that, p-adically,
k
λ 7→ limk→∞ λ(p−1)p is the characteristic function of Z×
p.
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 37

• Here’s what MST do. Recall that associated to any finite covering map g : Y → X
is an associated transfer map Σ∞ ∞
+ X → Σ+ Y , which induces an (additive) transfer
g! : E ∗ X → E ∗ Y in any cohomology theory.
Any infinite loop map commutes with transfers. MST prove:
5.12. Proposition. f ∈ [Ω∞ Kp , Ω∞ Kp ]H is infinite loop iff
f
Kp0 (X × ECp ) / Kp0 (X × ECp )
τp τp
 
Kp0 (X × BCp ) / Kp0 (X × BCp )
f

commutes, where τp is the transfer associated to the covering map ECp → BCp .
Proof. We have
Kp0 (X × BCp ) ≈ Kp0 (X)[T ]/(T p − 1),
where T = line bundle on BCp associated to the representation Cp ⊂ C× . Furthermore,
the transfer in K-theory is given by τp (x) = x  N , where N = 1 + T + · · · + T p−1 .
The proof follows from the calculation (for integers λ)
(
p if p|λ,
ψ λ (N ) =
N if p - λ.
Since
τp ψ λ (x) = ψ λ (x)  N, ψ λ τp (x) = ψ λ (x)  ψ λ (N ),
we see that the image of Ω∞ : [Kp , Kp ] → [Ω∞ Kp , Ω∞ Kp ]H (i.e., the subset Zp [[Z×
p ]] ⊂
Zp [[Zp ]]) is exactly as asserted. 
We can turn this into a formula for a projection operator E on [Ω∞ Kp , Ω∞ Kp ]H , by
1
(Ef )(x) = f (x) − hf (τp (x)), ci.
p
0 0
Here h−, ci : Kp (X × BCp ) → Kp (X) ⊗ Z(ζp ) is a ring homomorphism defined by T 7→ ζp =
e2πi/p . (You can think of it as pairing with a homology class c ∈ K0∧ (BCp ) ⊗ C.) The point
is that
hN, ci = 0, h1, ci = 1.
5.13. Application to GL1 (Kp ). You can apply this to
Ω∞
[gl1 (Kp ), Kp ] −−→ [GL1 (Kp ), Ω∞ Kp ]H ⊂ [GL1 (Kp ), Ω∞ Kp ].
This is because of the theorem of Adams-Priddy [AP76]: there exists an equivalence
gl1 (Kp )≥4 ≈ (Kp )≥4
of spectra. The Adams-Priddy is non-constructive: they show that if Ω∞ F has the same
homotopy groups and k-invariants as BSUp , then F has the same homotopy type as (Kp )≥4 .
In fact, there is a splitting gl1 (Kp ) ≈ Z × (Kp )≥4 . As a consequence, you can show that
the image of Ω∞ is exactly the set of H-maps f : GL1 (Kp ) → Ω∞ Kp which commute with
transfer.
38 CHARLES REZK

Now, the infinite loop space structure on GL1 (K) comes from the multiplicative structure.
Thus, a map X → GL1 (K) has an ECp → BCp transfer map
X × BCp → X p ×Cp ECp → GL1 (K)p ×Cp ECp → GL1 (K)
which coincides with the power operation Pp on Kp0 (X).
5.14. Proposition ([MST77]). An H-space map f : GL1 (Kp ) → Ω∞ Kp is infinite loop if and
only if
f
Kp0 (X × ECp )× / Kp0 (X × ECp )
Pp τp
 
Kp0 (X × BCp )× / Kp0 (X × BCp )
f
commutes.
5.15. tom Dieck’s logarithm. Recall that the total power operation
Pp : Kp0 (X) → Kp0 (X × BCp ) ≈ Kp0 (X)[T ]/(T p − 1)
has the form
Pp (x) = ψ p (x) − θp (x)N,
where ψ p is the Adams operation, and N = 1 + T + · · · + T p−1 is the regular representation
of Cp . Furthermore, forgetting about the Cp -action, which amounts to setting T → 1, gives
the identity ψ p (x) − pθp (x) = xp .
5.16. Theorem (tom Dieck [tD89]). There is a spectrum map gl1 (Kp ) → Kp inducing a
cohomology operation ` : Kp0 (X)× → Kp0 (X) described by
1 xp X (−1)m pm−1
`(x) = log p = (θp (x)/x)m .
p ψ (x) m≥1 m

Proof. Verify that `Pp = τp `. We can compute

1 (Pp (x))p
`Pp (x) = log p
p ψ (Pp (x))
1 (ψ p (x) − θp (x)N )p
= log p p
p ψ ψ (x) − ψ p θp (x)p
p2
(
1
p
log ψpx(xp ) = p`(x) under T 7→ 0,
= 1 (ψ p (x))p
p
log ψ p (xp )
=0 under T →
7 ζp .
The same is true for τp `(x) = `(x)N . 
5.17. Corollary. The map ` : gl1 (Kp ) → Kp giving the tom Dieck logarithm is an equivalence
on 3-connected covers.
Proof. Compute its effect on π2n by computing it on K 0 (S 2n ) = Z[]/(2 ). We get
1 (1 + )p 1 1 + p
`(1 + ) = log p = log = (1 − pn−1 ).
p ψ (1 + ) p 1 + pn 

 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 39

So tom Dieck’s logarithm gives a “canonical” example of the equivalence gl1 (Kp )≥4 ≈ (Kp )≥4
proved by Adams-Priddy.
5.18. Where does this come from? Recall that we obtained a rational logarithm by
“linearizing” the map x 7→ x − 1 : GL1 (R) → R for a rational R.
The analogue of this can be done much more generally.
5.19. Proposition. Let F be any spectrum, then
Ω∞ : [F, Kp ] → [Ω∞ F, Ω∞ Kp ]
is injective, with image equal to the image of an idempotent E on [Ω∞ F, Ω∞ Kp ], computed
(on finite dimensonal X) by the formula
X (−1)m−1  1

∗ ∗ ×m
(Ef )(x) = Crm f (x, . . . , x) − hCrm f (π1 τp (x), . . . , πm τp (x)), c i .
m≥1
m p
Here πi : X × BCp×m → X × BCp is projection to the ith factor, and
h−, c×m i : Kp0 (X × BCp×m ) → Kp0 (X) ⊗ Zp [ζp ]
is “character evaluation” at the element (c, . . . , c) ∈ Cpm .
If f is an H-map, then Crm f = 0 for m ≥ 2, so this simplifies to the formula f (x) −
1
p
hf (τp (x)), ci.
Now apply this to the shift map s : GL1 (Kp ) → Ω∞ Kp defined by s(x) = x − 1.
5.20. Corollary. The map Es : GL1 (Kp ) → Ω∞ Kp is the one given by tom Dieck’s formula.
Proof. Explicitly, the formula reduces to
X (−1)m−1 1 1
(x − 1)m − (ψ p (x) − 1)m = log x − log ψ p (x).


(Es)(x) =
m p p

It turns out we can replace Kp with many other commutative S-algebras, which are
“K(n)-local”. The idea is (to some extent) suggested by the “Bott peroidicity” construction
of the idempotent for K-theory.
5.21. Bousfield-Kuhn functor. A finite CW-complex V is type n if K(n)∗ V = 6 0 but
K(n − 1)∗ V = 0.
The periodicity theorem of Hopkins-Smith says that, for any type n finite V , there exists a
k
(stable) map f : Σd V → V with d = 2(pn − 1)pk > 0 so that K(n)∗ f is multiplication by vnp .
Such a map is called a vn -self map.
Given a finite CW-complex V and a map f : Σd V → V with d > 0, we can define
ΦV,f : Top∗ → Sp
by sending a space X to the spectrum E = {E k } with
E kd := Map∗ (V, X), k ≥ 0,
with structure map E kd → Ωd E kd+d given by
◦f
Map∗ (V, X) −→ Map∗ (Σd V, X) ≈ Ωd Map∗ (V, X).
40 CHARLES REZK

Since E = hocolim Σ−kd Σ∞ E kd , we can write this as

ΦV,f X ≈ hocolimk Σ−kd Σ∞ Map∗ (V, X).
Observe that if X = Ω∞ F , then we can write this as
ΦV,f Ω∞ F ≈ hocolimk Σ−kd Hom(Σ∞ V, F ) ≈ (Σ∞ f )−1 Hom(Σ∞ V, F ) =: TelV,f F.
In other words, we have a factorization
TelV,f
Sp / Sp
<
Ω∞
 ΦV,f
Top∗
The localization functor LK(n) : Sp → Sp factors through an inverse limit of telescopes:
LK(n) = LK(n) holimi Σ−qi TelVi ,fi . Bousfield and Kuhn show ([Bou87], [Kuh89]) that with
the above technology, you can factor this through spaces:
LK(n)
Sp / Sp
<
Ω∞
 Φn
Top∗
As a consequence, if E = LK(n) E, then you get a canonically defined retraction
[Ω∞ F, Ω∞ E] → [ΦΩ∞ F, ΦΩ∞ E] ≈ [LF, LE] ≈ [F, E]
of the map [F, E] → [Ω∞ F, Ω∞ E].
5.22. Computing the Bousfield-Kuhn idempotent. In many cases you can compute
the resulting idempontent E on [Ω∞ F, Ω∞ E], whose image is infinite loop maps.
One way is suggested by the construction of Φ, and should lead to a formula generalizing
the “Bott periodicity” definition for K-theory. Stacey and Whitehouse [SW08] carried this
out in the case that E and F are K(n) or similar.
5.23. A sketch of the idea. In the approach I describe here, the key case turns out to be
F = S. Given a ∈ E 0 (Ω∞ S) = [Ω∞ S, Ω∞ E], we have
Ω∞ S
a / Ω ∞
E
7
ηΩ∞ S
 Ω∞ e
a
∞ ∞ ∞
Ω Σ (Ω S)
a : Σ∞ Ω∞ S → E is the adjoint to a. Applying Φ gives
where e
LS
Φa / LE
7 =E
λ:=Φη
 Le
a
LΣ∞ Ω∞ S
Here, λ ∈ π0 LΩ∞ S.
In other words, [Ω∞ S, Ω∞ E] → [S, E] is computed by
h−, λi : E 0 Ω∞ S → E 0 S,
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 41

which can be thought of in terms of pairing with the Hurewicz image of of λ in E0∧ Ω∞ S.
Fix L = LK(n) .
5.24. Theorem ([Rez06]). Given a space X, spectra E, F with E = LE, and a map
f : Ω∞ F → Ω∞ E, and x ∈ F 0 X, we have
(Ef )(x) = hf (τ x), λi.
That is, the diagram
Ef
F 0X / E 0O X
τ h−,λi

F 0 (X × Ω∞ S) / E 0 (X × Ω∞ S)
f

commutes, where τ is the “transfer”.

5.25. A formula for En . Computing the Bousfield-Kuhn idempotent amounts to “naming”
the element λ ∈ π0 LΩ∞ S. It is easier to name its image in E0∧ Ω∞ S, because we have better

+
access to the E homology of symmetric groups, using Ω∞ S ≈
`
BΣm , and HKR character
theory.
5.26. Proposition. Let E be a height n Morava E-theory. The BK idempotent on
[Ω∞ F, Ω∞ E]H is computed by the formula
n
r
(−1)r p(2)−r
X X
(Ef )(x) = hf (τpr (x)), ω(α)i.
k=0 [α]

Here
• τr (x) : X × B(Z/p)r → Ω∞ F is the transfer of x along X × E(Z/p)r → X × B(Z/p)r ;
• the inner sum is over conjugacy classes of surjective homomorphisms α : Znp → (Z/p)r ;
• h−, αi : E 0 (X × B(Z/p)r ) → E 0 X ⊗E0 D is evaluation of the HKR character map at
α.
5.27. Corollary. If [Ω∞ F, Ω∞ E] is p-torsion free, then an H-map f : Ω∞ F → Ω∞ E is infinite
loop iff f τr = τr f for all r = 1, . . . , n.
Proof. Amounts to hτk (1), αi = 0 when r ≥ 1, by the transfer formula from HKR. 
There is a more complicated formula for non-H-maps, using cross-effects. Using this, you
get the formula for `n = Es, where s : GL1 (E) → Ω∞ E is the shift map.
5.28. Theorem ([Rez06]).
n Y

(−1)r p(r2)−r+1
Y 
1
`n (x) = log ψα (x) .
p k=0 [α]

The functions ψα : E 0 X → E 0 X ⊗E0 D are certain power operations of the form

E 0 X → E 0 X ⊗ E 0 BΣpr /I → E 0 X ⊗ E 0 B(Z/p)r /I → E 0 X ⊗ D.
42 CHARLES REZK

5.29. The height 2 case. Suppose E has height 2. Then we can write the above formula
in the following form.
1 xp Np2 (x)
`2 (x) = log .
p Np (x)
Here Np (x) = ψα0 (x) . . . ψαp (x) corresponding to the p + 1 conjugacy classes of Z2p → Z/p,
while N2 (x) = ψα (x) corresponding to any projection Z2p → (Z/p)2 .
Recall that power operations for E-theory produce maps
s
ψr : E 0 (X) → E 0 (X) ⊗E0 Ar .
• The map Np is the composite
ψ1 s norm
E 0 (X) −→ E 0 (X) ⊗E0 A1 −−−→ E 0 X,
where the second map is the Galois norm associated to s : A0 → A1 , which is finite
and free of rank p + 1. The operation Np is multiplicative but not additive.
• The map Np2 is the composite
ψ2 s
E 0 (X) −→ E 0 (X) ⊗E0 A2 → E 0 (X),
using A2 → A0 which classifies the subgroup G[p] ⊂ G.
The fact that this is well defined relies on a congruence
xp Np2 (x) ≡ Np (x) mod pE 0 (X),
which is a consequence of the Frobenius congruence for power operations.
To compute `2 on homotopy groups, compute it on E 0 (S 2n ) = E0 []/(2 ). The formula
becomes linearized, so on π2n we have
`2 (f ) = f − Tp (f ) + pTp2 (f ).
−1
ψαi , or equivalently, p−1 times the trace version of Np , and Tp2 = p−2 Np2 .
P
Here Tp = p
The maps are versions of Hecke operators.
5.30. Application to tmf. This can be applied to tmf, obtaining a map of spectra
`2 : gl1 tmf → LK(2) tmf
whose effect on π2k (up to torsion) is f 7→ f − Tp (f ) + pk−1 f .
Using the Hasse square and the K(1)-local version of this, we show [AHR06] that you can
factor this through a map
`tmf
gl1 tmf −− → tmf p → LK(2) tmf.
The existence of `tmf is ad hoc, and relies on some calculations. Such a map doesn’t exist for
general commuative S-algebras. We obtain a commutative diagram
`2
,
gl1 tmf / tmf p / LK(2) tmf
ι2

`1 ι1 ι1
  
LK(1) tmf / LK(1) tmf L / LK(1) LK(2) tmf
1−U K(1) ι2

In the case of tmf, here is a clue for why this ought to work: the above formula given for `2
actually makes sense for any elliptic curve with descent for isogenies.
 ELLIPTIC COHOMOLOGY AND ELLIPTIC CURVES (FELIX KLEIN LECTURES, BONN 2015) 43

Getting control of gl1 tmf in this way is one of the steps in the construction of the string
orientation for tmf.
References
[AA66] J. F. Adams and M. F. Atiyah, K-theory and the Hopf invariant, Quart. J. Math. Oxford Ser. (2)
17 (1966), 31–38.
[AP76] J. F. Adams and S. B. Priddy, Uniqueness of BSO, Math. Proc. Cambridge Philos. Soc. 80
(1976), no. 3, 475–509.
[And95] Matthew Ando, Isogenies of formal group laws and power operations in the cohomology theories
En , Duke Math. J. 79 (1995), no. 2, 423–485.
[AB02] Matthew Ando and Maria Basterra, The Witten genus and equivariant elliptic cohomology, Math.
Z. 240 (2002), no. 4, 787–822.
[AHR06] M. Ando, M. J. Hopkins, and C. Rezk, Multiplicative Orientations of KO-theory and of the
spectrum of topological modular forms (2006), in preparation, available at http://www.math.
uiuc.edu/~mando/papers/koandtmf.pdf.
[AHS01] M. Ando, M. J. Hopkins, and N. P. Strickland, Elliptic spectra, the Witten genus and the theorem
of the cube, Invent. Math. 146 (2001), no. 3, 595–687.
[AHS04] Matthew Ando, Michael J. Hopkins, and Neil P. Strickland, The sigma orientation is an H∞
map, Amer. J. Math. 126 (2004), no. 2, 247–334.
[ADL16] G. Z. Arone, W. G. Dwyer, and K. Lesh, Bredon homology of partition complexes, Doc. Math. 21
(2016), 1227–1268.
[Ati66] M. F. Atiyah, Power operations in K-theory, Quart. J. Math. Oxford Ser. (2) 17 (1966), 165–193.
[Beh06] Mark Behrens, A modular description of the K(2)-local sphere at the prime 3, Topology 45 (2006),
no. 2, 343–402.
[Beh07] , Buildings, elliptic curves, and the K(2)-local sphere, Amer. J. Math. 129 (2007), no. 6,
1513–1563.
[BL06] Mark Behrens and Tyler Lawson, Isogenies of elliptic curves and the Morava stabilizer group, J.
Pure Appl. Algebra 207 (2006), no. 1, 37–49, available at arXiv:math/0508079.
[BE13] Daniel Berwick-Evans, Perturbative sigma models, elliptic cohomology and the Witten genus
(2013), available at arXiv:1311.6836.
[Bou87] A. K. Bousfield, Uniqueness of infinite deloopings for K-theoretic spaces, Pacific J. Math. 129
(1987), no. 1, 1–31.
[Bou96] , On λ-rings and the K-theory of infinite loop spaces, K-Theory 10 (1996), no. 1, 1–30.
[Bui05] Alexandru Buium, Arithmetic differential equations, Mathematical Surveys and Monographs,
vol. 118, American Mathematical Society, Providence, RI, 2005.
[DFHH14] Christopher L. Douglas, John Francis, André G. Henriques, and Michael A. Hill (eds.), Topological
modular forms, Mathematical Surveys and Monographs, vol. 201, American Mathematical Society,
Providence, RI, 2014.
[Gan07] Nora Ganter, Stringy power operations in Tate K-theory (2007), available at arXiv:math/0701565.
[Gan13] , Power operations in orbifold Tate K-theory, Homology Homotopy Appl. 15 (2013), no. 1,
313–342.
[HL16] Michael Hill and Tyler Lawson, Topological modular forms with level structure, Invent. Math. 203
(2016), no. 2, 359–416.
[HBJ92] Friedrich Hirzebruch, Thomas Berger, and Rainer Jung, Manifolds and modular forms, Aspects of
Mathematics, E20, Friedr. Vieweg & Sohn, Braunschweig, 1992. With appendices by Nils-Peter
Skoruppa and by Paul Baum.
[Joy85a] André Joyal, δ-anneaux et vecteurs de Witt, C. R. Math. Rep. Acad. Sci. Canada 7 (1985), no. 3,
177–182 (French).
[Joy85b] , δ-anneaux et λ-anneaux, C. R. Math. Rep. Acad. Sci. Canada 7 (1985), no. 4, 227–232
(French).
[Kas98] Takuji Kashiwabara, Brown-Peterson cohomology of Ω∞ Σ∞ S 2n , Quart. J. Math. Oxford Ser. (2)
49 (1998), no. 195, 345–362.
44 CHARLES REZK

[KM85] Nicholas M. Katz and Barry Mazur, Arithmetic moduli of elliptic curves, Annals of Mathematics
Studies, vol. 108, Princeton University Press, Princeton, NJ, 1985.
[Kuh89] Nicholas J. Kuhn, Morava K-theories and infinite loop spaces, Algebraic topology (Arcata, CA,
1986), Lecture Notes in Math., vol. 1370, Springer, Berlin, 1989, pp. 243–257.
[Lur09] J. Lurie, A survey of elliptic cohomology, Algebraic topology, Abel Symp., vol. 4, Springer, Berlin,
2009, pp. 219–277.
[MST77] I. Madsen, V. Snaith, and J. Tornehave, Infinite loop maps in geometric topology, Math. Proc.
Cambridge Philos. Soc. 81 (1977), no. 3, 399–430.
[Rez06] Charles Rezk, The units of a ring spectrum and a logarithmic cohomology operation, J. Amer.
Math. Soc. 19 (2006), no. 4, 969–1014 (electronic).
[Rez08] , Power operations for Morava E-theory of height 2 at the prime 2 (2008), available at
arXiv:0812.1320(math.AT).
[Rez09] , The congrugence criterion for power operations in Morava E-theory, Homology, Homotopy
Appl. 11 (2009), no. 2, 327–379, available at arXiv:0902.2499.
[Rez12a] , Rings of power operations for Morava E-theories are Koszul (2012), available at arXiv:
1204.4831.
[Rez12b] , Modular isogeny complexes, Algebraic & Geometric Topology 12 (2012), no. 3, 1373–1403,
available at :arXiv:1102.5022.
[Rez13] , Power operations in Morava E-theory: structure and calculations (2013), available at
http://www.math.uiuc.edu/~rezk/power-ops-ht-2.pdf.
[Rez14] , Etale extensions of λ-rings (2014), available at http://www.math.uiuc.edu/~rezk/
etale-lambda.pdf.
[Ros01] Ioanid Rosu, Equivariant elliptic cohomology and rigidity, Amer. J. Math. 123 (2001), no. 4,
647–677.
[SS15] Tomer M. Schlank and Nathaniel Stapleton, A transchromatic proof of Strickland’s theorem, Adv.
Math. 285 (2015), 1415–1447.
[Sch16] Stefan Schwede, Global homotopy theory (2016), available at http://www.math.uni-bonn.de/
~schwede/global.pdf.
[Seg88] Graeme Segal, Elliptic cohomology (after Landweber-Stong, Ochanine, Witten, and others),
Astérisque 161-162 (1988), Exp. No. 695, 4, 187–201 (1989). Séminaire Bourbaki, Vol. 1987/88.
[Seg07] , What is an elliptic object?, Elliptic cohomology, London Math. Soc. Lecture Note Ser.,
vol. 342, Cambridge Univ. Press, Cambridge, 2007, pp. 306–317.
[SW08] Andrew Stacey and Sarah Whitehouse, Stable and unstable operations in mod p cohomology
theories, Algebr. Geom. Topol. 8 (2008), no. 2, 1059–1091.
[ST11] Stephan Stolz and Peter Teichner, Supersymmetric field theories and generalized cohomology,
Mathematical foundations of quantum field theory and perturbative string theory, Proc. Sympos.
Pure Math., vol. 83, Amer. Math. Soc., Providence, RI, 2011, pp. 279–340.
[Str97] Neil P. Strickland, Finite subgroups of formal groups, J. Pure Appl. Algebra 121 (1997), no. 2,
161–208.
[Str98] N. P. Strickland, Morava E-theory of symmetric groups, Topology 37 (1998), no. 4, 757–779,
available at arXiv:math/9801125.
[tD89] Tammo tom Dieck, The Artin-Hasse logarithm for λ-rings, Algebraic topology (Arcata, CA, 1986),
Lecture Notes in Math., vol. 1370, Springer, Berlin, 1989, pp. 409–415.
[Wal09] Barry Walker, Orientations and p-adic analysis (2009), available at arXiv:0905.0022.
[Wil82] Clarence Wilkerson, Lambda-rings, binomial domains, and vector bundles over CP (∞), Comm.
Algebra 10 (1982), no. 3, 311–328.
[Zhu14] Yifei Zhu, The power operation structure on Morava E-theory of height 2 at the prime 3, Algebr.
Geom. Topol. 14 (2014), no. 2, 953–977.
[Zhu15] , The Hecke algebra action on Morava E-theory of height 2 (2015), available at arXiv:
1505.06377.

Department of Mathematics, University of Illinois, Urbana, IL

E-mail address: [email protected]