# What is the sphere spectrum?

This post is speculative, and is meant to attempt to understand what the sphere spectrum *is* from this derived point of view.

The sphere spectrum is absolutely essential to everything that’s done in homotopy theory – it’s the whole reason the field was created! It acts on everything, and in fact, modules over are exactly spectra. Categorically, we can write this as

We know another interpretation of the left hand side, coming from algebraic geometry;
morally, it should be .
It is quite reasonable to ask what **is**.
For instance, what are its points?
What is the structure sheaf?

# The thick subcategory theorem

The thick subcategory theorem is a theorem that’s part of the Ravenel conjectures, that was proved by Hopkins-Smith. We’ll recall a proof of this result here. Let’s recall a definition from homological algebra.

**Definition:** A subcategory of a triangulated category is *thick*
if it’s closed under retracts, weak equivalences, and
satisfies the 2-out-of-3 property for cofiber sequence.

**Theorem (Hopkins-Smith):** The only thick subcategories of the category of
-local finite spectra are the categories of
-local spectra that are -acyclic.

*Proof:* Let be any thick subcategory of the triangulated category of -local finite spectra.
We’ll show that any object in that’s in (with minimal)
gives an inclusion .
Let be any other -acyclic spectrum that is in , and
let denote its Spanier-Whitehead dual.
The identity on adjuncts to a map (everything is -local),
and for , we therefore have
an injection .
If is the fiber of , then this tells us that
is zero if .

Now, the composite is zero on -homology for , but this argument. Since is -acyclic – this implies that it is -acyclic for – it follows that is zero for . But if is null for all , it follows from nilpotence that is null for some . The evaluation map from therefore shows that is null, i.e., the map is null. The cofiber of this map is thus ,

We know that is in , so we use the cofiber sequence

to show by induction that is in for all . Since is a retract of this, we have .

It’s easy to see that (since otherwise, would contain some spectrum that’s -acyclic for , which contradicts minimality). This concludes the proof.

# What does this mean for us?

Let denote the the collection of thick subcategories of which are *prime ideals*,
in the sense that if and , and if
, then either or .
This theorem shows that the only elements in are the thick subcategories ,
for every integer and every prime .
This means that we would expect to be as sets.

This is an awful lot like – but it’s not, since

as triangulated categories. But perhaps this is true in the derived context?

In particular, I think the following proto-theorem should be true.

**Proto-Theorem:** If a ‘‘derived moduli stack’’ of formal groups exist, then as -categories.

There is evidence against this theorem, though, as the legends tell us in this MO question. The Adams-Novikov spectral sequence (described below) not degenerating is supposed to be like differenting from the category of spectra.

Still, morally, one wants to be as close as humanly (chromotopically?) possible to , but if you’re to believe Lurie’s definition of a derived stack, presented in a previous blog post, the derived scheme is given by the data of – with the Zariski topology – and a structure sheaf of -rings. I think this is wrong: the underlying set of doesn’t contain all the Morava -theories, that should morally be there (by the thick subcategory theorem)!

Let’s try to construct the derived moduli stack of formal groups. We can lift the infinitesimal neighborhood of a point; this is exactly the derived scheme that we were considering in previous posts. Suppose that the proto-theorem was literally true, so that . This would admit a map , where is the spectrum of periodic complex cobordism. For any spectrum , we’d then know that the (quasicoherent) sheaf on corresponds to the quasicoherent sheaf on associated to , and that . Then the descent spectral sequence for would run

A quasicoherent sheaf on is just a -comodule, and the cohomology groups are then just groups. In particular, the -page would just be – and now the spectral sequence is just the Adams-Novikov spectral sequence!

Recall that maps over correspond to homotopy commutative maps , where and are the cohomology theories associated to and (I guess here I’m assuming that and ) are homotopy commutative. This is Proposition 1.2 in this paper of Paul Goerss. One feature of (that follows from the above discussion) would be that it’d classify maps between -rings: maps over would correspond to -maps – at least if and are -complete. But this is literally telling us that should be ! This is some form of a functorial Landweber exact functor theorem.

If we were to go about trying to actually construct , we would need to know that its underlying stack, i.e., , is locally trivial in some Grothendieck topology. As we said in a previous post, the best topology to use when considering derived stuff is the etale topology. I don’t think it’s true, though, that is locally trivial in the etale topology. So:

**Question:** In what topology is locally trivial?
What is the appropriate notion of a ‘‘derived stack’’ in that topology?

Here is a table that I made when trying to understand this story.

Note that with Lurie’s definition, the last line is not true, even for derived affine schemes!
A fabulous reference for the right hand side, not in the derived context,
is Goerss’ *Quasicoherent sheaves over the moduli stack of formal groups*, available on the arXiv.

In some sense, the whole goal of chromatic homotopy theory is to construct .