Topology. It’s something that every math student has to become comfortable with. First you get to learn metric topology (usually on ), followed by point set topology. Then in graduate school, pretty much everyone gets an introduction to algebraic topology. However, it all really just studies sets with distinguished collections of subsets. There is, however, a more general version of topology, in fact, a way to put a topology onto a category due to, of course, Grothendieck.

So the first question is: how should topologies be defined? The traditional way is that, given a set , a topology is a collection of subsets satisfying the following properties:

1.

2. If with an arbitrary index set, then

3. If then .

There are all sorts of other ways to define a topology, but that’s the standard one. It works out rather nicely, as any topologist can tell you, but somehow, it doesn’t satisfy algebraic geometers. Why, you might ask, would this be inadequate? The simple reason is that in algebraic geometry, there are a lot of situations where there isn’t a “good” topology to put on the set in question.

When algebraic geometers work over the complex numbers, they get two topologies to work with, the classical topology, which has the small open sets that analysts and differential geometers love so much. Also, though, there is the Zariski topology, which has the property that every open set is dense. It is beloved by algebraists for the simple reason that it captures a lot of the algebraic information. For instance, dimension can be defined by using irreducible subsets in the Zariski topology, which just correspond to irreducible polynomials. There’s a very strong algebra-geometry dictionary for this topology.

Unfortunately, when algebraic geometers decide to work in positive characteristic, say in the algebraic closure of for some prime number , there’s only the Zariski topology. So we need some way to generalize the notion of topology so that all varieties have “small” open sets.

Now we return to the math. Let be a topological space. Then we can define a category whose objects are the open sets in and such that, for open, consists of a single element if and is empty otherwise. We’ll call this , and we can construct this for any topological space.

Grothendieck came up with the brilliant insight that the important things are *covers*. He looked at coverings and came to the conclusion that covers also had three important properties, just like topologies do:

1. If is an open set, then is a cover, that is, open sets cover themselves.

2. Coverings pull back, which means that if is a covering and , then is a covering.

3. Coverings compose, that is, if and are all covers, then is a cover.

If you think about it, these are all categorical notions. The first one sounds a lot like the identity morphism, the second and third are even named suggestively as pullback and composition. So let’s move on to what Grothendieck suggested:

A *Grothendieck Topology* (note, the wikipedia article defines Grothendieck Topologies differently than I do) on a category is a collection of sets of morphisms called *coverings* such that

1. Any isomorphism is a covering

2. If is a covering and any map, then exists for each and is a covering

3. If is a covering and is a covering, then is a covering.

We call a category with a Grothendieck topology a *site*.

The only thing here that requires an explanation is the second condition. We should think of that fiber product as an intersection. Look at sets, and take two inclusions of sets into a third. The pullback is then going to turn out to be intersection, because it is the largest object that injects into each of them and makes all the maps commute.

So we should think of this as a generalization of a topology by first taking the coverings as fundamental, and then looking at what it means in a category. So why does this help with the loss of small open sets in algebraic geometry? Because we always have MANY natural sites to choose from, given a variety. I’ll define a few:

1. The Small Zariski Site has objects the open immersions of varieties into , with morphisms the maps that commute with immersion. The coverings are the collections of morphisms whose image covers , this gives something that can be thought of as the standard Zariski topology.

2. The Big Zariski Site has as objects varieties over , that is, varieties equipped with a morphism to with the covers being collections of open immersions that cover the target.

3. The Small Etale Site has objects varieties over where the structure map is etale, which just means smooth of relative dimension zero. The coverings are then collections of arrows such that the arrows are etale and commute with the structure maps and that the images cover .

4. The Big Etale Site is much like the Big Zariski Site, except that the maps must all be etale.

5. The fppf Site is similar to the Big Zariski and Big Etale sites, as it is on the category of schemes over with structure and covering maps all flat and of locally finite presentation. (fppf is from the French for this set of conditions.)

That’s plenty for now, at some point I’ll come back and define sheaves on sites, descent, and perhaps get to stacks, but that will have to wait until I understand them. I will mostly be following the lecture notes from the MSRI Deformation Theory and Moduli Spaces workshop, with additional references brought in, because that’s where I first got rigorous definitions of these things.

You might want to take a look at the wonderful article by Vistoli, titled: Notes on Grothendieck topologies, fibered categories and descent theory. The newest version is available from his website.

Michael Artin’s book on Grothendieck Topologies is also a good reference despite being a little dated.

Thanks, both of you. I’ve got access to a copy of FGA Explained, and have specifically been looking at Vistoli’s notes, as well as my notes from the lectures as MSRI, and I’ll definitely look for Artin’s book.

I’m currently in the process of writing up some notes on Artin’s book, supplemented with some generalizations (with proof) of the category theory results he uses in more modern language. If you are interested I can let you know when I am done.

I’d say though that most of it is probably covered in the notes you’ve already been looking at.

I’m never one to turn down an extra reference, please let me know when your notes are available.

Pingback: Sheaves on Sites « Rigorous Trivialities

Don’t forget the canonical Grothendieck topology, which is the weakest grothendieck topology such that every functor is a sheaf. It shows you can do this for categories that aren’t just schemes, and proving some other topology is stronger than the Grothendieck topology is usual the first interesting thing you prove about it.

Pingback: Locally Ringed Spaces and Varieties « Rigorous Trivialities

Greg: Is the existence of the canonical topology easy to see? I’ve seen its existence asserted many-a-time, but I’ve never seen the proof. (I assume it’s in SGA.)

I know this thread is super old, but to answer Walt’s question, it is easy to see that if {J_a | a \in A } is a family of Grothendieck topologies on a category, then J defined by J(C) = intersection of all the J_a(C) is also a Grothendieck topology. Thus given any collection of sieves on a category the smallest Grothendieck topology making the sieves into covers is well defined (simply take the intersection of all such topologies – the discreet topology always works, so you will always be able to do this). So to show that the Canonical topology exists, simply intersect all the Grothendieck topologies such that the representable presheaves are sheaves.

To expand upon what Steve Gubkin said, if F is any presheaf, and C is an object in the category, then we may define J_F(C) to be the set of subfunctors R of the functor H(-,C) (i.e. R is a sieve) so that, if f:D\to C is any arrow, and \phi:R_f\to F (where R_f is the pull-back of the subobject R along f, to a subobject of H(-,D)) is any natural transformation, then \phi factors through H(-,D).

In other words, J_F(C) is the set of sieves at C such that F has the sheaf property with respect to any pull-back of the sieve.

This topology is clearly the largest topology such that F is a sheaf with respect to it. If F_i are a collection of presheafs, then interesecting all the J_{F_i}’s gives you the largest topology for which all of the F_i’s are simultaneously sheafs.

If you do this construction when the F_i’s are the representable presheafs, then you get the so-called canonical topology.

Pingback: Towards Stacks 1 « A Mind for Madness