Algebraic geometry studies sheaves, which are ways of gluing things together
Studies polynomials
A sieve on an object B is a subfunctor of Mor(-,B) = hB
In other words, it's a collection of morphisms that point into B.
Denote a sieve on B by sB Similarly, a cosieve is sB
A subfunctor F of Hom(-,B) is a presheaf such that F(A) ⊆ Hom(A,B) and F(f) is the restriction of hBf. Say f: A → X F(f): F(X) → F(A) ⊆ Hom(A,B)
But for there to be such a map, if F(X) ≠ ∅, it must be the case that F(A) ≠ ∅ So we have that if g ∈ F(X), so g: X → B then F(f)(g) = fg: A → X → B must be an element of F(A)
In this way, sB is something like a collection of left ideals
It follows then that it's rare for sB(B) to contain 1B: if it did, it would also contain all morphisms into B. So we'd have sB = hB
So 1B ∈ sB iff sB = hB.
Then hB is something like the full ring and sB is an ideal
Say we have a one-object category { * }. Then hB is the monoid of all morphisms in the category, and sB is a sub-monoid closed under left multiplication by any element in hB
In the case of enriched categories, taking abelian enrichment turns hB into a ring, and sB into literally an ideal
A sieve is something like a filter: it picks out regions of B
We can think of a sieve as an ideal in a ringoid (but not literally in a ringoid)
Let sB ≤ tB iff sBA ⊆ tBA for all A. Define the union and intersection objectwise (sB ∪ tB)(A) = sBA ∪ tBA (sB ∩ tB)(A) = sBA ∩ tBA
Then you get that for any fixed B, the collection of all sieves forms a complete lattice.
Let sB be a sieve and f: A → B. We get a sieve f* sB called the pullback, defined by
(f* sB)(X) = { g: X → A | gf ∈ sB (X) }
This is actually the pullback in the categorical sense:
We take sB ×hB hA = f* sB
This sends X → sB(X) ×hB(X) hA(X), which we compute in Set:
This is { (g,h) ∈ sB(X) × hA(X) | g = hf }
Personally, I don't like the f* notation for the pullback. I'll denote the pullback by
sB / f
since it corresponds to the set of h: A → X such that hf ∈ sB
Consider the one-object case Given an ideal s*, and a ring element x: * → *, the pullback is x*s* = { a | ax ∈ s* } This is the preimage of right multiplication by x ρx-1(s*)
This just ends up being the collection of all sieves of the form hB/f But since hB contains all of the maps into B, every map of the form gf will be in hB. So
hB / sB = { hA | there is a map f: A → B in sB(A) }
So hB/sB is something like taking all of the hom sets and then restricting to the ones that are connected to B through s.
This is the set of all sieves of the form tB/f, where f is any map into B
So the sieves in tB / hB are those such that
For each object, we get a sieve. So think of our sieves as collected into a map S: C × C → Set
The weird thing is that we want more than one sieve for each object. We choose a collection of sieves, which we call covering sieves. They have to satisfy the following:
hB is a covering sieve for all B
If sB is covering and f:A → B then sB/f is covering.
Let sB be covering and tB be any sieve. If tB/f is covering for all f ∈ sB(A), for all A, then tB is also covering
This is a much nicer way of looking at Grothendieck topologies, I think — it's very algebraic The conditions become
Let's define tB/sB to be all of the sieves of the form tB/f, with f ∈ sB(A) for some A.
Then we say that if tB /sB is covering, then so is tB
Given any ideal t, and a covering ideal s, if ρx-1(t) is covering for all x ∈ s, then t is covering
If sB ≤ tB and s is covering then so is t
TODO Proof
If xf ∈ s ≤ t then xf ∈ t. So ρf-1(s) ⊆ ρf-1(t) for all f
A site is just (C, S), a category and a Grothendieck topology
This is also called a topos
Given a contravariant functor F mapping into Set, we call the set F(X) the sections of F over X
From the Yoneda lemma, we have Hom(hB, F) ≅ F(B) naturally. So we can identify the elements of Hom(hB , F) with the sections.
We associate the two with η → ηB(1B) And if x ∈ F(B), we take ηA: hB(A) → F(A) to be the map which sends f: A → B to F(f)(x) ∈ F(A)
So ηx = F(-)(x) is the corresponding natural transformation
These are ways of taking something local and turning it into something global
A separated presheaf is a contravariant functor F, such that given any covering sieve sB, the map
Hom(hB, F) → Hom(sB, F)
is injective.
A sheaf on a site is a contravariant functor F, such that given any covering sieve sB, the map
Hom(hB, F) → Hom(sB, F)
is a bijection.
A η: hB ⇒ F is determined by ηB and what it does to 1B What is ηB doing on sB?
ηB: sBB → F(B)
We don't necessarily have that 1B ∈ sBB
The map above sends a natural transformation (a section of the presheaf) η: hB → F to basically the same transformation, just restricted to sB
So this basically says that every covering sieve has 'enough space' to contain all of the possible transformations hB → F. My guess is that means something like the covering sieves are all global?
A sheaf is a functor F : 𝒞 → 𝒟
It assigns to each of the objects in 𝒞 some local structure in 𝒟
We think of the morphisms as telling us how one local structure fits inside of another
If we zoom-in to a local region, it should tell us the local structure
If we take a pull-back, which we think of as an intersection,