These are various examples or counterexamples to the slogan “total space = base × fiber”.
General linear group over projective space
Here we fix an isomorpism P1≅Proj(k[x,y]), and similarly fix coordinates for
GL2,k after discussing the definition thereof. I believe this example holds in positive
characteristic, but I generally have been thinking about the characteristic zero case.
The general linear group is a group scheme by this example on the stacks
project, and is given by GL2=Spec(Z[x,y,z,w,1/(xw−zy)]). Base changing to a field k gives GL2,k=Spec(k[x,y,z,w,1/(xw−zy)])
(this is effectively the classical definition). Consider the map ψ1:GL2,k→U1≅A1 given by the ring morphism k[x]→k[x,y,z,w,1/(xw−zy)],x↦(xw−yz) and the map
ψ2:GL2,k→U2≅A1 given by the ring morphism k[x]→k[x,y,z,w,1/(xw−zy)],x↦1/(xw−yz). Viewing U1→P1 and U2→P1 as affine patches, we see that
U1×P1U2≅Speck[x,x−1]; under this isomorphism, we see that ψ1 and
ψ2 agree on U1×P1U2, the overlap. This means that we can glue them to a map
ψ:GL2,k→P1.
We now seek to compute the fiber ψ−1(p)=GL2,k×P1Speck(p) for a point
p∈P1. Let U≅Speck[x] be an affine patch containing p. Then we may instead
compute the fiber product ψ−1(p)=GL2,k×A1Speck(p), which is affine. Thus
this map is a map who’s source and fiber are affine, but who’s target is not.
The blowup of the cone at a point over the cone itself
Recall the example the blowup of the
cone; let Y
denote the cone and Y~ it’s blowup. It is difficult to look at the scheme
Y~=Proj(k[x,y,z,X,Y,Z]/(x2−z2−y2,xY−yX,xZ−zX,zY−yZ))
and assert that this is a surface; in particular, it’s homogeneous ideal is generated by four
generators and there is no obvious way to reduce that number. However, we know that Proj(A) is a
scheme over SpecA0, the degree zero peice of A; we examine this map and use it to deduce that
Y~ is a surface.
Let f:X~→X be the natural map given by the Proj construction. We seek to compute for
a point P the sheaf-theoretic fiber
f−1(P)=X~×Speck[x,y,z]/(x2−y2−z2)Spec(k[x,y,z]/(x2−y2−z2,x−P1,y−P2,z−P3))
where P=V(x−P1,y−P2,z−P3) and P12−P22−P32=0. We dehomogenize and compute
U(X)×Speck[x,y,z]/(x2−y2−z2)Spec(k[x,y,z]/(x2−y2−z2,x−P1,y−P2,z−P3))
≅Spec(k[x,y,z,Y,Z]/(x2−z2−y2,xY−y,xZ−z,zY−yZ)
⊗k[x,y,z]/(x2−y2−z2)k[x,y,z]/(x2−y2−z2,x−P1,y−P2,z−P3))
Recall that A/(f)⊗AA/(g)≅A/(f,g), so
≅Spec(k[x,y,z,Y,Z]/(x2−z2−y2,xY−y,xZ−z,zY−yZ,x−P1,y−P2,z−P3))
By substitution we obtain
≅Spec(k[Y,Z]/(P12−P32−P22,P1Y−P2,P1Z−P3,P3Y−P2Z))
But P12−P22−P32=0. Supposing P1=0, it has an inverse, so
f−1(P)≅Spec(k[Y,Z]/(Y−P1P2,Z−P1P3,P3Y−P2Z))
Substitution again
≅Spec(k/(P3P1P2−P2P1P3))≅Speck.
If (on the other hand) P1=0 and either P2 or P3 are nonzero, we quotient by the unit ideal
and obtain f−1(P)={}, the empty set; otherwise, we obtain f−1(P)≅Spec(k[Y,Z]).
Similarly, we see
U(Y)×Speck[x,y,z]/(x2−y2−z2)Spec(k[x,y,z]/(x2−y2−z2,x−P1,y−P2,z−P3))
≅Speck[X,Z]/(P1−P2X,P1Z−P3X,P3−P2Z))
and
U(Z)×Speck[x,y,z]/(x2−y2−z2)Spec(k[x,y,z]/(x2−y2−z2,x−P1,y−P2,z−P3)).
≅Speck[z,y,z,X,Y]/(P1Y−P2X,P1−P3X,P3Y−P1))
When any of the Pi are nonzero, the points glue to a single reduced point; when
all are zero, the fiber is a copy of P2.
This shows that the blowup of a singular point is not always a copy of P1, which is
interesting; if we blew up a smooth point, the result would be different. I would venture to argue
that this demonstrates the intimate connection between the blowup and “tangents”; we can think of
this singularity as being a point through which the lines form a two dimensional space, not a one
dimensional space.
We’ve also showed that the map is an isomorphism on geometric points away from V(x,y,z), which
gives that the dimension of the blowup is 2. I might elaborate more on the many ways we can look at
this as a “base × fiber” argument in the future.