Normal schemes are nice, and happily there is a process for taking a scheme and producing “the same scheme but normal”, namely normalization. This can be thought of as a mild analogue of resolution of singularities (whose goal is to produce “the same scheme but nonsingular”), and indeed in the case of curves normalization succeeds in resolving singularities.
The normalization of an integral scheme is constructed as follows. If , the normalization of is , where is the integral closure of in its field of fractions . In general with integral; we can normalize each affine piece and then glue these together to obtain a normalization of . The normalization some universal property of being terminal among dominant morphisms from integral normal schemes to , so is unique up to unique isomorphism.
The examples here are exercises 9.7.E, 9.7.F, 9.7.H of Vakil’s notes (and are maybe also just standard examples, but that’s where the inspiration to do them came from).
First let’s normalize the nodal cubic . By intersecting with lines through the double point at the origin we can find a map from : setting ,
so that and . In fact the map
is the normalization.
Let be the field of fractions of . First note that is integral over , because e.g. . Furthermore (in particular they have the same field of fractions) because of the relations and and . And finally is integrally closed, because it is a unique factorization domain. Thus is the integral closure of in its field of fractions.
Let’s examine the normalization map more closely; in particular, the fiber over the node. The node is
and the fiber over this point is
Since , every element of can be written for some and in . We can transfer through the tensor, and on the left hand side it becomes zero, so
An element of the form on the right is zero precisely when (since is not in ), so the representation is unique. We can also compute that the multiplication is
This realizes as , via the isomorphism
Thus the fiber over the node is , two points with residue field , i.e. normalization splits the node into two points.
The cuspidal cubic is similar: essentialy the same argument shows that its normalization is also , with the following map.
Here the cusp is
and the fiber over this point is
The same analysis as above (using ) shows every element of this tensor product can be uniquely written . We can also compute that the multiplication is
This realizes as , via the isomorphism
Thus the fiber over the node is , one point with nilpotents, i.e. normalization straightens out the cusp into a single point by somehow adding first-order behavior.
For one last geometric example, consider the cubic . This has a node at the origin which appears as an isolated point e.g. if we are working over , but looks like a self-intersection over . That is, it has two distinct tangents, but they may live over a quadratic extension of the base field.
The normalization is again with , and the fiber over the origin is . So in this case the behavior of the normalization at the singular point depends on the ground field; if , then the node is split into two copies, but if then the fiber is a single point with residue field a quaratic extension of .
Now consider the arithmetic curve . The field of fractions of is , whose ring of integers is , so this is the integral closure of (since ). Thus the normalization is given by
corresponding to the inclusion .
If is a prime of lying over a prime of , then . In this case
which shows that the fiber of over is
just a copy of the point .
The primes and remain prime in , for
Recall from our knowledge of prime splitting in quadratic fields that
Thus the fiber of the normalization over is
and the fiber over is
So we can imagine to be a node, because normalization splits it into two copies of itself. We can also imagine to be a node, but of the “isolated point” type, in analogy with the isolated real point of .
2 thoughts on “Normalization of Algebraic and Arithmetic Curves”