]> Number Fields - The Discriminant

The Discriminant

Let K be a number field of degree n over . Let σ 1 ,...,σ n be the embeddings of K in . We write [a ij] for the matrix with a ij in the ith row and j column, and a ij for the determinant of this matrix. Define the discriminant of α 1 ,...,α nK by

disc(α 1 ,...,α n)=σ i(α j) 2

As with the trace and norm, the discriminant may be generalized by replacing with any number field.

Theorem: disc(α 1 ,...,α n)=T(α iα j)

Proof: Recall some basic facts about determinants:

a ij=a ji,AB=AB

Then the theorem follows from

[σ j(α i)][σ i(α j)]=[σ 1 (α iα j)+...+σ n(α iα j)]=[T(α iα j)]

Corollary: disc(α 1 ,...,α n), and if the α i are integral, then disc(α 1 ,...,α n).

Theorem: disc(α 1 ,...,α n)=0 if and only if α 1 ,...,α n are linearly dependent over .

Proof: If the α j are linearly dependent over then so are the columns of the matrix [σ i(α j)] and its determinant will be zero. Conversely if disc(α 1 ,...,α n)=0 , then the rows R i of the matrix [T(α iα j)] are linearly dependent. Suppose the α i are linearly independent. Then let a 1 ,..,a n be rationals, not all zero, such that a 1 R 1 +...+a nR n=0 . Then consider α=a 1 α 1 +...+a nα n0 . Note we have T(αα j)=0 for all j. The α j form a basis for K over , and since α0 , so do the αα j. But this implies T(β)=0 for all βK, a contradiction since there exist elements of nonzero trace (for example T(1 )=n).

Theorem: Let K=[α] and let α 1 ,...,α n be the conjugates of α over . Let f be the minimal polynomial of α. Then

disc(1 ,α,...,α n1 )= 1 r<sn(α rα s) 2 =±N K(f(α))

where the sign is positive if and only if n=0,1 (mod4 ).

Proof: First observe σ i(α j1 )=(σ i(α)) j1 =α i j1 is a Vandermonde determinant, yielding the first equality.

Next we have

r<s(α rα s) 2 =± rs(α rα s)

where the plus sign holds if and only if n=0,1 (mod4 ). Since f has rational coefficients, we have

N K(f(α))= r=1 nσ r(f(α))= r=1 nf(σ r(α))= r=1 nf(α r)

Then the second equality follows from the fact that for all r

f(α r)= sr(α rα s)

(To see this, let f(x)=(xα r)g(x). Then by the product rule, f(α r)=g(α r), and the roots of g are all the roots of f except for α r.)

Let us now apply this theorem to compute disc(1 ,ω,...,ω p2 ) for ω=e 2 πi/p where p is an odd prime. We have f(x)=1 +x+...+x p1 . Since x p1 =(x1 )f(x), we have px p1 =f(x)+(x1 )f(x) showing that

f(ω)=pω(ω1 )

Taking norms gives

N(f(ω))=N(p)N(ω)N(ω1 )=p p1 N(ω1 )

Since (1 ω)...(1 ω p1 )=p, we have N(ω1 )=N(1 ω)=p, hence N(f(ω))=p p2 .

Write disc(α) for disc(1 ,α,...,α n) where α is a degree n algebraic integer over and K is taken to be [α].

We now know that disc(ω)=±p p2 for prime p. For ω=e 2 πi/m for general m, the expression is more complicated, but it can be readily shown that disc(ω)m ϕ(m) as follows. Let f be the minimal polynomial of ω. Then Since x m1 =f(x)g(x) for some g[x] we have m=ωf(ω)g(ω). Taking norms gives

m ϕ(m)=±disc(ω)N(ωg(ω))

and we are done since N(ωg(ω)).