5.9. Norms and traces.

Back

The trace of a square matrix is the sum of its diagonal elements, Tr(aij) =

_

i aii. Since

Tr(UAU1) = Tr(A), we can define the trace of an endomorphism α of a finite-dimensional

vector space V to be the trace of the matrix of α with respect to any basis of V .

Similarly, we can define the determinant and characteristic polynomial of α to be the

determinant and characteristic polynomial of the matrix of α with respect to any basis of V .

In a little more detail, a direct computation shows that Tr(AB) = Tr(BA), which shows

that Tr(UAU1) = Tr(A) and hence Tr(α) is well-defined. The characteristic polynomial of

α can be defined to be

cα(X) = Xn + c1Xn1 + + cn, ci = (1)i Tr(α|ΛiV ), n= dimV ;

in particular, c1 = Tr(A) and cn = (1)n detA. If A is the matrix of α with respect to

some basis for V , then cα(X) = det(XIn A).

For α and β endomorphisms of a finite-dimensional F-vector space V , we have

Tr(α) F;Tr (α + β) =Tr(α) + Tr(β);

det(α) F;de t(αβ) =det(α) det(β).

Now let E be a finite extension of F of degree n, and regard E as an F-vector space. Then

α E defines an F-linear map αL : E E, x αx.

Define:

TrE/F(α) =Tr(αL);T r is a homomorphism (E, +) (F, +);

NmE/F(α) =det(αL);N m is homomorphism (E×,×) (F×,×);

cα(X) =cαL(X).

50 J.S. MILNE

Note that α αL is an injective ring F-homomorphism from E into the ring of endomorphisms

of E as a vector space over F, and so the minimum polynomial of α (in the sense of

Section 1.8) is the same as the minimum polynomial of αL (in the sense of linear algebra).

Example 5.32. (a) Consider the field extension C R;the matrix of αL, α = a + bi,

relative to the basis 1, i is

_

a b

b a

_

, and so

TrC/R(α) = 2"(α), NmC/R(α) = |α|2.

(b) For α F, Tr(α) = rα, Nm(α) = αr , r = [E : F].

(c) Let E = Q[α, i] be the splitting field of X8 2. What are the norm and the trace of

α? The definition requires us to compute a 16 × 16 matrix. We shall see a quicker way of

computing them presently.

Proposition 5.33. Consider a finite field extension E/F, and let f(X) be the minimum

polynomial of α E (in the sense of Section 1.8). Then

cα(X) = f(X)[E:F[α]].

Proof. Suppose first that E = F[α]. In this case, we have to show that cα(X) = f(X).

But f(X)|cα(X) because cα(αL) = 0 (Cayley-Hamilton theorem), and the injectivity of

E EndF-linear(E) then implies that cα(α) = 0. Since the polynomials are monic of the

same degree, they must be equal.

For the general case, write V for E regarded as an F-vector space. The endomorphism

αL of V defines an action of F[X] on V (see Math 593), and this action factors through

F[X]/(f(X)) = F[α]. Because F[α] is a field, V is a free F[α]-module, and in fact, V

F[α]m with m = [E : F[α]] (count dimensions over F). Hence the characteristic polynomial

of α acting on V is the mth power of its characteristic polynomial acting on F[α], which,

according to case already proved, is f(X).

Alternatively, we can be more explicit. Let β1, ..., βn be a basis for F[α] over F, and let

γ1, ..., γm be a basis for E over F[α]. As we saw in the proof of (1.10), {βiγk} is a basis

for E over F. Write αβi =

_

ajiβj;the n A = (aij) has characteristic polynomial f(X)

according to the first case proved. Note that αβiγk =

_

ajiβjγk. Therefore the matrix of

αL in End(E) breaks up into n × n blocks with As down the diagonal and zero matrices

elsewhere. Therefore its characteristic polynomial is f(X)m.

Corollary 5.34. Suppose that the roots of the minimum polynomial of α are α1, . . . , αn

(in some splitting field containing E), and that [E : F[α]] = m. Then

Tr(α) = m

_n

i=1

αi, NmE/F α =

_

_n

i=1

αi

_m

.

Proof. Write the minimum polynomial of α as

f(X) = Xn + a1Xn1 + + an =

_

(X αi).

Then

cα(X) = (f(X))m = Xmn + ma1Xmn1 + + amn

,

and so

TrE/F (α) = ma1 = m

_

αi,

FIELDS AND GALOIS THEORY 51

and

NmE/F(α) = (1)mnamn

= (

_

αi)m.

Example 5.35. (a) Consider the extension C R. If α C \ R, then

cα(X) = f(X) = X2 2"(α)X + |α|2.

If α R, then cα(X) = (X a)2.

(b) Let E = Q[α, i] be the splitting field of X8 2 (see Exercise 16). The minimum

polynomial of α = 8

2 is X8 2, and so

TrQ[α]/Q α = 0;T rE/Q α = 0.

NmQ[α]/Q α = 2;N mE/Q α = 4.

Remark 5.36. Assume E is separable over F, and let Ω be an algebraic closure of F;let

σ1, ..., σr be the distinct embeddings of E into Ω. Then

TrE/F α =

_

σiα

NmE/F α =

_

σiα.

When E = F[α], this follows from the observation (cf. 2.1b) that the σiα are the roots of

the minimum polynomial f(X) of α over F. In the general case, σ1α, ..., σrα are still roots

of f(X) in Ω, but now each root of f(X) occurs [E : F[α]] times (cf. the proof of 2.7).

For example, if E is Galois over F with Galois group G, then

TrE/F α =

_

σG

σα

NmE/F α =

_

σG

σα.

Proposition 5.37. For finite extensions E M F, we have

TrE/M TrM/F = TrE/F ,

NmE/M NmM/F = NmE/F .

Proof. If E is separable over F, then this can be proved fairly easily using the descriptions

in the above remark. We omit the proof in the general case.

Proposition 5.38. Let f(X) F[X] factor as f(X) =

_m

i=1(X αi) in some splitting

field, and let α = α1. Then, with f_ = df

dX , we have

disc f(X) = (1)m(m1)/2 NmF[α]/F f_(α).

Proof. Compute that

disc f(X) df =

_

i<j

(αi αj)2

= (1)m(m1)/2

_

i

(

_

j_=i

(αi αj))

= (1)m(m1)/2

_

f_(αj)

= (1)m(m1)/2 NmF[α]/F (f

_

(α)).

52 J.S. MILNE

Example 5.39. We compute the discriminant of

f(X) = Xn + aX + b, a, b F,

assumed to be irreducible and separable. Let α be a root of f(X), and let γ = f_(α) =

nαn1 + a. We compute its norm. On multiplying the equation

αn + aα + b = 0

by nα1 and rearranging, we obtain the equation

nαn1 = na nbα

1.

Hence

γ = nαn1 + a = (n 1)a nbα1.

Solving for α gives

α =

nb

γ + (n 1)a

,

from which it is clear that F[α] = F[γ], and so the minimum polynomial of γ over F has

degree n also. If we write

f(

nb

X + (n 1)a

) =

P(X)

Q(X)

,

then P(γ) = f(α) = 0. Since

P(X) = (X + (n 1)a)n na(X + (n 1)a)n1 + (1)nnnbn1

is monic of degree n, it must be the minimum polynomial of γ. Therefore Nmγ is (1)n

times the constant term of this polynomial, and so we find that

Nmγ = nnbn1 + (1)n1(n 1)n1an.

Finally we obtain the formula,

disc(Xn + aX + b) = (1)n(n1)/2(nnbn1 + (1)n1(n 1)n1an),

which is something Maple doesnt know (because it doesnt understand symbols as exponents).

For example,

disc(X5 + aX + b) = 55b4 + 44a5.