• 沒有找到結果。

Theorem 3.10.2 (E

N/A
N/A
Protected

Academic year: 2022

Share "Theorem 3.10.2 (E"

Copied!
6
0
0

加載中.... (立即查看全文)

全文

(1)

Dec. 15, 2006

3.10. solving cubic polynomials. In this section, we are going to review classical result on solving polynomials by using non-classical language. I think this experience also serve a good start for Galois theory in general.

Definition 3.10.1. A character from a group G to a field K is group homomorphism χ : G → K. The set of characters is denoted Homgp(G, K).

Let Hom(G, K) be the set of functions from G to K. It’s clear that Hom(G, K) is a K-vector space.

Theorem 3.10.2 (E. Artin). Homgp(G, K) is linearly independent in Hom(G, K).

Proof. Suppose on the contrary that Homgp(G, K) is not linearly in- dependent. Pick a linearly dependent subset {χ1, ..., χn} of minimal n.

There are ai ∈ K such thatP

aiχi = 0, i.e.

Xaiχi(g) = 0, (∗)

for all g ∈ G. We can rewrite it as

Xaiχi(gh) = 0, (∗∗)

for all g, h ∈ G. Multiply (∗) by χ1(h), we get

Xaiχi(g)χ1(h) = 0. (∗ ∗ ∗)

Compare (∗) with (∗ ∗ ∗), we get

Xaii(h) − χ1(h))χi(g) = 0 for all g ∈ G.

Thus Pn

i=2aii(h) − χ1(h))χi = 0 ∈ Hom(G, K). It follows that the n − 1 elements {χ2, ..., χn} is linearly dependent, which is a contradic-

tion to the minimality. ¤

Corollary 3.10.3. Let F/K be an extension. The set of K-homomorphisms from F to K is linearly independent in the K-vector space of linear maps from F to K.

Sketch. Take G = F. ¤

Let K be a field containing n-th root of unity ζ. Let F/K be a Galois extension with Galois group ∼= Zn generated by σ. We consider

ψζ := 1 + ζσ + ζ2σ2+ ... + ζn−1σn−1 ∈ Hom(F, K).

Any element of the form ψ(x) is called a Lagrange resolvent.

By direct computation, we have the following properties.

(2)

Proposition 3.10.4. Keep the notation as above, we have:

1. σ(ψζ(x)) = ζ−1ψζ(x).

2. ψ1(x) ∈ K.

3. (ψζ(x))n ∈ K.

4. (ψζ(x))(ψζ−1(x)) ∈ K.

5. P

ζ∈µnζ−rψζ(x) = nσr(x).

Now we can use this technique to solve cubic equations. Let f (x) = x3 + px + q ∈ K[x] be an irreducible polynomial with discriminant D = −4p3− 27q2 ∈ K. We assume that K contains a primitive 3-root of unity ζ. We have extension K ⊂ L := K[√

D] ⊂ F := K[u1, u2, u3].

Note that F/L is Galois with Galois group ∼= Z3. Step 1. ψζ 6= 0 ∈ Hom(F, K), in fact ψζ(u1) 6= 0.

Step 2. ψζ(u1) 6= L and (ψζ(u1))3 ∈ L, thus F = L[ψζ(u1)].

And similarly, ψζ2(u1) ∈ L, (ψζ2(u1))3 ∈ L. Moreover, ψζ(u1ζ2(u1) ∈ L.

Step 3. Solve ψζ(u1),ψζ2(u1) . Recall that

∆ := (u1−u2)(u2−u3)(u3−u1) = u21u2+u32u3+u23u1−u1u22−u2u23−u3u21. ψζ(u1)3 = u31+u32+u33+3ζ(u21u2+u22u3+u23u1)+ζ2(u1u22+u2u23+u3u21)+6u1u2u3. Let v1 = u21u2+ u22u3+ u23u1, v2 = u1u22+ u2u23+ u3u21, then

v1+ v2 = (u1+ u2+ u3)(u1u2+ u2u3+ u3u1) − 3u1u2u3 = 3q, v1− v2 = ∆.

Thus ψζ(u1)3 can be expressed in terms of p, q, ∆.

Step 4. solve u1, u2, u3 in terms of ψζ(u1),ψζ2(u1).

By the property 5 above, we have

3u1 = ψ1(u1) + ψζ(u1) + ψζ2(u1),

3u2 = 3σ(u1) = ψ1(u1) + ζ−1ψζ(u1) + ζ−2ψζ2(u1), 3u3 = 3σ2(u1) = ψ1(u1) + ζ−2ψζ(u1) + ζ−1ψζ2(u1).

And note that ψ1(u1) = 0. So one can solve cubic polynomial explicitly.

3.11. cyclic extension. The discussion in the previous section can be extended to a more general setting.

Definition 3.11.1. We say that an extension is cyclic (resp. abelian) if it’s algebraic Galois and GalF/K is cyclic (resp. abelian). An cyclic extension of order n is an cyclic extension whose Galois group is iso- morphic to Zn.

The following theorem characterize cyclic extension except some ex- ceptional case.

(3)

Theorem 3.11.2. Suppose that char(K) = 0 or char(K) = p - n.

Suppose furthermore that there is a primitive n-th root of unity in K, say ζ. Then F/K is a cyclic of order n if and only if F = K(u) where u is a root of irreducible polynomial xn− a ∈ K[x].

Before we get into the proof. Let’s consider the ”difference” be- tween u and σ(u) for σ ∈ GalF/K. Let F/K be a finite Galois exten- sion. Then in this circumstance, norm and trace (which we will define more generally later) are nothing but NF/K(u) :=Q

σ∈GalF/Kσ(u) and TF/K := P

σ∈GalF/Kσ(u). It’s easy to see that T (u − σ(u)) = 0 and N(u/σ(u)) = 1. The follows lemma says that the converse is also true for cyclic extension, which will play the central role in the study of cyclic extension.

Lemma 3.11.3. Let F/K be an cyclic extension with σ the generator of the Galois group.

(1) If TF/K(u) = 0, then there exists an v ∈ F such that u = v − σ(v).

(2) (Hilbert’s Theorem 90) If NF/K(u) = 1, then there exists an v ∈ F such that u = v/σ(v).

Proof of the Theorem 3.11.2. Let u be a root of xn − a, then all the roots are uζi for i = 0, ..., n − 1. Since ζ ∈ K. We can produce an element in Galois group by considering σi : u 7→ uζi. Thus we have i}i=0,...,n−1 ⊂ GalKF . It’s clear that GalKF = {σi}i=0,...,n−1 =<

σ1 >. Thus F = K(u) is a cyclic extension over K.

Conversely, suppose that F/K is a cyclic extension of order n. Since there is a primitive n-th root ζ ∈ K, one has N(ζ) = ζn = 1. By the Lemma, there exist an v such that ζ = v/σ(v). Let u = v−1, then σ(u) = ζu. Hence σ(un) = un∈ K. Therefore u satisfies xn− a ∈ K[x]

for some a ∈ K.

Moreover, for uζi and uζj, there is an automorphism sending uζi to j. So they have the same minimal polynomial p(x) dividing xn− a.

One the other hand, p(x) has n distinct roots uζi for i = 0, ..., n − 1.

It follows that p(x) = xn− a is irreducible. One has [K(u) : K] = n

and thus F = K(u). ¤

Theorem 3.11.4. Suppose that char(K) = p 6= 0. Then F/K is a cyclic extension of order n if and only if F = K(u), where u is a root of an irreducible polynomial xp− x − a ∈ K[x].

Proof. The proof is parallel to the previous one.

Let u be a root of xp − x − a, then all the roots are u + i for i = 0, ..., p − 1. It’s clear that F = K(ζ) is a cyclic extension over K with Galois group generated by σ such that σ(u) = u + 1.

Conversely, suppose that F/K is a cyclic extension of order n. One has T (1) = p = 0. By the Lemma, there exist an v such that 1 =

(4)

v − σ(v). Let u = −v, then σ(u) = u + 1. Hence σ(up) = up+ 1 and σ(up− u) = up− u. Therefore u satisfies xp − x − a ∈ K[x] for some a ∈ K.

Moreover, for u + i and u + j, there is an automorphism sending i to uζj. So they have the same minimal polynomial p(x) dividing xp − x − a. One the other hand, p(x) has p distinct roots u + i for i = 0, ..., p − 1. It follows that p(x) = xp− x − a is irreducible. One

has [K(u) : K] = n and thus F = K(u). ¤

It remains to define norm and trace, and prove the main lemma 3.11.3.

Definition 3.11.5. Let [F : K] be a finite separable extension. Let Σ be the set of K-embeddings of F into K. For any u ∈ F , we define the norm, denoted

NF/K(u) := (Y

σ∈Σ

σ(u)).

Similarly, we define the trace as

TF/K(u) := (Σσ∈Σσ(u)).

Example 3.11.6. If F/K is finite Galois extension, then the set of all K-embeddings of F is nothing but the Galois group of F (since F is normal). Therefore, NF/K(u) = Q

σ∈GalF/Kσ(u) and TF/K(u) = P

σ∈GalF/Kσ(u)

Proof of Lemma 3.11.3. We only prove that T (u) = 0 implies u = v − σ(v). The other implication is easy.

Step 1. Find an element z ∈ F with T (z) 6= 0. This is an immediate consequence of independency of automorphism.

Step 2. We normalize it to get w ∈ F with T (w) = 1. In fact, we take w := T (z)z .

Step 3. Let

v = uw + (u + σ(u))σ(w) + ... + (u + σ(u) + ... + σn−2(u))σn−2(w).

Then by direct computation and T (u) =P

σ(u) = 0, we are done.

For the norm, if N(u) = 1, then u 6= 0. Take

v = uy + uσ(u)σ(y) + ... + uσ(u)...σn−1(u)σn−1(y).

By independency of automorphism, there exist a y such that v is non- zero. One checks that u−1v = σ(v). We are done. ¤ 3.12. radical extension.

Definition 3.12.1. F/K is said to be an radical extension if F = K(u1, ..., un) such that for 1 ≤ i ≤ n, unii ∈ K(u1, ..., un−1).

For a polynomial f (x) ∈ K[x]. We say f (x) = 0 is solvable by radical if its splitting field E is contained in some radical extension.

(5)

Remark 3.12.2. In the definition, it’s not necessary that the splitting field itself is a radical extension over K.

The main observation is the following:

Proposition 3.12.3. Let F/K be a radical and Galois extension over K. Write F = K(u1, ..., un) such that for 1 ≤ i ≤ n, unii ∈ K(u1, ..., un−1).

Let m = Q

ni and assume that char(K) - m. Suppose furthermore that K contains a primitive m-th root of unity. Then GalF/K is solvable.

Proof. Let Ki := K(u1, ..., ui). And let Gi = Ki0. One sees that K1 is cyclic over K, hence Galois over K. Hence G1 C G0 = GalF/K. Consider next F/K1 which is radical and Galois. Then K2 is cyclic over K1 and hence similarly, G2C G1. Therefore, we have a solvable series {e} = GnC Gn−1C ... C G0 = GalF/K with Gi−1/Gi cyclic. We

are done. ¤

One can actually generalize it to the following general setting:

Theorem 3.12.4. Let F/K be a radical extension, and K ⊂ E ⊂ F . Then GalE/K is solvable. As a consequence, if f (x) = 0 is solvable by radical, then Gf is solvable.

Proof. We first reduce to simpler situation.

Step 1. Let G = GalE/K and K0 = G0. It’s clear that F/K0 is radical, and E/K0 is Galois for GalE/K0 = G00 = G and G000 = G0. Thus F/K0 is radical and E/K0 is Galois with Galois group GalE/K.

We thus replacing K by K0 and assume that E/K is Galois.

Step 2. Reduce to the case that E = F/K is Galois. To see this, let σ : F → K be an K-embedding. One can show that σ(F ) is again a radical extension. One can also prove that if F1, F2 ⊂ K are radical extension over K, then F1F2 is a radical extension over K. Hence let N be the compositum of σ(F ) for all σ. It follows that N is radical over K. Moreover, N is normal over K.

Since E/K is Galois, in particular, E is normal over K and E is a stable intermediate subfield of N/K. Then one has a homomorphism GalN/K → GalE/K. This is surjective because N is normal. Thus it suffices to prove that GalN/K is solvable.

Step 3. By the same trick an in Step 1. We may assume that N/K is Galois. Therefore, it suffices to show that if F/K is Galois and radical, then GalF/K is solvable.

Step 4. Since F/K is separable, we may assume that (char(K), ni) = 1. Let m =Q

ni.

Let ζ be a primitive m-th root of unity. We claim that F (ζ) is Galois over K. Grant this for the time being, then F (ζ) is Galois over K(ζ) and K(ζ)0 C GalF (ζ)/K. Moreover, GalF (ζ)/K/K(ζ)0 = GalK(ζ)/K. By Proposition 3.12.3, K(ζ)0 is solvable. K(ζ)/K is cyclotomic, hence GalK(ζ)/K is solvable. Thus, GalF (ζ)/K is solvable.

(6)

Now F/K is Galois, GalF/K = GalF (ζ)/K/F0 which is solvable.

Step 5. To prove the claim, suppose that F is a splitting field of separable polynomial f1, .., fn ∈ K[x]. Then F (ζ) is nothing but a splitting field of separable polynomials f1, ..., fn, xm − 1. Thus we are

done. ¤

Theorem 3.12.5. Let E be a finite dimensional Galois extension over K with solvable Galois group. Assume that char(K) - [E : K], then there is a radical extension F/K containing E.

Proof. We prover by induction on [E : K]. Let n = [E : K] and assume the theorem is true for all Galois extension of degree < n.

Let ζ be a primitive n-th root of unity. Then E(ζ)/K(ζ) is Galois.

If [E(ζ) : K(ζ)] < n then we are done by induction hypothesis and the fact that K(ζ)/K is radical.

By replacing E, K by E(ζ), K(ζ) respectively, we my assume that K has m-th root of unity.

GalE/K is solvable, let H be a subgroup of index q, for some prime q. Then H0/K is a cyclic extension, hence a radical extension. By induction hypothesis, E/H0 is radical. We are done. ¤ Corollary 3.12.6. Let f (x) ∈ K[x] be a polynomial of degree n > 0.

Suppose that char(K) - n!, then f (x) = 0 is solvable by radical if and only if Gf is solvable.

參考文獻

相關文件

[r]

Describe the Galois group G(K/Q) by the fundamental theorem of finitely-generated abelian groups.. Show that the order of the Galois group G(E/Q)

If it’s abelian, then it’s cyclic by fundamental the- orem of abelian groups plus Chinese remainder theorem.. Let’s suppose that

We conclude this section with the following theorem concerning the relation between Galois extension, normal extension and splitting fields..

Hence prove that any finite cyclic group is isomorphic to a direct product of cyclic groups of prime power

We conclude this section with the following theorem concerning the relation between Galois extension, normal extension and splitting fields..

It’s clear that the Galois group G f

Here is