Dienstag, 24. Januar 2017
A note in Galois Theory
Theorem 1:
Let G={σ1,⋯,σn} be a finite group, x1,⋯,xn be indeterminates
and let ρ be the regular representation of G. Let x=(x1,⋯,xn)T and G=(ρ(σ1)x,⋯,ρ(σn)x) , A′=diag(x1,⋯,xn).
Set A=GA′G−1. Suppose there exists α:=(α1,⋯,αn)∈Cn such that:
1) det(G(α))≠0
2) A=A(α) has rational coefficients
3) the characteristic polynomial χA(t) is irreducible over Q.
Then Gal(χA(t)/Q)=G.
Proof:
On the one hand we have
GA′=AG=GA′=(x1ρ(σ1)x,⋯,xnρ(σn)x)
On the other hand we have
GA′=AG=A(ρ(σ1)x,⋯,ρ(σn)x)=(Aρ(σ1)x,⋯,Aρ(σn)x)
and it follows that
(Aρ(σ1)x,⋯,Aρ(σn)x)=GA′=(x1ρ(σ1)x,⋯,xnρ(σn)x)
and so we get
Aρ(σi)x=xiρ(σi)x for i=1,⋯,n
Let K:=Q(α1,⋯,αn). From Aρ(σi)α=αiρ(σi)α follows
for i=1, ρ(σ1)=1 that Aα=α1α. Application of τ∈Gal(K/Q)
at the coefficients of the last equation gives us with τ(A)=A (since A has rational coefficients)
that
Aτ(α)=τ(α1)τ(α)=αττ(α)
where we have set ατ=τ(α1). This means that τ(α) is an eigenvector of A to the
eigenvalue ατ. But because of equation (1) we have:
Aρ(στ)α=ατρ(στ)α
Since the eigenvectors ρ(σ1)α,⋯,ρ(σn)α of A build a basis of Kn, it means
that every eigenspace has dimension 1. Especially there exists aτ∈K such that
τ(α)=aτρ(στ)α
If we build the polynomials with roots from the coefficients of the left and right side of (4), then they must
be because of (4) be equal:
χA(t)=n∏i=1(t−αi)=(4)n∏j=1(t−aταj)
This means that the coefficients of both polynomials must be equal, especially we get
n∑i=1αi=aτn∑i=1αi
By a theorem of Frobenius concerning the factorization of the group determinant, we have
det(G(x1,⋯,xn))=(x1+⋯+xn)p(x1,⋯,xn)
for some polynomial p. From (7) follows by assumption
0≠det(G(α1,⋯,αn))=(α1+⋯+αn)p(α1,⋯,αn)
which means that 0≠α1+⋯+αn and because of (6) we get
aτ=1
From (4) and (8) it follows that
Pτα=τ(α)=ρ(στ)α
where Pτ is the permutation matrix which corresponds to τ. Since
χA(t)=n∏i=1t−αi
is separable, it follows that αk≠αl for k≠l.
This means that the two permutation matrices Pτ and ρ(στ) from (9) must be equal,
that is
Pτ=ρ(στ)
From this we get a mapping:
ϕ:Gal(K/Q)→G,τ↦στ
ϕ is injective:
Let στ=ϕ(τ)=ϕ(τ′)=στ′. Then we get
Pτ=(10)ρ(στ)=ρ(στ′)=(10)Pτ′
and since the regular representation P is faithful, it follows that τ=τ′.
ϕ is surjective:
Let m=|Gal(K/Q)|, n=|G|=deg(χA(t)).
Since χA(t) is irreducible, deg(χA(t))=n is a divisor of m=|Gal(K/Q)| and
it follows that n≤m. Since ϕ is injective it follows that m≤n, which means that m=n.
Since the mapping ϕ is injective on two finite sets of equal size, this means that ϕ must also
be surjective.
ϕ is an antihomomorphism of groups:
τ(α)=ρ(σm)α→ϕ(τ)=σm
τ′(α)=ρ(σk)α→ϕ(τ′)=σk
(ττ′)(α)=ρ(σl)α→ϕ(ττ′)=σl
From this it follows that:
ρ(σl)α=(ττ′)(α)=τ(τ′(α))=τ(ρ(σk)α)=ρ(σk)τ(α)=ρ(σk)ρ(σm)α
Since αi are separable, it follows that the permutation matrices ρ(σl) and ρ(σkσm) must be equal:
ρ(σl)=ρ(σkσm)
and from this, since the regular representation is faithful, it follows that σl=σkσm.
From this we get:
ϕ(ττ′)=σl=σkσm=ϕ(τ′)ϕ(τ)
Since for two finite groups which are antiisomorph it follows that they are isomorph (if ϕ(xy)=ϕ(y)ϕ(x) is an antihomomorphism, then Φ(x):=ϕ(x)−1 is a homomorphism),
we conclude that
Gal(K/Q)=G
Definition:
Let G be a finite group with n elements. Let Q[x1,⋯,xn]G=Q[g1,⋯,gm]. Then there exist polynomials sj∈Q[y1,⋯,ym] for j=1,⋯,n such that
ej(x1,⋯,xn)=sj(g1(x1,⋯,xn),⋯,gm(x1,⋯,xn)) for j=1,⋯,n where ej is the j-elementary symmetric polynomial in xk. Consider the polynomial
p(t,y1,⋯,ym)=tn−s1(y1,⋯,ym)tn−1+⋯+(−1)nsn(y1,⋯,ym) which is a polynomial in Q[t,y1,⋯,ym]. We call p the corresponding polynomial of the group G to the invariants
g1,⋯,gm. Let I={h∈Q[y1,⋯,ym]|h(g1,⋯,gm)=0}. Define the variety VG={a∈Qm|h(a)=0∀h∈I}. We say that a polynomial is Hilbert irreducible over some variety,
if it is irreducible and there exists a point in the variety such that the polynomial remains irreducible after specializing to the point of the variety and that other polynomials are not zero on this point.
Lemma 1:
The corresponding polynomial is irreducible in Q[t,y1,⋯,ym].
Proof:
If the polynomial factored in a non-trivial way, then because of the tn term the factors must have degree less than n in t (consider the factorization in R=Q(y1,y2,…,yn)[t]; note also that we can assume that the factors over R are monic polynomials in t). Now specialise via yi↦gi and we get a non-trivial factorization of the specialised polynomial in Q[g1,…,gm][t] and hence in Q[x1,…,xn][t] and so in Q(x1,…,xn)[t]. But we know the complete factorization in this ring, it's just ∏(t−xi), so our given factorization must specialise into factors of the form ∏i∈I(t−xi) for some subsets I of {1,2,…,n} (with each I not empty or the whole thing), and the constant term of each factor must be in Q[g1,…,gm] and hence G-invariant. This is a contradiction.
Lemma 2:
Let f1,⋯,fr∈Q(x1,⋯,xn)G and suppose that the corresponding polynomial of G is Hilbert irreducible over the variety VG.
Then there exists α=(α1,⋯,αn)∈Cn such that:
1) f1(α),⋯,fr(α) are rational numbers.
2) n∏i=1t−αi is irreducible in Q[t].
Proof:
We can write
fi=Pi(g1,⋯,gm)Qi(g1,⋯,gm) for Pi,Qi∈Q[y1,⋯,ym]
The corresponding polynomial p(t,y1,⋯,ym) is irreducible in Q[t,y1,⋯,ym] by Lemma 1. By assumption, we can find (a1,⋯,am) in VG such that:
1) p(t,a1,⋯,am) is irreducible in Q[t].
2) Qi(a1,⋯,am)≠0 for i=1,⋯,r.
By application of the Hilbert Nullstellensatz we can find α=(α1,⋯,αn)∈Cn such that gi(α)=ai for i=1,⋯,m.
Then
ej(α)=sj(g1(α),⋯,gm(α))=sj(a1,⋯,am)
and the polynomial p(t,a1,⋯,am) factors over C as
p(t,a1,⋯,am)=n∏i=1t−αi.
Then we get
fi(α)=Pi(g1(α),⋯,gm(α))Qi(g1(α),⋯,gm(α))=Pi(a1,⋯,am)Qi(a1,⋯,am)
is a rational number, which was to be shown.
Lemma 3:
Let the notation be as in Theorem 1. Then the entries aij of the matrix A are elements of Q(x1,⋯,xn)G.
Proof:
We have A(x)=G(x)A′(x)G(x)−1 and want to show that A(ρ(σ)x)=A(x) for all σ∈G.
For this, it is sufficient to show that
A′(ρ(σ)x)=ρ(σ)A′(x)ρ(σ)−1
and
G(ρ(σ)x)=G(x)ρ(σ)−1
as it follows that
A(ρ(σ)x)=G(ρ(σ)x)A′(ρ(σ)x)G(ρ(σ)x)−1
=G(x)ρ(σ)−1ρ(σ)A′(x)ρ(σ)−1ρ(σ)G(x)−1
=G(x)A′(x)G(x)−1=A(x)
Corollary 1:
Let G be a finite group such that the correspoding polynomial is Hilbert irreducible over VG.
Then G is realizable over Q as a Galois group.
Proof:
Consider the regular representation of G and let A=(aij) be defined as in Theorem 1.
Then by Lemma 3 we have aij∈Q(x1,⋯,xn)G. Hence
by Lemma 2 there exists α∈Cn such that
1) aij(α) are rational numbers.
2) The characteristic polynomial χA(t)=n∏i=1t−αi is irreducible in Q[t].
Since aij(α) are rational numbers, the matrix A(α)=G(α)A′(α)G(α)−1
is defined for α, hence det(G(α))≠0.
By Theorem 1 therefore it follows that Gal(χA(t)/Q)=G.
Theorem 2:
The following are equivalent:
1) Hilbert Irreducibility can be applied to the corresponding polynomial p(t,y1,⋯,ym) over the variety VG
2) Every finite group G is Galois over Q.
Proof:
1) → 2): Corollary 1.
2) → 1)
Let G=Gal(K/Q)={σ1,⋯,σn} be a Galois group. By the normal
basis theorem there exists a θ∈K such that σ1(θ),⋯,σn(θ) build a
basis of K over Q. Let Q[x1,⋯,xn]G=Q[g1,⋯,gm] where
G acts through the regular representation. Let I={h∈Q[y1,⋯,ym]|h(g1,⋯,gm)=0}
and VG={a∈Qm|h(a)=0∀h∈I}.
Let now Q1,⋯,Qr∈Q[y1,⋯,ym] be given, such that Qi∉I for i=1,⋯,r. Let Q:=Q1⋯Qr.
Since I is a prime ideal and Qi∉I, we must have Q∉I, which means that there exists b∈Qn such that
Q(g1(b),⋯,gm(b))≠0
that is ˆf(x1,⋯,xn)=Q(g1(x1,⋯,xn),⋯,gm(x1,⋯,xn)) is a reduced polynomial ( since Q has infinite elements)
(see Algebra, Lang, p. 177, p. 312).
Since the σi by Artin (p. 311, Th. 12.2 Algebra, Lang) are algebraically independent and ˆf is reduced,
there exists α∈K, such that
0≠Q(g1(σ1(α),⋯,σn(α)),⋯,gm(σ1(α),⋯,σn(α)))
Let α=n∑i=1αiσi(θ) with αi∈Q
Then we have for i=1,⋯,m that gi(σ1(α),⋯,σn(α))=pi(α1,⋯,αn) is a rational number,
where pi is some polynomial in Q[x1,⋯,xn].
But then also
0≠Q(g1(σ1(α),⋯,σn(α)),⋯,gm(σ1(α),⋯,σn(α)))
=Q(p1(α1,⋯,αn),⋯,pm(α1,⋯,αn))
=q(α1,⋯,αn)
for some polynomial q∈Q[x1,⋯,xn]. We can choose the αi such that αi≠αj for i≠j,
otherwise replace ˆαj=αi+u for some rational number u≠0 and we get the polynomial in u:
q(u):=q(α1,⋯,αj−1,αi+u,αj+1,⋯,αn)
Since q(0)≠0, q is not the zero polynomial in Q[u]. By choosing u≠0 as a rational number to avoid the
finitely many roots of q, we get q(u)≠0 and ˆαj=αi+u≠αi.
But then also σi(α)≠σj(α) for i≠j, since α has pairwise distinct coordinates αk≠αl to
the given basis σ1(θ),⋯,σn(θ). The rational and separable polynomial
n∏i=1(x−σi(α))
is irreducible in Q[x], since the Galois group operates transitively on the roots.
Let ai:=gi(σ1(α),⋯,σn(α)) for i=1,⋯,m.
Then we have a:=(a1,⋯,am) is an element of VG and also
p(t,a1,⋯,am)=n∑i=0(−1)isi(a1,⋯,am)tn−i
=n∑i=0(−1)isi(g1(σ1(α),⋯,σn(α)),⋯,gm(σ1(α),⋯,σn(α)))tn−i
=n∑i=0(−1)iei(σ1(α),⋯,σn(α))tn−i
=n∏i=1(t−σi(α))
is an irreducible polynomial in Q[t]. Furthermore we have
Q(a1,⋯,am)≠0, which means that Qj(a1,⋯,am)≠0 for all j=1,⋯,r, which was to be shown.
This note can be downloaded from here (A note in Galois Theory, Orges Leka) as a pdf file.
Abonnieren
Posts (Atom)