Bourbaki

§ 18. Simple algebras over a finite field

1. Polynomials over a Finite Field

Theorem 1 (Chevalley–Warning) 00SA

Let K be a finite commutative field of characteristic pp. Let nn be an integer ⩾1\geqslant 1 and (fi)i∈I(f_i)_{i\in I} a finite family of nonzero elements of K[X1,...,Xn]K[X_1, . . . ,X_n]. Denote by Z the set of elements x\mathbf{x} of KnK^n such that we have fi(x)=0f_i(\mathbf{x}) = 0 for i∈Ii\in I. If we have n>∑i∈In >\sum_{i\in I} deg(fif_i), then the cardinal of Z is divisible by pp.

Lemma 1 00LB

Let L be a field, G a finite group, and χ\chi a nontrivial homomorphism from G to the multiplicative group L∗L^*. We have ∑x∈Gχ(x)=0\sum_{x\in G}\chi (x) = 0.

By assumption, there exists an element aa of G such that χ(a)≠1\chi (a)\not= 1; since multiplication by aa is a permutation of G, we have

∑x∈Gχ(x)=∑x∈Gχ(ax)=χ(a)∑x∈Gχ(x) \sum_{x\in G}\chi (x) =\sum_{x\in G}\chi (ax) =\chi (a)\sum_{x\in G}\chi (x)

this gives Lemma 1.

Lemma 2 00LC

Let qq be the cardinal of K. For any integer m⩾0m\geqslant 0, set Sm=S_m= ∑x∈Kxm\sum_{x\in K}x^m. We have Sm=−1S_m=-1 if mm is a nonzero multiple of q−1q-1 and Sm=0S_m= 0 in all other cases.

Recall that 00=10^0= 1 (I, §2, No. 1, p. 14). Suppose that mm is a multiple of q−1q-1. Since the abelian group K∗K^* has order q−1q-1, we have xm=1x^m= 1 for every x∈K∗x\in K^* and Sm=0m+(q−1)⋅1S_m= 0^m+ (q-1)\cdot 1, which gives the assertion in this case.

Suppose that mm is not a multiple of q−1q-1. Set χ(x)=xm\chi (x) =x^m for x∈K∗x\in K^*. Since the multiplicative group K∗K^* is cyclic of order q−1q-1 (V, §12, No. 1, p. 93, Proposition 1), there exists an element aa of K∗K^* such that χ(a)≠1\chi(a) \neq 1. By Lemma 1 applied to G=K∗G = K^*, we have

Sm=0m+∑x∈K∗χ(x)=0; S_m = 0^m + \sum_{x \in K^*} \chi(x) = 0;
this gives Lemma 2.

Let us now prove Theorem 1. Let x=(x1,…,xn)\mathbf{x} = (x_1, \ldots, x_n) be an element of KnK^n. We have 1−fi(x)q−1=01 - f_i(\mathbf{x})^{q-1} = 0 if fi(x)≠0f_i(\mathbf{x}) \neq 0 and 1−fi(x)q−1=11 - f_i(\mathbf{x})^{q-1} = 1 if fi(x)=0f_i(\mathbf{x}) = 0. Set P=∏i=1r(1−fiq−1)P = \prod_{i=1}^r (1 - f_i^{q-1}). We have

P(x)={1if x∈Z,0if x∉Z. P(\mathbf{x}) = \begin{cases} 1 & \text{if } \mathbf{x} \in \mathbf{Z}, \\ 0 & \text{if } \mathbf{x} \notin \mathbf{Z}. \end{cases}
Let us expand the polynomial PP as ∑α∈NncαXα\sum_{\alpha \in \mathbf{N}^n} c_\alpha X^\alpha; by assumption, it has degree <(q−1)n< (q-1)n. Let α\alpha be an element of Nn\mathbf{N}^n such that cαc_\alpha is nonzero. Since we have α1+⋯+αn<(q−1)n\alpha_1 + \cdots + \alpha_n < (q-1)n, there exists an integer ℓ\ell such that 1≤ℓ≤n1 \leq \ell \leq n and 0≤αℓ<q−10 \leq \alpha_\ell < q-1. By Lemma 2, we then have ∑x∈Kxαℓ=0\sum_{x \in K} x^{\alpha_\ell} = 0, and therefore
∑x∈Knxα=∏j=1n(∑x∈Kxαj)=0. \sum_{\mathbf{x} \in K^n} \mathbf{x}^\alpha = \prod_{j=1}^n \left( \sum_{x \in K} x^{\alpha_j} \right) = 0.
We consequently have
∑x∈KnP(x)=∑α∈Nncα(∑x∈Knxα)=0. \sum_{\mathbf{x} \in K^n} P(\mathbf{x}) = \sum_{\alpha \in \mathbf{N}^n} c_\alpha \left( \sum_{\mathbf{x} \in K^n} \mathbf{x}^\alpha \right) = 0.
Now, by formula (1), we have ∑x∈KnP(x)=Card⁡(Z)⋅1\sum_{\mathbf{x} \in K^n} P(\mathbf{x}) = \operatorname{Card}(\mathbf{Z}) \cdot 1, and therefore Card⁡(Z)⋅1=0\operatorname{Card}(\mathbf{Z}) \cdot 1 = 0, which means that Card⁡(Z)\operatorname{Card}(\mathbf{Z}) is divisible by pp.

Corollary 00LD

Let VV be a vector space of finite dimension nn over KK and II a finite set, and for each i∈Ii \in I, let Fi:V→KF_i : V \to K be a homogeneous polynomial mapping of degree di>0d_i > 0. If we have ∑i∈Idi<n\sum_{i \in I} d_i < n, then there exists a nonzero element xx of VV such that Fi(x)=0F_i(x) = 0 for every i∈Ii \in I.

Let (e1,…,en)(e_1, \ldots, e_n) be a basis of VV over KK. By the definition of homogeneous polynomial mappings (IV, §5, No. 9, p. 55, Definition 3), for every i∈Ii \in I, there exists a homogeneous polynomial fif_i in K[X1,…,Xn]K[X_1, \ldots, X_n] of degree did_i such that we have Fi(ξ1e1,…,ξnen)=fi(ξ1,…,ξn)F_i(\xi_1 e_1, \ldots, \xi_n e_n) = f_i(\xi_1, \ldots, \xi_n). Let ZZ be the set of elements xx of VV such that Fi(x)=0F_i(x) = 0 for every i∈Ii \in I. By Theorem 1, the cardinal of ZZ is divisible by pp, and since 00 belongs to SS, we have Card⁡(Z)≥p>1\operatorname{Card}(Z) \geq p > 1.

2. Simple Algebras over Finite Fields

Theorem 2 (Wedderburn) 00SB

Every finite field is commutative.

Let D be a finite field, and let K be its center. The K-algebra D is central simple of degree m2m^2, where mm is a strictly positive integer. The reduced norm is a homogeneous polynomial mapping Nrd:D→K: D\rightarrow K of degree mm (VIII, p. 345, Proposition 6), and we have Nrd(aa)≠0\not= 0 for every a≠0a\not= 0 in D (VIII, p. 340, Proposition 3). The corollary above implies that m⩾m2m\geqslant m^2, and therefore m=1m= 1. So we have D = K.

Corollary 1 00LE

Every finite simple ring is isomorphic to a matrix ring Mn(L)\mathbf{M}_n(L), where nn is a strictly positive integer and L is a finite commutative field.

This follows from Theorem 2 and the structure theorem for simple rings (VIII, p. 120, Theorem 1).

Corollary 2 00LF

Let K be a finite commutative field. Every central simple algebra over K is isomorphic to a matrix algebra Mn(K)\mathbf{M}_n(K), where nn is a strictly positive integer.

This follows from Theorem 2 and the structure theorem for central simple algebras (VIII, p. 252, Theorem 1).

Remark 1 00LG

Here is another proof of Theorem 2. Let D be a finite field, let K be its center, and let L be a maximal commutative subfield of D. Let xx be an element of D∗D^*. It belongs to a maximal commutative subfield L1L_1 of D. We have the equality

[D:K]=[L:K]2=[L1:K]2 [D : K] = [L : K]^2= [L_1: K]^2

by Corollary 2 of VIII, p. 265, and therefore [L:K]=[L1:K][L : K] = [L_1: K]. By Proposition 3 of V, §12, No. 2, p. 94, the extensions L and L1L_1 of K are isomorphic. By VIII, p. 263, Corollary, there exists an element aa of D∗D^* such that aLa−1=L1aLa^{-1}= L_1, so a−1xaa^{-1}xa belongs to L. We then have (ay)−1x(ay)=a−1xa(ay)^{-1}x(ay) =a^{-1}xa for every y∈L∗y\in L^*. Consequently, if S is a set of representatives of the left cosets of D∗D^* modulo L∗L^*, then every element of D∗−{1}D^*-\{1\} can be written as sxs−1sxs^{-1} with s∈Ss\in S and x∈L∗−{1}x\in L^*-\{1\}. We denote the order of D∗D^* by dd and that of L∗L^* by ℓ\ell. Since the cardinal of S is equal to d/ℓd/\ell, we have d−1⩽(d/ℓ)(ℓ−1)=d−d/ℓd-1\leqslant (d/\ell )(\ell -1) =d-d/\ell. It follows that ℓ=d\ell =d, and therefore L = D, which proves that the field D is commutative.

Remark 2 00LH

Let L be a commutative field with the following property: (C1)(C_1) Let V be a vector space of finite dimension nn over the field L, and let dd be such that 0<d<n0< d < n. For every homogeneous polynomial mapping F:V→LF : V\rightarrow L of degree dd, there exists a nonzero element xx of V such that F(x)=0F(x) = 0.

The proof of Theorem 2 shows that every field of finite degree over L with center L is equal to L.

By the corollary of VIII, p. 356, every finite field has property (C1)(C_1). We can prove (VIII, p. 360, Exercise 7) that the following fields have property (C1):(C_1):

– every algebraic extension of a finite field

– every field of rational fractions in one variable with coefficients in an algebraically closed field (Tsen’s theorem).

– ∗*every field endowed with a discrete valuation for which it is complete and

whose residue field is algebraically closed (VIII, p. 332, Exercise 17).∗.*

Remark 3 00LI

Suppose that the field K satisfies the following condition:

– If L is an extension of K of finite degree, then it is cyclic and the norm mapping N:L∗→K∗N : L^*\rightarrow K^* is surjective.

This condition is satisfied, in particular, when the field K is finite (V, §12, No. 2, p. 95, Proposition 4). We can then prove that every field of finite degree over K with center K is equal to K (Exercise 10 of VIII, p. 329).

Exercises

See the exercises for § 18.

Transcribed from the printed edition, 2023, Springer Nature. Not translated.