Irish Math. Soc.

Bulletin
Number 88, Winter 2021, 57–68
ISSN 0791-5578

Four Group-theoretic Proofs of Wedderburn’s Little Theorem

ALAN ROCHE

Abstract. Wedderburn proved in 1905 that a finite division ring is always a field.
His result has intrigued generations of mathematicians, spurring generalizations and
alternative proofs. The shortest, most elegant proof is surely Witt’s from 1931, now
the standard textbook treatment. Following a strategy of Zassenhaus, we present four
overlapping group-theoretic proofs. The first uses a counting argument; the others
hinge on properties of special classes of finite groups.

1. Introduction
Wedderburn’s Little Theorem says that a finite division ring is a field. That is, if each
nonzero element of a finite ring with identity has a multiplicative inverse then the ring is
commutative. In the words of I. N. Herstein, the theorem “has caught the imagination
of mathematicians because it is so unexpected, interrelating two seemingly unrelated
things, the number of elements in a certain algebraic system and the multiplication of
that system” [9]. In the words of Emil Artin, “this result of Wedderburn has fascinated
most algebraists to a very high degree” [2]. As often in mathematics, the first proofs
were somewhat clumsy. Quoting Artin again, Wedderburn [17] makes use of “divisibility
properties which are hard to establish” and “several attempts were made to simplify the
proofs.” In 1931, Witt gave a short, elegant proof. It uses only elementary group theory
in the form of the class equation (which Wedderburn also used) and easy divisibility
properties. It is indeed a proof from THE BOOK (see [1, Chap. 6]).
Witt’s argument, however, was far from the last word. There have been many other
proofs—ones like Witt’s that use essentially elementary methods and ones that stem
from a broader perspective. As an example of the latter, note that Wedderburn’s Big
Theorem [18] puts the Little Theorem in a wider setting—the theory of central simple
algebras and the Brauer group. This is a profound theory concerned, in its classical
incarnation, with describing division algebras that are finite dimensional over their
centres. Various results in the overall theory yield proofs of the Little Theorem. For
instance, by properties of what are called cyclic algebras the theorem reduces to an easy
calculation—checking surjectivity of the norm map for a finite extension of finite fields
(see, for example, [10, Section 8.4]).
Our starting point is a technical lemma due to Zassenhaus [20]. It says that if E is
a subfield of a finite division ring D then the normalizer and centralizer of E× in D×
coincide. Zassenhaus goes on to deduce the Little Theorem by a purely group-theoretic
argument (see Remark 1 below). We follow the same path: with Zassenhaus’s lemma
in hand, we apply tools from finite group theory to obtain the Little Theorem—four
times. Out of a perverse sense of symmetry, we also include four proofs of Zassenhaus’s
lemma.

2020 Mathematics Subject Classification. 20D99, 16K20.

Key words and phrases. division ring, finite solvable group, Frobenius group.
Received on 24-7-2021; revised 19-11-2021.
DOI:10.33232/BIMS.0088.57.68.

c 2021 Irish Mathematical Society

57
58 ALAN ROCHE

The first proof of the Little Theorem is direct. Beyond the technical lemma, it
uses only elementary group theory and simple counting. The other proofs rely on
more advanced material: Burnside’s Normal Complement Theorem, Carter subgroups,
a property of Frobenius groups. For anyone new to these topics, we’ve tried to fill in
enough background so that the bulk of the paper still makes sense. More precisely, a
reader on friendly terms with (a) finite groups up to the Sylow theorems plus the notion
of a nilpotent group and (b) fields and basic Galois theory should be able to follow the
main thread of argument.
Three of our proofs share a common strategy: they proceed by showing that the
maximal fields in a noncommutative finite division ring D (with D sometimes of minimal
order) form a single conjugacy class (under the action of D× ). Since each element of D
is contained in a maximal field, it follows that if E is a fixed maximal field in D then the
group D× is a union of conjugates of E× . A finite group, however, is never a union of
conjugates of a proper subgroup (see Lemma 2 below), and so a noncommutative finite
division ring cannot exist. The same strategy underlies the classic proof of the Little
Theorem in van der Waerden’s Moderne Algebra, the influential text based on lectures
by Artin and Noether, first published in 1930–31. Van der Waerden’s argument, which
is due to Noether, uses early results in the theory of central simple algebras. It is
probably the most reproduced proof after Witt’s of the Little Theorem. 1
Except for the first, our proofs are more artificial than Zassenhaus’s. We could be
accused (we have been) of using a series of sledgehammers to crack the Little Theorem.
But they are such beautiful sledgehammers. The actual arguments, once the necessary
background is in place, are short, requiring just a tap from any sledgehammers we
wield. The proofs are meant as mathematical entertainments—an amusing application,
we hope, of parts of finite group theory.
Acknowledgements. The treatment of Zassenhaus’s lemma in Section 7 owes much to com-
ments provided by a reviewer of a previous (pre-Bulletin) version of the paper. In addition, I’ve
adopted several suggestions made by the referee of the current (Bulletin) version.
I would also like to note an older, more personal debt to two master expositors of mathematics,
T. J. Laffey and D. L. McQuillan, under whose stimulating guidance I first learned of the world
containing Wedderburn’s theorem.

Notation
Throughout D denotes a finite divison ring. The centre of D is a finite field which we
write always as F. For any ring R (with identity), R× denotes the group of units of R.
We write q for the cardinality of F and q N for the cardinality of D, so that N is the
dimension of D as an F-vector space.
For H a subgroup of a group G, the normalizer and centralizer of H in G are written
as N (H) and C(H). The ambient group G should be clear from context. In fact,
we work mostly with a subfield E of D. Then N (E× ) and C(E× ) always mean the
normalizer and centralizer in D× (that is, the ambient group is invariably D× ).

1Van der Waerden, Witt and Zassenhaus were part of the remarkable flowering of abstract algebra
in Germany in the 1920s and early 1930s in which Artin at Hamburg and Noether at Göttingen were
leading figures. Witt was Noether’s doctoral student and Zassenhaus was Artin’s. Van der Waerden
studied at Göttingen and was in the group (or ring) of young mathematicians centred around Noether;
he also spent a year at Hamburg where he worked closely with Artin [16]. And then the cataclysm.
In March 1933 the Nazis took power. In April a decree dismissed “non-Aryans” from government
institutions including universities. Within weeks, a vibrant mathematical culture was destroyed [12].
Hilbert, when asked in 1934 by the new minister of education about mathematics at Göttingen “now
that it was freed of the Jewish influence,” replied: “Mathematics at Göttingen? There is really none
anymore” [13, p. 205].
 Wedderburn’s Little Theorem 59

For S a finite set, |S| denotes the number of elements in S. Given a subset T of S,
we write S \ T for the difference of S and T , that is, the set of elements of S that do
not belong to T .

2. Zassenhaus’s Lemma
The following technical result, due to Zassenhaus ([20, p. 59]), underpins each of our
approaches to Wedderburn’s theorem.
Lemma 1. If E is a subfield of a finite division ring D, then the normalizer and cen-
tralizer of E× in D× coincide.
We only need the statement for now and so defer the task of proving the lemma to
Section 7.
Remark 1. The formulation in [20] is superficially different: it says that if A is an
abelian subgroup of D× then N (A) = C(A). To obtain this version for a given abelian
subgroup A of D× , simply apply Lemma 1 to F(A), the subfield of D generated over F
by A. Zassenhaus goes on to prove the Little Theorem by establishing a pleasant result:
if a finite group G has the property that N (A) = C(A) for every abelian subgroup A
then G is abelian ([20, Theorem 7]).
Remark 2. Let E be a subfield of D that strictly contains the centre F, so that the
automorphism group Aut(E/F) is nontrivial. A foundational result in the theory of
central simple algebras—the Skolem–Noether Theorem—implies that each element of
Aut(E/F) is implemented by conjugation by some element of D× (see, for example,
[5, Theorem 3.14]). The action of N (E× ) on E by conjugation therefore induces an
isomorphism N (E× )/C(E× ) ≃ Aut(E/F), and so N (E× ) 6= C(E× ). Thus, if we were
willing to make use of the theory of algebras, Lemma 1 yields a one-line (or one-
paragraph) proof of the Little Theorem.

3. First Proof: Counting Maximal Fields

We assume that D has minimal order among noncommutative finite division rings
and hope to find a contradiction.
Let E and E′ be distinct maximal fields in D. Then the subring hE, E′ i generated by
E and E′ is noncommutative. It’s also a division ring: in fact, each nonzero subring of
D is a division ring (since each element of D× has finite order). By minimality of |D|,
it follows that hE, E′ i = D. Thus E ∩ E′ lies in the centre of D, and so
E ∩ E′ = F. (1)
Recall our notation: |F| = q and |D| = q N . Let E1 , . . . , Er be representatives of
the distinct conjugacy classes of maximal fields in D. We write |Ei | = q mi for integers
mi ≥ 2 (for i = 1, . . . , r). We use some counting and a simple estimate to show that
r = 1, that is, the maximal subfields of D form a single conjugacy class.
Each element of D is contained in a maximal field. Using (1), we see that D \ F is the
disjoint union of the sets E \ F as E varies through the maximal fields in D. The number
of distinct fields xEi x−1 as x varies through D× is [D× : N (E× i )] (for i = 1, . . . , r).
qN − 1
Thus, by Lemma 1, the field Ei has m conjugates, and so the conjugates of Ei \ F
q i −1
q N − 1 mi
account for m (q − q) elements in D \ F (for i = 1, . . . , r). Therefore
q i −1
q N − 1 m1 q N − 1 mr
qN − q = (q − q) + · · · + (q − q).
q m1 − 1 q mr − 1
60 ALAN ROCHE

Rearranging, we obtain
qN − q q m1 − q q mr − q
= + · · · + . (2)
qN − 1 q m1 − 1 q mr − 1
1
The left side is less than 1. Further, each term on the right side is greater than .
2
Indeed, for m ≥ 2,
qm − q 1
m
> ⇐⇒ 2(q m − q) > q m − 1
q −1 2
⇐⇒ q m − 2q + 1 > 0,

and q m − 2q + 1 ≥ q 2 − 2q + 1 = (q − 1)2 > 0. Hence

1 1
1 > + ··· + (r terms)
2 2
r
= ,
2
so 2 > r and r = 1.
From here, there are two ways to complete the argument.
Method 1. Since r = 1, the equality (2) now says that there is a positive integer m < N
such that
qN − q qm − q
= . (3)
qN − 1 qm − 1
Clearing denominators, we have

(q m − 1)(q N − q) = (q m − q)(q N − 1).

Expanding each side, cancelling common terms and rearranging then gives

q N +1 − q N = q m+1 − q m .

Thus q N (q − 1) = q m (q − 1) and N = m—a contradiction.

Alternatively, the real function
qx − q q−1
= 1− x
qx − 1 q −1
is increasing (since q x is increasing). Again, we see from (3) that N = m which is
absurd.
Method 2. We’ve proved that if E is a maximal subfield of D then
[
D× = x E× x−1 .
x∈D×

This is impossible: a finite group is never a union of conjugates of a proper subgroup.

Lemma 2. Let H be a proper subgroup of a finite group G. Then

[
xHx−1 6= G.
x∈G

Proof. The subgroup H has at most [G : H] distinct conjugates in G. Indeed, for x ∈ G,

the conjugate xHx−1 depends only on the left coset xH of H in G. Moreover, each
 Wedderburn’s Little Theorem 61

conjugate of H contains the identity element. Hence

[ |G|
| xHx−1 | ≤ (|H| − 1) + 1
|H|
x∈G
|G|
= |G| − +1
|H|
< |G|,
which gives the result. 
Remark 3. Lemma 2 features in several proofs of the Little Theorem. As noted
in the introduction, it provides the final step in the classic proof recounted by van
der Waerden. Further, the group-theoretic principle that underlies Zassenhaus’s proof
(recalled in Remark 1) rests ultimately on Lemma 2. The lemma also plays a role in
our third and fourth proofs.

4. Second Proof: Cyclic Sylow Subgroups

Next we rework an argument from 1964 due to T. J. Kaczynski [11] 2. The key
observation is the following which we derive from Lemma 1 (the first several paragraphs
of [11] give another route).
Proposition 1. Every Sylow subgroup of D× is cyclic.
Proof. We use induction on |D|. The base case |D| = 2 is trivial. Assuming that each
Sylow subgroup of the multiplicative group of a division ring of order less than |D| is
cyclic, we wish to show that D has the same property.
Let S be a Sylow l-subgroup of D× for some prime l. If S ⊂ F× then S is cyclic
(by cyclicity of F× ). If S is not contained in F× , then S F× /F× is a nontrivial Sylow
l-subgroup of D× /F× . In this case, we choose a nontrivial element βF× in the center
of S F× /F× . Then, for each α ∈ S, we have βαβ −1 = λα for some λ ∈ F× . Thus
conjugation by β takes the field F(α) to itself. Using Lemma 1, it follows that α and β
commute. Hence S ⊂ C(β), the centralizer of β in D. Observe that C(β) 6= D as β ∈ / F.
Using our inductive hypothesis, we conclude that S is cyclic. 

For use in this section and the next, we list some properties of finite groups in which
all Sylow subgroups are cyclic. More can be said: in fact, a finite group has cyclic
Sylow subgroups if and only if it’s a semidirect product of cyclic groups whose orders
are relatively prime ([8, Theorem 5.16]).
Proposition 2. Let G be a finite group in which every Sylow subgroup is cyclic and
write l for the largest prime divisor of |G|. Then:
(a) a Sylow l-subgroup of G is normal;
(b) the group G is solvable;
(c) if G is nonabelian, it contains a normal abelian subgroup that is strictly larger
than the centre.
The proposition is a consequence of a famous result of Burnside ([8, Theorem 5.13]).
Burnside’s Normal Complement Theorem. Suppose a finite group G admits a
Sylow subgroup S such that N (S) = C(S). Then S has a normal complement in G.
That is, there is a normal subgroup N of G such that G = N S and N ∩ S = {1}.

2The author of the argument is better known today for activities outside mathematics.
62 ALAN ROCHE

The result is also known as Burnside’s Transfer Theorem as it follows from properties
of the transfer map—a natural homomorphism from a group G to the commutator group
H/H ′ of a subgroup H of finite index . Isaacs’ book [8] contains a cogent account of
this map and several of its applications including Burnside’s result.
Proof of Proposition 2. If G has prime-power order then (a) is obvious and (b) and
(c) are well known. For the remainder of the proof, we assume that the order of G is
divisible by at least two primes.
To establish part (a), we argue by induction on |G|. Write l1 for the least prime divisor
of G and let S1 be a Sylow l1 -subgroup of G. First, we note that Burnside’s Theorem
implies that S1 has a normal complement in G. To this end, observe that S1 ⊂ C(S1 ),
so that [N (S1 ) : C(S1 )] is not divisible by l1 . The action of N (S1 ) on S1 by conjugation
induces an embedding from N (S1 )/C(S1 ) into Aut(S1 ). We have |S1 | = l1e for some
positive integer e. Since S1 is cyclic, Aut(S1 ) has order φ(l1e ) = l1e−1 (l1 − 1). Thus
[N (S1 ) : C(S1 )] divides l1 − 1, and so N (S1 ) = C(S1 ) (as l1 is the least prime divisor
of |G|). Hence, by Burnside’s Theorem, G = N S1 for a normal subgroup N of G such
that N ∩ S1 = {1}.
Every Sylow subgroup of N is again cyclic. Our inductive hypothesis therefore implies
that a Sylow l-subgroup S of N is normal in N , and thus is the unique Sylow l-subgroup
of N . Now S is also a Sylow l-subgroup of G. Further, for g ∈ G, the group gSg −1 is
a Sylow l-subgroup of N (by normality of N ). Using uniqueness of S, we see that S is
normal in G.
For part (b), we also use induction on |G|. With notation as in the proof of part
(a), the quotient G/S inherits from G the property that each of its Sylow subgroups is
cyclic. Thus, by our inductive hypothesis, G/S is solvable. As S is certainly solvable,
it follows that G is solvable.
For part (c), suppose G is nonabelian and write Z for the centre of G. Then the
nontrivial group G/Z has cyclic Sylow subgroups. Using part (a), we see that G/Z
admits a nontrivial cyclic normal subgroup, say H/Z. Note that H is abelian (since
H/Z is cyclic), normal in G and strictly contains Z. We’ve proved part (c). 
We now use Proposition 1 and Proposition 2 (c) to prove Wedderburn’s theorem.
Proof. We assume that D is noncommutative and derive a contradiction.
By Proposition 1 and Proposition 2 (c), the group D× contains a normal abelian
subgroup S that is strictly larger than F× . We set E = F(S), the subfield of D generated
over F by S. Since S is normal in D× , we see that E× is also normal in D× .
As in the first proof, we write down two ways to complete the argument (see the last
two paragraphs of [11] for yet another way).
Method 1. Since N (E× ) = D× , Lemma 1 says that C(E× ) = D× , and so E is contained
in the centre F of D—a contradiction.
Method 2. Recall that |F| = q and |D| = q N . As E strictly contains F, we have
C(E) 6= D. Thus
|E| = q m and |C(E)| = q n
for integers m and n with m ≤ n < N . Note that n divides N : in fact, N = ne where
e is the dimension of D as a C(E)-vector space.
The action of D× on E by conjugation induces an embedding of groups
D× /C(E)× ֒→ Aut(E/F).
In particular, [D× : C(E× )] is at most |Aut(E/F)| = m, that is,
q ne − 1
≤ m.
qn − 1
 Wedderburn’s Little Theorem 63

The left side is q n(e−1) + · · · + q n + 1. Certainly n < q n , and thus a fortiori

n < q n(e−1) + · · · + q n + 1 ≤ m,
in contradiction to m ≤ n. 

5. Third Proof: Carter Subgroups

In 1961, R. W. Carter published a striking result [4].
Carter’s Theorem. Let G be a finite solvable group. Then G contains a nilpotent
subgroup C such that N (C) = C. The subgroup C is unique up to conjugacy in G.
There is an analogous statement in the theory of Lie algebras which prompted
Carter’s discovery. A Lie subalgebra of a Lie algebra that is nilpotent and coincides
with its normalizer is called a Cartan subalgebra. Under suitable hypotheses, Cartan
subalgebras exist and are unique up to a natural notion of conjugacy. 3
Carter’s Theorem, once formulated, is not difficult to prove. Indeed, it’s an exercise
without hints in Isaacs’ text [8] (see p. 91) though the author does acknowledge that
the problem may be “a bit harder than most of the problems in this book.”
A nilpotent subgroup C of a group G such that N (C) = C is now called a Carter
subgroup of G. Non-solvable finite groups need not have Carter subgroups: for example,
A5 , the smallest non-solvable group, has none. However, a Carter subgroup of an
arbitrary finite group—when it exists—is unique up to conjugacy. This was put forward
as a conjecture in 1976 and stood for around thirty years (see [15] for some of the
history). It was finally proved by E. P. Vdovin as a culmination of a series of reductions
and calculations based around the classification of finite simple groups—in particular,
detailed properties of finite groups of Lie type [19].
Using Lemma 1 and the discussion in Section 4, we can quickly deduce the Little
Theorem from Carter’s result.
Proof. Let D be a noncommutative finite divsion ring. By Proposition 1 and Proposi-
tion 2 (b), the group D× is solvable. Let E be a maximal field in D, so that C(E× ) = E× .
Then N (E× ) = E× by Lemma 1, and so E× is a Carter subgroup of D× . Carter’s theo-
rem therefore says that the maximal subfields of D form a single conjugacy class (under
the action of D× ). Since each element of D lies in a maximal field, it follows that D× is
a union of conjugates of the proper subgroup E× . Lemma 2, however, tells us that this
is impossible. Thus a finite division ring is a field. 
Remark 4. We only used solvability of D× to ensure that the Carter subgroups of D×
are conjugate. Can we instead appeal to conjugacy of Carter subgroups in a general
finite group [19]? If so, this would yield an outrageous proof of the Little Theorem
given that Vdovin’s paper and its antecedents rest on the classification of finite simple
groups. This approach, however, is not only wildly inefficient—it’s circular: Wedder-
burn’s theorem is a step, a tiny one, in the classification results that Vdovin uses.

6. Fourth Proof: Frobenius Groups

The final proof uses a uniqueness property of Frobenius groups. It relies on a pow-
erful result of J. G. Thompson—we are indeed wielding a sledgehammer. For a more
elementary approach to the Little Theorem via properties of Frobenius groups, see [6, 3].
Definition. A finite group G is a Frobenius group if it admits a nontrivial proper
subgroup H such that
gHg −1 ∩ H = {1} for all g ∈ G \ H. (4)
3Thanks to T. J. Laffey for confirming the influence on Carter of the theory of Cartan subalgebras.
64 ALAN ROCHE

Following [14], we say in this case that (G, H) is a Frobenius pair.

To draw out the definition, let’s rephrase it in terms of the action of G by left
multiplication on the space G/H of left cosets of H in G. The defining property (4)
says exactly that each element of G \ {1} fixes at most one point in G/H. Moreover,
since H is nontrivial, some non-identity element of G fixes some point in G/H—each
element of H \ {1}, for example, fixes the coset H. Conversely, suppose a finite group
G acts transitively on a set X with |X| > 1 so that these properties hold, that is,
(a) no element of G \ {1} fixes more than one element of X,
(b) some element of G \ {1} fixes some element of X.
Let x ∈ X and write H for the stabilizer in G of x. Then H satisfies (4) by (a), is
nontrivial by (b), and is proper since |X| > 1. In all, (G, H) is a Frobenius pair.
In other words, a Frobenius group is a transitive permutation group on a finite set
such that each nonidentity element fixes at most one point and some point is fixed by
some nonidentity element (so that the action is not just the regular action).
There is a fascinating structure theory of Frobenius groups. For compelling accounts
of part of this theory, see [8, Chap. 6] and [14, Chap. 6]. The first main result is due
to Frobenius in 1901. It says that in any Frobenius pair (G, H) the elements of G that
act without fixed points on G/H plus the identity element form a normal complement
N to H in G, that is, G = H ⋉ N . The proof was an application and early triumph of
the emerging theory of characters of finite groups, itself initiated by Frobenius in 1896.
The group N is now called a Frobenius kernel.
Special properties of Frobenius groups gave rise to a conjecture that Frobenius kernels
are always nilpotent. Thompson proved this conjecture in his celebrated 1959 doctoral
thesis. The proof drew as much attention as the result: it hinged on a new, powerful
criterion for an odd Sylow subgroup to admit a normal complement. Isaacs’ book
[8] includes a thorough discussion of Thompson’s criterion (a slight modification of a
simpler but still involved version that Thompson arrived at after his thesis) and its
connection with Frobenius kernels (see Theorem 6.24 and Chapter 7). Another source,
a more condensed one, is Feit’s classic book [7] (see 22.2 and 25.10).
Nilpotence of Frobenius kernels leads to the following uniqueness property of Frobe-
nius groups—the key to our final proof of the Little Theorem.
Uniqueness of Frobenius Structures. Let (G, H1 ) and (G, H2 ) be Frobenius pairs
in the same group G. Then H1 and H2 are conjugate in G.
This is Exercise 6.6.6 in Serre’s book [14] (see p. 86). Despite its label, the exercise
is not in the least diabolical. In fact, Serre gives a generous hint that’s effectively a
solution.
We’ve sketched (more than) enough background to write down the fourth proof.
Proof. Let D be a noncommutative division ring of minimal order and let E be a maximal
subfield of D, so that C(E× ) = E× . Then, by Lemma 1, xE× x−1 6= E× for x ∈ D× \ E× .
Using minimality of |D|, we noted at the start of Section 3 that a pair of maximal fields
in D must intersect in the centre F (see equation (1) and the preceding discussion). In
particular,
xE× x−1 ∩ E× = F× , for all x ∈ D× \ E× .
Working in the quotient group D× /F× , it follows that
xF× · E× /F× · (xF× )−1 ∩ E× /F× = (xE× x−1 ∩ E× )/F×

= {1}, for all x ∈ D× \ E× .

That is, (D× /F× , E× /F× ) is a Frobenius pair. Using uniqueness of Frobenius structures,
we see that the maximal fields in D form a single conjugacy class.
 Wedderburn’s Little Theorem 65

To finish, we argue as in Method 1 or Method 2 of our first proof. 

7. Four Proofs of Zassenhaus’s Lemma

It remains to prove Zassenhaus’s lemma. As with the Little Theorem, we record four
overlapping proofs. All share the same first step which is borrowed from [20]. It makes
crucial use of surjectivity of the norm map for finite fields. From the theory of cyclic
algebras, the Little Theorem is equivalent to this surjectivity property, so it’s not a
surprise that it lies at the base of the arguments below.
As noted earlier, this section owes much to the suggestions of a reviewer of a pre-
Bulletin version of the paper. In particular, the fourth proof is a variant of an argument
provided by this reviewer. The third proof is close to Zassenhaus’s original argument.
For convenience, we recall the statement of the lemma.
Lemma 1. If E is a subfield of a finite division ring D, then the normalizer and cen-
tralizer of E× in D× coincide.
Proof. Suppose N (E× ) 6= C(E× ) for some subfield E of the finite division ring D. Choose
β ∈ N (E× ) with β ∈/ C(E× ).
Let l be the least positive integer such that β l ∈ C(E× ) and set K = hE, β l i, the
subring of D generated by E and β l . Note that K is a field since β l ∈ C(E× ). We write
σ for the automorphism of K given by conjugation by β, that is,
σ(λ) = βλβ −1 , λ ∈ K.
Observe that σ has order l.
Let D1 = hK, βi, the subring of D generated by K and β. For later use, note that
β i λ = σ i (λ)β i , i ∈ Z, λ ∈ K. (5)
We put K1 = Kσ , the fixed field of σ. Thus the automorphism group of the extension
K/K1 is generated by σ and [K : K1 ] = l. Let N denote the norm map from K to K1 ,
so that
N (α) = ασ(α)σ 2 (α) · · · σ l−1 (α), α ∈ K.
Now, for α ∈ K,
(αβ)l = αβ · αβ · · · αβ (l terms)
= α · βαβ −1 · β 2 αβ −2 · · · β l−1 αβ −(l−1) · β l
= ασ(α)σ 2 (α) · · · σ l−1 (α)β l
= N (α)β l .
The element β l ∈ K is visibly fixed under conjugation by β, that is, β l ∈ K1 . Since
norm maps are surjective for finite fields, we can choose α ∈ K such that N (α) = β −l ,
equivalently (αβ)l = 1. Thus we can and do adjust the element β so that β l = 1. We
still have D1 = hK, βi and the automorphism σ of K is still given by conjugation by β.
We rewrite β l = 1 as (1 − β)(1 + β + · · · + β l−1 ) = 0. Since β 6= 1, we have
1 + β + · · · + β l−1 = 0, (6)
and so
λ + βλ + · · · + β l−1 λ = 0, for all λ ∈ K.
Equivalently, via (5),
λ + σ(λ)β + · · · + σ l−1 (λ)β l−1 = 0, for all λ ∈ K. (7)
From here, we write down four ways to complete the proof.
66 ALAN ROCHE

Method 1. Fix ζ ∈ K such that K = K1 (ζ). Substituting ζ 0 = 1, ζ, . . . , ζ l−1 in (7) gives

ζ i + σ(ζ)i β + · · · + σ l−1 (ζ)i β l−1 = 0, for i = 0, 1, . . . , l − 1. (8)
We can rewrite these l equations as a single matrix equation
    
1 1 ··· 1 1 0
 ζ σ(ζ) ··· σ l−1 (ζ) 
 β 
   0 
 .. .. .. .  = .. . (9)
  
..   .. 
  
 . . . .  . 
ζ l−1 σ(ζ)l−1 · · · σ l−1 (ζ)l−1 β l−1 0
Thus the square matrix, say M , is singular. On the other hand, it’s of Vandermonde
form, so Y
σ j (ζ) − σ i (ζ)


det M =
0≤i<j<l
is nonzero. This contradiction proves the lemma.
Method 2. We derive a contradiction from (7) by a slightly different matrix argument.
Write tr : K → K1 for the trace map of the extension K/K1 , so that
tr(λ) = λ + σ(λ) + · · · + σ l−1 (λ), λ ∈ K.
Since K/K1 is separable, the symmetric bilinear form
(λ, µ) 7→ tr(λµ) : K × K → K1
 
is nondegenerate. Fixing a K1 -basis λ1 , . . . , λl of K, the matrix tr(λi λj ) is therefore
invertible.
Substituting the basis elements λ1 , . . . , λl in (7), we have
λi + σ(λi )β + · · · + σ l−1 (λi )β l−1 = 0, for i = 1, . . . , l. (10)
As in Method 1, we gather these l equations into a single matrix equation
λ1 σ(λ1 ) · · · σ l−1 (λ1 )
    
1 0
λ2 σ(λ2 ) · · · σ l−1 (λ2 )  β  0
 .. .. ..   ..  =  ..  . (11)
    
..
. . . .   .   . 
λl σ(λl ) · · · σ l−1 (λl ) β l−1 0
Writing M for the given square matrix and M ⊤ for its transpose, the product M M ⊤
has ij entry
λi λj + σ(λi λj ) + · · · + σ l−1 (λi λj ) = tr(λi λj ).
Thus M M ⊤ is invertible, so M is invertible, in contradiction to (11).
Method 3. Using (5) and that β l ∈ K, we see that
l−1
X
D1 = hK, βi = Kβ i . (12)
i=0
We claim that the sum is direct. Note that our claim means that (6) cannot hold and
thus establishes the lemma.
The claim follows from a powerful technique that goes back to Dedekind, often called
“linear independence of characters.” In detail, suppose we have a dependence relation
λ1 β i 1 + · · · + λr β i r = 0 (13)
for λ1 , . . . , λr ∈ K and 0 ≤ i1 < · · · < ir < l. In search of a contradiction, we assume
that r is minimal, so that each λi 6= 0.
Multiplying (13) on the right by γ ∈ K and rewriting via (5), we have
λ1 σ i1 (γ)β i1 + · · · + λr σ ir (γ)β ir = 0, γ ∈ K. (14)
 Wedderburn’s Little Theorem 67

Choose γ1 ∈ K such that σ i1 (γ1 ) 6= σ i2 (γ1 ). Replacing γ by γ1 γ in (14) gives

λ1 σ i1 (γ1 )σ i1 (γ)β i1 + λ2 σ i2 (γ1 )σ i2 (γ)β i2 + · · · + λr σ ir (γ1 )σ ir (γ)β ir = 0.
On the other hand, multiplying (14) on the left by σ i1 (γ1 ) gives
λ1 σ i1 (γ1 )σ i1 (γ)β i1 + λ2 σ i1 (γ1 )σ i2 (γ)β i2 + · · · + λr σ i1 (γ1 )σ ir (γ)β ir = 0.
Subtracting the last equation from the previous one, we see that
λ2 σ i2 (γ1 ) − σ i1 (γ1 ) σ i2 (γ)β i2 + · · · + λr σ ir (γ1 ) − σ i1 (γ1 ) σ ir (γ)β ir = 0.

By our choice of γ1 , the coefficient of β i2 is nonzero (for γ nonzero) which contradicts

minimality of r in (13). Therefore (12) is a direct sum and we’ve proved the lemma
once more.
Method 4. Consider K as a K1 -vector space. The field K embeds in End K1 (K) via
m : K → End K1 (K) where m(λ)(µ) = λµ (for λ, µ ∈ K). We identify K with its image
under m. That is, for λ ∈ K, we simply write λ for the map m(λ) ∈ End K1 (K). We
also have σ ∈ End K1 (K). In parallel to (5), these various elements of End K1 (K) satisfy
σ i λ = σ i (λ)σ i , i ∈ Z, λ ∈ K, (15)
Now, by Dedekind’s lemma (linear independence of characters), the sum
l−1
X
Kσ i ⊆ End K1 (K)
i=0

is direct, and hence has dimension l2 over K1 . The containment is therefore an equality.
That is, each element of End K1 (K) admits a unique expression as l−1 i
P
i=0 λi σ with each
λi ∈ K. Using the same notation, we see that there is a well-defined map of K-vector
spaces
l−1
X Xl−1
i
λi σ 7−→ λi β i : End K1 (K) −→ D1 . (16)
i=0 i=0
We claim that the map is a ring homomorphism. Note that the lemma follows. In-
deed, given the claim, the kernel of (16) is a two-sided ideal in the simple ring End K1 (K)
and hence is trivial. This means that (16) is injective and so the division ring D1 con-
tains a copy of the matrix ring End K1 (K)—a contradiction. Alternatively, the map is
surjective (its image, for example, contains K and β). Using simplicity of End K1 (K)
again, the rings End K1 (K) and D1 must be isomorphic which is absurd.
To establish the claim, we appeal to the following principle whose proof is immediate.
(α) Let R and S be rings and let φ : R → S be a homomorphism of abelian groups.
Suppose R is generated as an abelian group by a subset Γ and that
φ(γ1 γ2 ) = φ(γ1 )φ(γ2 )
for all γ1 , γ2 ∈ Γ. Then φ is a homomorphism of rings.
From our discussion, End K1 (K) is generated as an abelian group by the elements λσ i
for λ ∈ K and i ∈ Z. Writing φ for the map in (16) and using (α), we see that we only
have to check one family of relations:
φ(λσ i µσ j ) = φ(λσ i )φ(µσ j ), λ, µ ∈ K, i, j ∈ Z.
By (15), λσ i µσ j = λσ i (µ)σ i+j , and so the left side is λσ i (µ)β i+j . By (5), the right side
is
λβ i µβ j = λσ i (µ)β i+j .
Thus (16) is a ring homomorphism and the (fourth) proof is complete. 
68 ALAN ROCHE

Remark 5. Dedekind’s lemma (linear independence of characters) is at the heart of

Zassenhaus’s lemma. Its central role in Methods 3 and 4 is evident. For Methods 1
and 2, the key is that the square matrices in equations (9) and (11) are invertible—and
invertibility of these matrices is a quick consequence of Dedekind’s lemma (a preferable
approach perhaps to the more computational one used above). To check the implication,
write Ci for the ith column of either matrix and suppose there exist γ1 , . . . , γl ∈ K such
that
γ1 C1 + · · · + γl Cl = 0,
the zero column vector. In each case, the relation says that l−1 i
P
Pl−1 i=0 γi σ ∈ End K1 (K)
vanishes on a K1 -basis of K. Thus i=0 γi σ i = 0, so that γi = 0 for all i, and the
matrices are indeed invertible via Dedekind’s lemma. Noether’s motto, quoted in [16],
was “it is already all in Dedekind.”

References
[1] M. Aigner and G. M. Ziegler, Proofs from THE BOOK, 6th ed. Springer, 2018.
[2] E. Artin, The influence of J. H. M. Wedderburn on the development of modern algebra. Bull. Amer.
Math. Soc. 56 (1950), 65-72.
[3] R. P. Burn and D. M. Madaram, Frobenius groups and Wedderburn’s theorem. Amer. Math.
Monthly. 77(9) (1970), 983-984.
[4] R. W. Carter, Nilpotent self-normalizing subgroups of soluble groups. Math. Zeit. 75 (1961), 136-
139.
[5] R. K. Dennis and B. Farb, Noncommutative Algebra. Graduate Texts in Math., 144, Springer-
Verlag, 1993.
[6] S. Ebey and K. Sitaram, Frobenius groups and Wedderburn’s theorem. Amer. Math. Monthly. 76(5)
(1969), 526-528.
[7] W. Feit, Characters of Finite Groups. W. A. Benjamin, 1967.
[8] I. M. Isaacs, Finite Group Theory. Graduate Studies in Math., 92, Amer. Math. Soc., 2008.
[9] I. N. Herstein, Topics in Algebra, 2nd ed. John Wiley & Sons, 1975.
[10] N. Jacobson, Basic Algebra II, 2nd ed. W. H. Freeman and Co., 1989.
[11] T. J. Kaczynski, Another proof of Wedderburn’s theorem. Amer. Math. Monthly. 71(6) (1964),
652-653.
[12] S. Mac Lane, Mathematics at Göttingen under the Nazis. Notices Amer. Math. Soc. 42(10) (1995),
1134-1138.
[13] C. Reid, Hilbert. Reprint of 1970 original. Copernicus, 1996.
[14] J.-P. Serre, Finite Groups: An Introduction. International Press, 2016.
[15] M. Chiara Tamburini, The conjugacy conjecture for Carter subgroups. Milan J. Math. 75 (2007),
357-377.
[16] B. L. van der Waerden, On the sources of my book Moderne Algebra. Hist. Math. 2 (1975), 31-40.
[17] J. H. M. Wedderburn, A theorem on finite algebras. Trans. Amer. Math. Soc. 6 (1905), no. 3,
349-352.
[18] J. H. M. Wedderburn, On hypercomplex numbers. Proc. London Math. Soc. (2) 6 (1908), 77-118.
[19] E. P. Vdovin, Carter subgroups of finite groups. Siberian Adv. Math. 19 (2009), no. 1, 27-74.
[20] H. J. Zassenhaus, A group-theoretic proof of a theorem of Maclagan-Wedderburn. Proc. Glascow
Math. Assoc. 1 (1952), 53-63.

Alan Roche studied at University College Dublin and the University of Chicago, and has
worked at the University of Oklahoma since 2001. His principal mathematical interests are in
number theory and representation theory.
Department of Mathematics, University of Oklahoma, Norman, OK 73019, USA.
E-mail address: [email protected]