An Extension of Galois Theory to Non-Normal and Non-Separable Fields

Author(s): N. Jacobson
Source: American Journal of Mathematics, Vol. 66, No. 1 (Jan., 1944), pp. 1-29
Published by: The Johns Hopkins University Press
Stable URL: http://www.jstor.org/stable/2371892 .
Accessed: 07/12/2014 21:32

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

The Johns Hopkins University Press is collaborating with JSTOR to digitize, preserve and extend access to
American Journal of Mathematics.

http://www.jstor.org

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
AN EXTENSION OF GALOIS THEORY TO NON-NORMAL AND
NON-SEPARABLE FIELDS.*1

By N. JACOBSON.

As is well known, modern Galois theory is a theory of the automorphisms

of an arbitrary field P. Its principal result is the establishment of a (1 - 1)
correspondence between the finite groups g of automorphisms in P and the
subfields bg over which P is finite, separable anidnormal, such that 1) gi 2 g2
if and only if Slg?1< g92 and 2) 92 is invariant in gi if and only if tg2
is normal over Jgl. A second type of Galois theory has been given by the
author in a previous paper.2 It associates the subfields JDof a field P of char-
acteristic p over which P has the form 1 (xi, , x,,,) where xi = i in 1D
with certain algebras of derivations in P. Ill seeking an extension of these
theories we have been led to a study of the self-repi-esenttations of P, i. e. the
representations of P by matrices with elements in this field.3 One is led to
seek an extension along these lines by the following two remarks. First, if S
is an automorphism, thei a--> a8 is a self-representation by 1-dimnensional
nmatricesand second, if D is a derivation, then

/a aD\

is a representation bv 2-rowed matrices.

Now by our definition of a self-representation any isomorphism of P into
a subfield is a self-representation. It is therefore desirable to restrict the

* Received July 12, 1943.

Presented to the Society, September 11, 1943.
2
"Abstract derivation and Lie algebras," Transactionts of the Amer-ican Mathe-
matical Society, vol. 43 (1937), p. 220.
3 A Galois theory of separable extensions which makes use of the coiicept of self-
r epresentation has been aninounced recently by L. Kaloujnine, " Sur la the'orie de
Galois des corps non galoisielis separables," Compies Revdus de l'Acad6mie des Sciences,
vol. 214 (1942), pp. 597-599. (Abstracted in Mathematical Reviews, vol. 4 (1943),
p. 130). It appears that Kalouijniiie's results are special cases of those obtailled
here. See 12. It should be mentioned also that Galois theory for separable extensions
based on the classical theory for separable normal extensions had been developed pre-
viously. See Krasner, M. " Sul la. theorie de la ramificationl des ideaux des corps non-
galoisiens de nombres aluebriques," These, Paris, 1938. The present theory (and
apparently that of Kaloujnine) is independent of the classical theory.
I

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
2 N. JACOBSON.

class of self-representationlsin such a way that any self-representation in this

class having rank one (one-rowed) is an automorphism of P. This has been
done by defining non-singular self-representations. There is a natural way
to define the product of two self-representations: One substitutes for the
elements of the matrices of one representation the matrices representing these
elements in the second representation. This, of course, is a generalization
of the product of isomorphisnms.
With any self-representation-or more exactly, with any class of similar
self-representations-we may associate a double P-module 9i having 4the
property that its right dimensionality (R: P,) = m < oo. If, in addition,
(#: Pi) = m, then M is said to be non-singular. Modules of this type corre-
spond to non-singular self-representations. One may also define a product
for double modules corresponding to the product of self-representations.
Now with each self-representation we may associate a composite (ring)
of the field P with itself. This consists of the ring of transformations in P
generated by the scalar multiplications x -- ax and x -?> xa. These composites
may also be defined abstractly: A system (K, S, T) is a composite of P with
itself if K is a ring and S aind T are isomorphisms between P and subfields
PS and pT of K such that 1) K - pSpT; 2) K is commutative, 3) iS = T,
4) (K: PT) < oo. If (K', S', T') is a second comnpositeand the mapping
:Ea 8T >/S/3T' is a homomorphism, then (K, S, T) is a cover of (K', S', T')
((K, S, T) ? (Ki', S'',T')). If this mapping is an isomorphism, the two
composites are equivalent. Throughout our discussion equivalent composites
are idenitified. One may define non-singular composites and the product
(K, S, T) X (L, UT,V) of ally two composites. A composite (K, S, T) is
called a Galois composite if (K, S, T) ? (K', S, T) X (K, S, T), (K, S, T)
> the identity composite (Pu, U, U) and (K, S, T) = (K, T, S). Our main
result is the establishment of a (1 - 1) correspondence between the Galois
composites r of P and the subfields 4r of P over which P is finite, such that
rX? P2 if and only if (Dr, ?< (2r
Any composite may be decomposed in a certain sense into indecomposable
composites (K,, Si, T,). Conditions that (K, S, T) be a Galois composite
may be given in terms of the components (K,, Si, Ti). If each K, is a field,
(K, S, T) is called semi-simple. In this case the condition that (K, S, T) be
a Galois composite is that the components (Ki, Si, Tj) form a hypergroup
relative to the product (K.1,S,, T,) (Kj, Sj, Tj) defined to be the set of com-
ponents of the composite (Ki, Si, T,) X (Kj, Sj, Tj). Our fundamental
correspondence between composites and subfields induces a (1 - 1) corre-
spondence between semi-simple Galois composites and subfields (P over which

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAL FIELDS. 3

I is finite and separable. We investigate also the fields 1 over which P is

normal. Combining our results for normal and separable fields we obtain a
(1 - 1) correspondence between the semi-simple composites whose hyper-
groups are groups and the subfields 1' over which P is finite, separable and
normal. Finally, it is easy to obtain the connection between these composites
and finite groups of automorphisms and this gives the classical theorem.
Our results could have been formulated in terms of self-representations
but this seems to be unnatural except in the case where P has a primitive
element over (d. In a later paper we hope to ilnvestigate in greater detail the
theory for subfields 1 over which P is purely inseparable.

1. Self-representations of fields. If P is an arbitrary field, we define a

self-representation of P as an isomorphism between P and a field of rn-rowed
natrices with elements in P such that 1 -> 1 the identity matrix. The integer
ri is the rank of the representation. If E : a - aCEin P. is a self-representa-
tion, we define E,j as the transformation sending a into the element in the
i-th row and j-th column of aE.4 Thus aE- (aEij) and the Eij satisfy the
following conditions:

(1) (a, +-- )Eij - caE,j+ pE1,j (a/q)Eij =

>x(cx) (/3E)1),
lEij -= Sij

Conversely any m2 transformations E,j in P that satisfy (1) determine a self-

representation a -- (caEjj) of P.
We shall call y a fixed element under E if yE iS the diagonal matrix
{y,~* , y}. Hence y is fixed if yEi y and yEij 0O for i 7 j. Evidently
=

the set of fixed elements is a subfield of P.

We shall recall now some of the important concepts from general repre-
sentation theory. Two self-representations E and F are similar if there exists
a fixed non-singular matrix S =- (orj) such that cF - S-laES for all a. Similar
representations evidently have the same fixed elements. A representation is
reducible if it is similar to a representation F in which Fij 0 for i > r > 0 =

and j ? r. In this case the correspondences a -> (ac#F7) kI, I = 1,. . , r and
a > (aFcpq), p?,q = r + 1, * , m are self-representations. The first of these
is a subrepresentation and the second a difference representation of E. If
besides F,j =0 for i> r and j?r we have FXj - 0 for i<r and j> r,
then E is decomposable into the components . * (aFjkc), 1X * (aFpq). In a
similar manner decomposability inlto more than two components may be de-

4
Pm denotes the ring of mnX mnmatrices with elements in P.

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
4 N. JACOBSON.

" E1 +
fined. We shall write =A + Es when E is decomposableinto
E, +
the components E'. If E = E11 + + E where the E, are irreducible,
E is said to be completely reducible. If an element y of P is a fixed element
relative to E, it is fixed under any subrepresentation ancdunder any difference
representation of E.

2. Double P-modules. We shall now give a module formulatioln of the

representation problem. We define a double P-module I as a commutative
group SRwhich is both a right P-module and a left P-module such that

1. lx= x- x.
2. = (x[3).
~~~~~~~(ax)B-
Thus if we denote the endomnorphisnm x --> ax by a, anld the endomorphism
x -- xxa by a,. then ca,i,r = /3ra for all a, 8 in P. We denote the field of
eiidomorphisms a (o) by P1(Pr). For our purposes it suffices to consider
only double P-modules that satisfy the following condition

3. (91: Pr)= m < o.

Hence from now on we use the term " double P-module " for "double P-module
satisfying condition 3."
Let xl, * *, x. be a basis for N over Pr. Then ax? = xjajj ancl it is
readily verified that the correspondence a -> (aij) is a self-representation E
of P. The rank of E is m. Conversely if E is any self-representation; then
we let R be a right P-module such that (%R:Pr) m. If x,, , xi, is a
=

basis for NP,we write x = xe and define ax == Xpqj where qj (aEii) i. =

Then it is readily verified that X is a double P-module and ax = xj (aEjp).

Thus any self-representation is obtained from some double P-module in the
manner indicated. If yi, , ym, yi :4xjuj, is a second basis over Pr of the
=

double P-module X-, then ay- yj (aFj) and (aFij) -F S-aES where
S = (oij ). hIence the differenitright bases for X correspond to the different
self-representations similar to E.
It is clear that an element y of P is a fixed element under E if and only
if yt yr.
If (E is a submodule of R, ay aindyaEe ( for any y in ( and any a in P.

5 Cf. E. Noether, " Hyperkomplexe Grossen und Darstellungstheorie," Mctathe-

matische Zeitschrift, vol. 30 (1929), p. 669; v. d. Waerden, Moderne Algebra, vol. 2,
p. 131; or Jacobson, The Theor-y of Rings, New York, 1943, p. 95 (referred to here-
after as R).

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAL FIELDS. 5

We may choose a basis y,* , ym of 1 over Pr such that yi, . . ., yr is a basis

for (E over Pr. Then ay7,= :y itp, i, 1 1, , r. Hence E is reducible
and the representation F determined by 5 is a subrepresentation of E. Also
we have ay_ Egqp (mnod25) for p, q = r + 1, *, m. Hence the repre-
sentation G determined by the difference P-module X - (a is a difference
representation of Fi. In a similar manner we see that the condition that E
be decomposableis that W1be a direct sum of submodules X 0 and the condition
that F be completely reducible is that X1 be a completely reducible double
miodule.

3. Composites. If X is a double P-module, the set of transformations

that are finite sums of products ac/3ris a ring PPPr PrP1. The elements of
PiPr belong to 2 the algebra of linear transformations of 91 over Pr. Now we
recall that the scalar multiplications in the algebra 2 are the mappings
A >Aa, ar.A and that the dimensionality of 2 over Pr is M2. Since
PIPr 2 Pr, PiPr is a subalgebra of Q. Hence (PiPr: Pr) = q ? M2.
We nlow define a composite of a field with itself as a system (K, S, T)
consisting of a ring K and two isomorphisms S and T between P and subfields
pS and pT of K such that the following conditions hold:

1. K_ pSpT
2. =S#T ,8TaS for all a,,G
1
3. J

4. (K: pT) o.

Evidently by 1. and 2., K is a commutative ring and by 1. and 3. iS JT iS -

anl identitv element 1 in K. We have seen that any double P-module 91

cletermines a composite (PiPr, L, R) where L denotes the isomorphism a -- al
and R denotes the isomorphism a- > ar.
In all of our work we shall identify composites (K, S, T) and (K', S', T')
that are equivalent in the sense that there exists an isomorphism k -> k'
betweenK and K' such that (aS)t S', (aT)f
- aT'. The composite (K, S, T)
=

will be called a cover-of (K', S', 1) ( (K, S, T) ? (K', S', T') ) if there exists
a homomorphism k -* k' between K and K' such that (aS)y aS't (T) aT- .

Since K pSpT it is clear that if such a homomorphism exists, it is unique.

In fact, it may be characterized by the fact that it maps S,/S)3T into :aS'/3T'.
This, of course, implies that the mapping S/aT >,aS1/3T' is single-valued.
Thus if ,aS/3T = 0, then caS'/3T' 0. Conversely if the latter condition
holds, the mapping YSfjT aS'13T' is a homomorphism and hence (K, S, T)
? (K', S', T'). The condition that (K, S, T) and (K', S', T') be equivalent

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
6 N. JACOBSON.

is that aS/3T - 0 holds in K if and only if 0S',T holds in K'. Thus -

(K, S, T) = (K', ', T') if and only if (K, S, T) ? (KX,S', T') and (K', S', T')
(K, S, T). If (K, S, 7) ? (K', S', T') and (K', S', T') ? (K", S", T"),
then (K, S, T) ? (K', S", T"). By the fundamental theorem of homo-
morphisms of rings anv composite covered by (K, S, T) is equivalent to a
composite (K - B, $, T) where B is an ideal in K and S and T are obtained
from S and T by applying the natural homomorphism that maps the element
khof K into its coset -k +- B.
Any composite is equivalent to a composite determined by a double P-
nodule. For we may take 9Jto be K and define ax = asx xas and xa = aTx
a= xcT for any x in 91 and any a in P. Then (%: Pr) (K: PT) is finite.
It is readily seen that the composite (Rt,L, R) of 9i is equivalent to (K, S, T).
If 5 is a submodule or a difference module of a double P-module 9 then
the correspondence between 711,8in R and Eal/3r in (E is clearly a homo-
morphism. Thus the composite (P Pr, L, R; i) associated with R is a cover
of the composite (PlPrp L., R; ). If NR el3D 2 then if ta/l3r 0 in both
91 and 2P :Fall3r = 0 in RX. It follows that (PiPrp L, R; R) is a least common
cover of the composites (PiPr, L, R; RI), i- 1, 2, in the senise that anly cover
of the latter two composites is a cover of (P XPr L, R; X). We may also define
the least common cover (K, Sp,1) + (L, U, V) for any two composites
(K, S, T) and (L, U, V) abstractly: For let K ] L be the direct sum of the
rings K and L. The elements of K E) L are uniquely representable iin the
form, kc+ 1, k E C and I EL and (k-+ I) (k' + I') -k1l' + II'. We set
-x aS + aU and aY -= 4- aV. Then the set of elements oaX(aY) is a
field PX(PY) isomorphic to P. If k1, , kq is a basis for K over pT and
. . . lr is a basis for L over Pv, the elements leC, , qkq;li, , 1r form
lip P
a basis for K ( L over PY. Thus ((K + L): PY) is finiite. Hence if
I = pxp Y. we may conelude also that (M: PY) is finite. Thus (M, X, Y)
is a composite of P with itself. Since :/x3 Y .(aS s aU) (/3T + lV)
_ (ZS/3T + U/lv) -0 if and only if YaS/3T 0 and Yu/3v 0, (M, X, Y) -

is a least common cover of (K, S, T) and (L, U,1V). The least common cover
is uniquely determined in the sense of equivalence. In a similar fashion we
may define a least common cover for any finite number of composites.
We have seen that the fixed elements of a self-representation E are the
elements y such that yt 7r in the double P-module associated with E.
=

Evidently these elements are determined by the composite (P,Pr, L, R).

Accordingly we define an element y of P to be a fixed elementt of a composite
r = (K, S, 1) if /S yT. These elements form a subfield rp of P.
-

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS TTHEORYTO NON-NORMAL FIELDS. 7

4. The relations space of a composite. Let r- (IK,S, T) be a com-

posite of P with itself and suppose that (K: PT) - q. Since K pSpT, there -

exists a basis for K over pT consisting of elements a1s, , xaS in ps. For
any a in P we may write

(2) as - alstki(a)1 + ***+ aqSlq (a)T

Since this expression is unique, the mapping M?: a > i (a) is single-valued.
Since (x+/3)s xs+/3s, ,ii(o+P)T=,yi(X)T+,yi(P)T and hence

(3) (a + ,#)Mi ,-am, + #Mi,.

If y is a fixed elenment under r; (y) S Z7S-s aST Hence

(4) (ay) Mi -(am ?,)y

Now suppose that

(5) =tsajS--XkS E k-ijT

Then

eS/S (.S,( X) 1') (jSlj (p)T) Xk-YSEkijT_tQ (X) TMj ( p) T.

On the other hand, SS - (X)S - (X)T So that

(6) (a/3)Mi
j Y.X,yEjX
(aMIx) (#M11).

These equations mav be written ill a more useful form as follows: If A8is any
element of P, we let /3 denote the multiplication a-- a,/ = /3a. With this
rnotationequation (4) becomes

(4') yAfX-M,y

and (6) becomes

(6') i3MI =

where gXi = 2iX; (/3Mg) Equation (3) states that Mi is an endomorphism

of the additive group of P. Of course, the multiplications /.6 are also endo-
miiorphisms. Hence this is true for any transformation of the form YMipi
pi in P.
We suppose now that /11, , fq is a second set of elements such that

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
8 N. JACOBSON.

p1S,. /3 is a basis for K over pT. Then fis

* ,qS =
XjpjSjT where (Pij) R
is a non-singular matrix. If B-1 (ofj) and =

_- (~
S1,1 a)T + * * + f3qSVq(a) T

theni /_L(a) T pi j7vj (a) T alnd vi (a) T Yg.jjTj (a) T. Thus

(7) mi --Y.NjTij and N7 = MjJjj.

lience the totality W of eiidomorphisms (of the additive group of P) of the

form :lMi-i is indepenadentof the choice of the basis. It is clear also that
any two equivalent composites determine the same sets W. The set W is
evidently closed under multiplicationi on the right by the multiplications -.
By (6') W is also closed under multiplication by T on the left. We assert now
that the Mi are right linearly independent relative to the p. For by (2),
c'iJIj = 8ij. Hence a =Mj pi and so if YMjpj= 0, pi = 0 and pT 0.
We shall call W the relationzsspace of the composite (K, S, T) and we shall
-
denote the set of multiplications by P. We have therefore proved the
following

THEOREM1. If f is the relations space of the composite (K, S, T) aind

F is the set of multiplications, Iheit %P?< W and Pff1? W. The right di-
mrensionality ("f p)r = ( pT).6

5. Conditions for covering composites. If N is a two-sided P-module

we have defined the composite of P as (PiPr, L, R) where L is the corre-
spondence c and I? is the correspondence >a r. Now let x1, ,x.
be a basis for P over Pr aicl set axt - x(aEij). As we hate, seen the
correspondlence a --> (aE E'j) is a self-representation.By the well-known
isomorphism between linear transformations and matrices, we may substitute
for a, the matrix cE and for a, the matrix AD -_ {, , } and obtain in

this way a nlew composite equivalelit to (PiPr, L, R). We denote this com-
posite as (pEpD, E, D) where E is the mapping a -_ aE and D is the mapping
> aRD

Consider now the transformations E]j. By (1) Eij is an endomorphism

of the additive group of P and the set of Ei1 satisfy the following conlditions:

6,Using (5) and (6') one can show that the self-representation determined by the
basis Ml of the double P-miiodule W is B' the transposed representation of the self-
representation E determined by the basis a1S, . . *, aqS of K. See a forthcoming paper
of the author's entitled "Conistruction of central simple associative algebras."

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAIL FIELDS. 19

(8) yEj Eijy

if y E 1 the field of fixed elements and

(9) =3Eij Ex(/3Exj).

As in the preceding section we determine the transformations Mi by choosinlg

elements a,, * * 2q such that each a1 ia,7i1 (2) r + **qi[q ((X)r uniquely.

Then aE 1E 1, (2)D + * * * + 2qE[qQ(a)D and

aEii iQ(a)(a1Ejj) + + [q(2) (2qEjj).

Thus
(110) M1(lEfj) + M2(22Ei ) + + Mq(2qEij)

is in the relations space A of our composite. It follows that the set of endo-
nrorphisms of the form YEijpiu, where the p are arbitrary multiplications, is a
subspace e of SKregarded relative to P on the right. We assert that B- W.
For otherwise there exist p < q endomorphisms N1, , Np in 9l such that

Eij N17hjj1 + + NcriPn

or aEi (a^N1) ii, + * + (atNv)o-ijp. Thus each matrix aE is a linlear

combination of the p matrices (r =k (cjj1c) *= k 1, . p, with coefficients
(X7VsD and this contradicts the fact that al, . , CqE are linearly independent
over pD. This proves

THEOREM 2. If E,Tjare the E's determined by a self-representation E,

then the set of endomrorphismsYE,jpjj is the comnpleterelations space f of
thtecornposite associated with E.

Suppose now that Ffi=(K, S, T) is a cover of the composite P' = (K', S', T').
We determine the Mi for K by equation (2). Then

aSt ~ a, S't 1(i )T' + . .. + 2qS'[q(a)T'

and every element of K' is a linear combination of the a St with coefficients

il_ pT'. Now let p31, *, /3p,p ? q be elements of P such that the POs' form.
a basis over pT' of K'. Then we have a7 .35kS'pk5T'. Hence -

a S_Slvl () T' T * * +spS'Vp (a) T'

where Vk(a) T' - t (a) T'pk5T'. The mappings Nk: a -

Vks(a) form a basis

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
10 N. JACOBSON.

for the relations space f(P) of r' and we have shown that N7cYM= p.
Thus %(F) ? L(F).
f
Now suppose converse]y that W(r') ? W(r). Let 9%anid (E be double
P-modules that. have composites equivalent to (K, S, T) anld (K', S', T')
respectively. Let E be a self-representation determined by % and F one
determined by (B. Since A(i') ? f(r) it follows from Theorem 2 that
there exist elements pij.kl such that F1Ci EjTji,71 for all kcand 1. These
=

equations may be writtell in the form

(11) aFk= tr (cEpkl)

where pkl is the matrix (pij,lk ). Now suppose that we have a relatioll of the
form aE/aD -O whereA3D = , p}. Then it follows readily from (11)
that y.aF/3D _ 0. Thus (K. S, T) is a cover for (K', S', T'). This completes
the proof of the following important

THEOREM 3. A necessary and sufficient condition that the composite

r ? rP is that the relations space SC(r) ? W(r').

This implies

THEOREM4. A necessary and sufficient condition that r = r' is that

SE(F) = f(F') .

6. Non-singular composites. If (K, S, T) is a composite such that

(K: PS) < oo, it defines a new composite (K, T, S). It may happen that
both (K: PS) and (K: pT) are finite but that these dimensionalities are not
equal. For example -let K P d>D
(t) the field of rational functions in one
indeterminate over (. Let T be the identity and S the automorphism defined
by ts t2. Then the self-representation determined by K is the isomorphism
a:(t) a (t2) that maps P into a proper subfield. We may avoid this type
of situation by restricting our attention to composites (K, S, T) that are non-
singular in the sense that (K: PS) = (K: PT). If (K, S, T) is non-singular,
the composite (K, T, S) will be called the inverse of (K, S, T). If (K, S, T)
is equivalent to (K, T, S) we shall call this composite symmetric.
Corresponding to the conlcept of non-singular composite, we define a
double P-module 9 to be non-singular if (%R: Pi) (P: Pr). Suppose that
91 has this property. We prove first the following

LEMMA. There exists a set of elements yi, , ym that constitute both

a left and a righlt basis for 9%over P.

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAL FIELDS. 11

The element yi miav be taken to be any element # 0 in 9P. Then yla # 0

and ayl #L0 for all a # 0. Now suppose that yi, , yr are elements of R
which are left linearly independent anld right linearly independent over P.
Then if r < i = (%k:Pi) (%: Pr), there is an element y not of the form
=

>4ayi. If y does not have the form Yyica' either, y may be taken to be the
e]ement Ym+i. Hence suppose that y Yyi'. Similarly we choose an element
=

z not of the form Yyip/3and we may suppose that z =- Y3',y,. Now form
w y + z. Then if wv Xy ,oyi,, z w- y= y, (ai -',) contrary to
assumption. Similarly, wvcannot be represented in the form Y,8,yi. Hence
we may take y,m+i- wv. This process leads to a basis of the required type.
The same method may be used to show that if (E is a non-singulair sub-
module of R, there exists a two-sided basis for 9P that includes a two-sided
basis for (B.
Now if. is a non-singular double P-module, we shall define the inverse
-1 of R to be the module whose elements may be put in (1 - 1) correspondence
with the elements of 9X in such a way that if x -> x' in X,

(x+y)' x'+y'

( 12 ) x)'- x.

Thus X-1 is obtainied from R bv interchanging the roles of Pi and Pr. If

x1, , xm is a left (right) basis for X, then x'1, *, x', is a right (left)
basis for X-1. Let y,, , ym be a two-sided basis for R and asyi yj (aE> ),
yic,=a- (aE*ji)yj. Then we recall that the correspondence a > a-' _ (QxEij)
is a self-representationidetermined by SR. In %-- we have osy'i Ey'j(aE*jj) =

and y'io =
(ji) y'j. Thus -- aE* (aE*j) is the self-representation
-

associated with 9p-1. We note that

=y - Eyj (aEj1 ) E (aEjiE*kj)yk

=,.C ((aEj j) yj - = yk (aE*jjEkj).

Htence (8uik =EEjjE*&k and Avc = :E*jiEkj or

(13) 8EjiE*tj k
ikp Vi]*jiEkj Sik.

Thes2 equations may be written in a simpler form by introducing the ring

(rn of rn-rowedmatrices with elements in the ring e of endomorphisms of the
additive group of P. For we set (E) (Eii) and call this matrix in em the
-

inatrix of the self-representation. If (E*)-(E*ij) then (13) may be

replaced by the single equation

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 12 N. JACOBSON.

(14) = 1
E*E' -

where the prime denotes the transposedmatrix. The representation a >('E* ij)
will be called the inverse E1-1 of the representation E.
If (K, S, T) is a non-singular composite, we have seen that we may take
K' to be a double P-module by setting ax -= ax = xca8 and xc' aTx - XT.

The dimensionality (: Pi) = (K: pS) = Hence i is

(K: PT) (%Y Pr).
non-singular and its inverse is 9-1 (K, T, S) with Pi here pT and Pr PS. - -

If we choose two-sided bases in these modules we obtain a pair of inverse

self-representations of P.
If (K, S, T) and (1L,U, V) are non-singular composites, their least
common cover (M, X, Y) is also non-singular. For, let R be a double P-
module such that R 9l1 P91where the Ri are non-singular submodules
=

and 9Phhas the composite (K, S, T) and W2 has the composite (L, U, V).
'rhen (9P:PI) - (91: Pl) + (9c2>:PI) ~=(91,: Pr) + (o2: Pr) ( Pr) and
so R is non-singular. Thus (MV,X, Y) is non-sinlgular. Its inverse (M, Y, X)
is evidently the least common cover of (K, T, S) and (L, V, U). It follows
from this that if (K, S, T) is aii arbitrary non-sinigular composite, then
(K, S, T) -n- (K, T, S) is a symmetric composite.

7. The product of self-representations. If E and F are two self-

representationsof rank imand r respectively, they determine a self-representation
OL rank mr obtained by substituting for the elements aE,j of aE the matrices

(aEij)F that represent these elements under F. We shall call this repre-
sentation the product E X F of the representations E and F. The product
L X F is an associative one. Moreover if 1 denotes the identity self-
representation a -* (a) of one row, then E X 1 E = 1 X E for all E. =

Hence the set of self-representationis forms a semi-group with an identity

relative to our composition. If (E) is the matrix of E and (F) is the matrix
of F, then the matrix (P) of P =- X F is the (right) direct product
'E) X (F), i. e. P(jil)r it, (j-1)'+l EjiFkl. =

We shall define next a product of double P-modules that corresponds to

the product of self-representations. Let R and (E be two double P-modules.
We wish to construct a double P-module $ and a function xy of x in T and
of y in (E such that the following conditions obtain:

'1. a<(xy) (acx)y

2. (xy) a =x(ya)
3. (xa))y x(aty)
4. x(y+y')= xy+y'

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOlS THEORY TO NON-NORMAL FIELDS. 13

a. (x + X')y xy + X'y
6. Any element of $ has the form xry for suitable x in 1 and y in (.
7. ($: Pr) (R: Pr) ((B: Pr)

Let xi, , m be a right basis for i and axi = Exj (aEji ) and let yi, , Yr
be a right basis for (E and ayk y (cF 1,). Then if x
= Exiej and y = = y:7 k
conditions.1.-5. imply that
(15) xy = Exjy 1(ejFj7z) q,.

By 6. every element of $ is a linear colubination of the elements xiy7. Hence

by 7. these elements form a right basis for $ over P. The procedure for
defining $ and xy is now clear. We let V be a right module of dimensionality
rj? over P and define xy by (15) where Xiyk = Z(i l)r+k form a basis for $
Then 2., 3., 4., a. and 6. are valid. Next we definieal to be the linear trans-
formation in $ over P,. that sends xiy, into :xjyi (cEjiFik). We set acz=aZ
and we observe that c (xy) - :xxyc (c#ExjFcu)(ejF,!7c) . We may now verify
that 1. holds and hence - is a double P-module and xy is a function satisfying
our conditions.
If X'i, . . , X'm anld y'i, . . , Y"r,' respectively, form a second pair of
right bases for 9R and for (E, x = :xie% and y :y'7'V0. Hence =

xy :XIjoytl(cay'1k)q Lk. Thus all the xy are linear combinations of the ele-
mnents x/iy'k and so these elements form a second right basis for $. Now
suppose that we construct a second space $' and a second function xy which
we shall denote as [xy] by using the bases x'j and y'k in place of the xi and
the yk. Then it is readily verified that the mapping :'iy,'keik -> [X'iyk'] ek
is an isomorphism between the double P-miodulesl anld$'. In this seiise the
module l does not depend on the choice of the bases xi and yk. We shall
therefore denote this module as Sc X (a and shall call it the product of the
minodules9c and (B.
By definition
a(Xiyk) = sxj(aEjik).
Hence the representation determinied by the basis zs, Z(i1)r+1 Xiyk is the
product E X F. The independence of $ =T X (E of the bases of 9 and (E
may be stated as

THEOREM 5. If E' and F' are self-representations similar respectively

to E and F, then E' X F' is similar to E X F.

If x'1, . , x'p are right linearly independent in 91 and y',, y'q are
right linearly independenit in (D then the elements x'jy', are right linearly

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
14 N. JACOBSON.

indepenidentin . :For we may supplement the x' and the y' to obtain bases
* '1, , X'r;I/i, , y'. Then we have seen that the elements X'iY'k are
linearly independent. Now suppose that the elements x'/, ,x'p form a
right basi3 for a submodule ' of EPcand the elements Y'i, ,ytq form a
right basis for a submodule (E' of (5. Then the elements x'jy'l form a basis ior
a submodule $' of $ and 43'is essentially the module 9t' X (D'. This proves

THEOREM 6. If E' is a subrepresentation,of E and F' is a subrepresenta-

tion of F, then E' X F is a subrepresentation of E X F.

In a similar manner we mav prove

THEOREM 7. If E is decomposable into the components E, and F is

decomposable into the components Fj, then E X F is decomposable into the
components Et X Fj.

We suppose n1ow that R and c5 are non-siiigular double P-modules and

that the xi and yk are two-sided bases. Let the elements x' and y'k form the
corresponding two-sided bases for X-1 and (-1'. If ax, xj (aEji), ax',
=,xfj (aE*ji) where the self-representation E* thus determined is the
inverse of E. Similarly ayk ,:yj(aFl:k) where F*
and ay'k Ey'z(a>F*Ik)
is the inverse of F. Now x,a = Y(aE*j)xj and yka E(akF*lk)y, and so
(XiYVk)a (akF*lkE*ji) Xjy. iHence every element of 9 X e is a left linear
combination of the elements Xiyk. These form a basis. For if :IKUikXiYk = 0,
0
Yxjyl/ilicEjiFl7k and hence :KCE =jiF 0 for all j and 1. It follows that
Ai'k'-El,ThlikEjiFV7cF*'llEi','j = 0. We have therefore proved that 9 X e is
non-singular and that the elements xiy7 form a two-sided basis for this module.
The elements (Xiyk)' form a two-sided basis for the inverse module and we
have a (xiyc) (xjyl' (aF*L i). Our argument shows also that the
elements y'k'i form a two-sided basis for (5-1 X 9-1 and we may verify that
ay'kx'i = :y'xl'j (aTl.kE,'*ji). Hence the modules (N X 5)-1 and 25-1 X 9V-1
are isomorphic.

8. The product of composites. It is easy to see from Theorems 2 and

4 that the composite determined by R X (E depends only on the composite of
R and the composite of 5. For the composite of NPX (E has as its relations
space the smallest subspace over F (on the right) containing all of the products
MN where M is in the relations space of the composite of RPand N is in the
relations space of the coinposite of (B. Thus the composite of 9 X (E may be
regarded as the product of the composite of %cby the composite of (. It is of
iiiterest to give a direct definition of the product of composites. We shall

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAL FIELDS. 115

obtain such a definition by introducing a type of three-fold composite of P

with itself. Aside from its uise in the present coniiection, the concept of a
three-fold composite will play an important role in the proof in 9 of one of
the main results of the present theory.
We define first a three-fold composite of P with itself as a system
(A, A, B, C) consistinig of a ring A and the isomorphisms A, B, C of P into
s-Lbfields pA pB PC of A suLchthat
1. A pApBpC

2. A is commutative
3. 1A =1B jC.

Evidently by 1. and a., 1A _= - lC is the identity 1 of A.

Coversof three-fold composites are defined in the obviousway: (A, A, B, C)
is a cover of (A', A',B', C') if there is a homomorphism a -- a' betweell A
and A' suchthat (aA)' - aA', (a'B)' - aB" (acC)'= aC'. Equivalenceis defined
in a similar manner.
Now suppose that (K, S, T) and (L, U, V) are arbitrary (two-fold)
composites. We wish to construct a three-fold composite (A, A, B, C) having
the followilig properties:
4. (PAPB, A, B) is equivalent to (K, S, T) and (PBPC, B, C) is equivalent
to (L, U, V1).
5. (A: pC) = (K: PT) (L: Pv).

We shall show first that any two three-fold composites having these properties
are equivalent. For this purpose let x,, , %7be a set of elements of P such
that alS,. . as form a basis for K over pT and let * q be elements j&iy

such that ,lu, . /. , u form a basis for L over pv. As before, we determine
the endomorphisms a -> li (a) =- acMi and a -> vj(a) aNj by writing

as- a,Sl1(a)T +. . a
+qSyq (a)T
= V + + V.
-U U,UV,(a) /q,TVq, (a)

Suppose also that

=
tais(zS aXXSCX,,T /jUfj,U- pvuvjj'v.

Iii1 Av we have 4A=_ aA (a)B. aB - .jBVj (a) C and -A-,A AEXA B'
/3jB/3JB k/3VBqvjj,C. Hence

(16) XAt, =-- C =

- 4XAEX,i,B xAjBVj(,EX,,)C 3.B/j,B :/v8^vjj,C.

Let 2i denote the totality of elements of the form ,a,A8jBC,jC. By (16), Z is

a subring of A. Since

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 16 N. JACOBSON.

(17) aA =_:a 1AP' (a) IB :>ajApjIVj (11 (a))C

pA C IHence 1 and PC_ &. Atoreover,pjB

2. (a.A)
-1
((A/jB)
-
EIa anld so
each aB =-:jBVj(a)C iS in 5. Thus =A and every element of A is a
linear combination with coefficients in pC of the qq' elements cXAf/jB. Hence
by 5. the elements cxiA/jB form a basis for A over pC.
Now suppose that (A', A', B', C') is a second three-fold composite having
the properties 4. and 5. Theni the mapping :>/iA3jBtjC -> 'aiA1/jB1C,jC1 is

clearly (1 - 1) and by (16), it is aii isomorphism. It is evident from the

above considerations that this isomorphism maps AA into AA', aB into aB' and
aC into aC'.
We now construct the composite (A, A, B, C). Suppose that the elements
acS and /3jU have been choseii in the way indicated above. We shall suppose
in addition that a,s - S and ,u -- lu. Now let pC be a field isomorphic
to P under the isomnorphisma --> ac anid let A be the algebra over pC with the
basis aiAajB and the mnultiplicationtable

(18) (aiA/jB) (a-,,fj,B) =

axA[B Cr,vCqvjj,CVk (EX,,) C.

We shall show that A is associative and that A determines a composite satis-

fying our conditions. We note first that a,A,8/B acts as the identity 1 of A.
The elements/jB ,= aA#jB satisfythe multiplicationtable ajB/3j,B = CavB vjjlC
and these elements are linearly independent over pC. It follows that the set
of elements aB I3Bvj(g)C is a subfield pB Of A isomorphic under the
correspondence aB a( to P. The totality of elements :/3jBpjC where the pj
are arbitrary defines a composite (pBpC, B, C) equivalent to (L, U, V). We
note next that if we set aiA a.A/_B the product of this element with /3jB
(in either order) is the element tA/3jB of A. The elements aiA satisfy the
relations
a .A(iA - x AuXABv1 (,EX,,,) C - -aXAEXi,'B

and these elements are linearly independent over pB. If we make use of
equations for the vk analogous to (6), we may prove that (acApB) (aiA0B)
,>ax Exji,BpB aI. It follows that the totality of elements of the form
aiApiB is a ring isomorphic to K. The set of elements AA ( is a =

field pA isomorphic under the correspondence aA A. a to P and the set of

elements >aXApi3 determines a composite (pApB, A, B) equivalent to (K, S, T).
Now the elements of A may be written in one and only one way in the form
ostAI, where 11,E pBpC. We may prove that (acAl) (ai,AM) -caXA (E ,BlM)
and this implies that A is associative. Since A is generated by pA, pB and

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAL FIELDS. 17

pC_ lpcp A is commutative. Conlditionis4. and 5. are evidently satisfied.

Heence (A, A, B, C) is the required three-fold composite.
We iiow define the pr-oductof the composites (K, S, T) X (L, U, V) as
the composite (PAPC, A, C) determined from the three-fold composite
(A, A, B, C) that we have constructed. This product is uniquely determined.
It can be seen that if (K, S, T) is a cover of (K', S', T') and (L, Up,V) is a
cover of (L', U', V'), thell the three-fold composite (A, A, B, C) is a cover
of the three-fold composite (A', A', B', C') constructed from (K', S', T') and
(L', U', 1'). It follows that (K, S, T) X (L, U, V) is a cover of (K', S', T')
X (L', U', 17'). If (K, S, T) and (L, U, V) are non-sinigular, the three-fold
composite (A, C, B, A) has the same relation to (L, V, U) ancd (K, T, S) as
(A, A, B, C) has to (K, S, T) ancd(L, U, T). Heence (L, V, U) X (K, T, S)
_ (pApC, C,A4) andl the latter is the inverse of (pAPC, A, C). It is not
difficult to give a direct proof of the associative law for multiplication of com-
posites. For this purpose it is necessarv to construct of four-fold composite
(E, A, B, C, D) such that (pApBpC, A, B, C) is equivalent to (A, A, B, C) and
(pBpCpD, B, C, D) is equivalent to the composite (A1, A1, B1, C1) determined
by (L, U, 1) and (M, X, Y). We shall not carry through this proof but
instead we shall obtain the associative law indirectly by determining the
relations space of the product of the composites (K, S, T) and (L, U, V).
Let 81A, , 8A be a basis for pApC over pC and suppose that
(19) aAA= 1(a)C+ C . *+ 82 A7r(a)a

Then the endomorphisms a ->>7r,(a) P7 form a basis over P of the rela-

tions space of the composite (PAPC,A, C). Since akA =aA1,(k)B, BA =
C and since B
AiAyi(&) B,rk (a) ui (&k) = :jBVj (y *(8k) ) CP

YA= atjVj (/X (8k) ) Cvk (a) C

lIence by (I7) VJ(Ili (a)) =YVj (pt (87k))7rk(a), or

Milyj - PkVj (/ti (8k))

is in the relations space of '(pAPC, A, C).

Now let RX, , 1R be a basis for the space over P (on the right)
generated by the product 1/liNj and write aRk p;k(a). Then by expressing
thle elements M Nj in terms of the Rk we may replace the relations
AA &a,A&jBvj (j,(a,) )C by

(20) E gkpk(2)C

where the gk EA. We wish to show that the gk E pApC. For this purpose we
require the
2

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
18 N. JACOBSON.

LEMMA. If R1, , R, are enidomorphismsin P that are linearly in-

dependent over P, then there exist elements A1, 7Ic 1, , s such that the

m,atrix (XkRT) is non-sin.gular.

Suppose that we have already determined r elements Xk such that the r
vectors (XkRi, ,kX,R) are linearly independent. Then we assert that if
r < s, there exists an element A+i such that the (r + 1) vectors (XkRt) are
linearly independent. For otherwise for each X we have
(21) XR1z=XiRi J
(X) + * * + XeRlr (X), (l=, , S).

If EXkRtok 0 for all 1, by the linear independence of the vectors (XjRl)

for k 1, ,r, each ok 0. It follows that the a- in (21) are uniquely
=

determined and so the correspondence X -> ok (X) XSk is single valued. It

is clear that Sk is an endomorphism of P and the relation (21) states that
each R, is a linear combination with coefficients in P of the r endomorphisms
S.k. This contradicts the linear independence of the R's and proves the lemma.
Now choose the X's as in the lemma and write XiA g,(kpX(i)C. Since
(pk(Xl) ) is noin-sinigular,there exist T7ru'such that Ypk(xA) CTlklcC &8k'. Then

=k' XA7tlkC is in pApC. We now write gk 1tA/ikC and substitute in

=

(20). This yields AA .8A/3lkCpk(a)C and so by (19) 7n-t(a) = /3zkpk(c&)

T'hus Pi = )Rkf-+,i is in the space generated by the products .IiNj. This
completes the proof of
THEOREM 8. The relations space of a product of two composites is the
smallest space over P containing all products. MN where M is in the relations
space of the first composite and N' is it -the relations space of the second
composite.
This along with Theorem 4 proves that multiplication of comnpositesis
associative. Now it is readily seen that the relations space of the least common
cover (K, S, T) + (L, U, V) is the join of their respective spaces. It follows
from Theorem 8 that the distributive laws hold: [(K, S, T) + (L, U, V) ]
X (M, , Y) (K. S, T) X (M, X, Y) + (L, U, V) X (M, X, Y) and
(MP X, Y) X [(K, S, T) + (L, U,pV)] (M, X, Y) X (K, S, T)
+ (M, X, Y) X (L, Up,V). We note also that any two composites
(PU, U, U) are identical and (PU, U, U) satisfies the equation (K, S, T)
X (PU, U, U) = (K, S. T) (PU, U, U) X (K, S, T). For this reason
we shall call (PU, U, U) the identity composite.
If we refer to the preceding section we see that the composite of a product
of two representations or modules is the product of the corresponding
composites.

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAL FIELDS. 19

9. Closed composites. We shall call a composite rP (K, S, T) closed

(under multiplication) if (K, S, T) ? (K, S, T) X (K, S, T). By Theorem
3 and the considerations of the preceding section, (K, S, T) is closed if and
only if its relations space W(r) is a ring. A composite (K, S, T) will be
called a Galois composite if 1) (K, S, T) is closed, 2) (K, S, T) is a cover
of the identity composite (PU, U, UT)and 3) (K, S, T) (K, T, S). It will =

be a consequence of the results of 9 and 10 that any closed composite is a

Galois composite. Hence the conditions 2) and 3) can be deleted in the
definition of a Galois composite. We have preferred, however, to state them
as separate conditions since they correspond to the existence of the identity
and of inverses in a group.
We suppose now that r _ (K, S, T) is a closed composite and we let (r
denote the set of fixed elements relative to r. Let a,, , aq be elements
of P such that a1S, , qS form a basis for K over pT. We assert that these
elements form a basis for P over 4r. To prove this we consider the three-fold
composite (A, A, B, C) that satisfies conditions 4. and 5. for (K, S, T) and
(L, U, V) = (K, S, T). In (K; S, T). we have as alspl(ax)T + -

+ aqSlq(a)T. Hence in A we have the relation aA aA/Q(a)B. Since

(K, S, T) is closed, (pAPC, A, C) is covered by (K, 8, T). Hence
aA :ApQ(a)C Thus a,A (K(a) -(a) C)
- 0. We have seen that
the elements are independent over pBpC. Iience this relation implies that
a,A
,}(a)B p(a)c anld so each yi(a) Epr. Thus each a in P is a linear com-
bination with coefficieiits in (r of the elements (i, *, aq. Since the elements
aiS are linearly independent over pT, the elements ai are linearly independent
over4r.
THEOREM 9. Let r = (K, s, T) be a closed cormposite,(K: pT) q and
let Fr deinote the field of fixed elements relative to r. Then (P: (r) q =

and if (i, , .q are q elements of P, they form a basis for P over (r if and
only if the elements 21s, a*, ,qs form a basis for K over pT.

The necessity of the conditionl is trivial. For if a,, a, from a basis

for P over (r, each as (xaS'yT+ . + aqSlT where the 7T E rT rS. =

Hence the as are linearly independent over pT

We consider now the relations ring 9f(r). We have seen that its dimen-
sionality over P is q. Hence the dimensionality of af(r) over 1r is q2. On
the other hand we have seen that the transformations belonging to af(r)
commute with the elements of pr, i. e., they are linear transformationlsin the
space P over Ip. Since (P: 1r) q, it follows that 9I(r) is the complete
P
set of linear transformatioiis of over 4Ir.

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 20 N. JACOBSON.

THEOREM 10. if r (K, S, T) is a closed composite and 4r is its field

=

of fixed elements, the relations space gI(r) is the complete ring of linear
transformations of P over 1p.

Suppose nlow that r' (K' S', T') is any composite of P which leaves
the elements of Fr fixed. Then if N is any element of the relations space
%1(r'), N is a linear transformation in P over 4or. Hence by Theorem 10,
N1Ve(r), and so %f(r') ?< Wf(r). This implies the followinlg

THEOREM 11. Let r be a closed composite and BI, its field of fixed ele-
ments. Then if P' is any composite sutchthat 4)rP ? r, r' ? r.

This, of course, implies that there is only one (in the sense of equiva-
lence) closed composite having J = (Dr as its field of fixed elements.
It may be remarked that if A (1, U, V) is any composite such that
=

(P: () = r < co, then (L, U, 17) is non-singular. For if we set DA ,

then (Pu: (Du) = r = (pv: 4V). Hence if (L: PV) - q', (L: v) - rq' and
since (DV- (DU, (L: =") rq'. It follows that (L: Pu) = q'. A similar
argument may be used to show that if (E is any double P-module with com-
posite equivalent to A, then (E is non-singular. This may be stated as follows:

THEOREM 12. If F is a self-representation of P such that the dimen-

sionality of P over the field of fixed elements is finite, then F is similar to a
non-singular self-representation.

We again suppose that A = (L, U, V) is a composite such that (P: iz )

-r < oo. If A is not closed, we form the least common cover A(2')
+ A X A. Then the relationsspace W(A(2)) > %(A). It is readilyseen that
the field of fixed elements 1(A2M (D so that (p: 1(2)) =- r also. If A(2)
=

is not closed we set A(3) -A- + (A X A) + (A X A X A) and note that

9((A(3)) > %f(A(2)) and 'A(3) (A(2), Since WI(A), I(A (2)), f(A'(3)) all
have dimensionalities over ESA? r2, the dimensionality over PA of the com-
plete ring of linear -transformations of P over (DA,this process leads after a
finite number of steps to a closed composite r AI+ (A X A) + * +
(A X * X A). By Theorem 10 we obtain

THEOREM 13. Suppose that a -> (aFki) is a self-representation of P

such that (P: 1) < oo for 41 the field of fixed elements. Then if L is a linear
transformation of P over (, L is expressible as a polynomial in the transforma-
tions ki with coefficients in P the field of mutltiplicationsin P.

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GAIOIS THEORY TO NON-NORMAL FIELDS. 21

10. The fundamental Galois correspondence. We suppose now that

P is an arbitrary field aiid that (F is a subfield of P such that (P: (F) q < ?o. =

We take two copies Ps and PIT of P and form the direct product K of these
fields relative to D.7 Thus K is the set of sums 0SfT, AS in pS and AT in pT
where ys 7T if y e b and addition and multiplication are defined in the
-

obvious way. The dimensionality (K: (bS)- q2 and (K: PS) = q = (K: PT).
WATerecall also that ps A\ pT jS - T. It is evident that rP (K, S, T),
where S is the mapping a -> as and T is the mapping a > aT, is a composite
of P with itself. Since the mapping aS aT, A aT aS
A is an automorphism
in K, (K, S, T) is symmetric. Sinice pS A pT -S 1T, the field (r
of fixed elements is (. Now let (T1,U, 1V) be any composite of P leaving the
elements of ( fixed. Suppose that >SfT 0 in K. Then if 1, , aq
form a basis for P over cD,the elements aisaj form a basis for K over -S T

and the elements xiUajv are generators of L over (DU -v. We replace 0S
by Eysi85 yi in 4) and 8T by ,8jTajT5 8j in ( and substitute in the relation
-
>SfaT 0. Then the coeflicients of the products Saj.T in the resulting
expression are all 0. Hence if we substitute au = EY-u(ju and /3V- Y8jVxjV
in XU,pv, we see that au/jv - 0. We have therefore proved that (K' S, T)
is a cover for (L, UT,V). This shows, in particular, that (K, S, T)
(K,S,T) X (K5,S,T) and (K,S,T) ? (PU, U, U).
THEOREM 14. If P is a field and (Da subfield such that (P: (D) = q < oo,
then the direct product of P ove- (Dwith itself determines a Galois composite
whose field of fixed elements is ?.
Now suppose that r is anly closed composite aiid let (r be the subfield
of P of fixed elements under r. We have seen that (P: or) < oo and hence
we may form the direct product F' of P over 1r by itself. By Theorem 11,
r' ? r and by what we have just proved r ? r1. Henice r is equivalent to
r' and so we have proved

THEOREM 15. Any closed composite is a Galois composite.

As a consequeniceof Theorems 9, 11, and 14 we have the following
fundamental
THEOREM 16. Let P be a fixed field and r a Galois composite of P. If I)r
denotes the field of fixed elements under r, then P is finite over 1r and the
correspondence r -> pr is (1 - 1) between the Galois composites and the
subfields ( over which P is finite. rP1r? if and only if Dr. ? 1p2.

XFor the definition and properties of the direct product see, for example, R, p. 88.

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
22 N. JACOBSON.

11. Decomposition of composites. Let r (K, S, T) be an arbitrary =

composite. Since K is a commutative algebra with a finiite basis over pT,

we may decompose K as a direct sum

(21) hZ K, 0 Ks

of indecomposable algebras Ki. The Ki are uniquely determined. Moreover

if B is any ideal in 1f, B = 1 . (9 B, where Bi-B A Ki. Hence if
N is the radical of K NT -T1 E NY, where
= Ni eN A Ki is the =

radical of Ki. We recall now that any indecomposable commutative algebra

is completely primary in the sense that the difference algebra with respect to
its radical is a field.8 Anarideal of such an algebra is contained in the radical.
It follows that the homomorphic image of such an algebra is also completely
primary and hence is also indecomposable. These results apply in particular
to the algebras IKi.
Corresponding to the decomposition (21) we write
1 el + * *+eS
where
e,i2 ei, eiej Of #i j.

Then Ki =- Kei. The mapping k--> k-ei is a homomorphism Ei of K into Ki.

\Vre set Si = SEi, Ti = TEL. These are isomorphisms between P and sub-
fields P83and pTm of KI. Since K pSpT, Ki pSipTi. Evidently (K: PT) =

-- (Ki: PT) and sinice 7i(T tgaTi if ki EKj, (K: pT) (K.: pTi).
H[ence (Ki: Pt7i) is finite. Thus (Ki, St, YTj) is a composite of P. It is
evident that (K,,5 T) is the least common cover of these composites. We
shall now show that the composites (Ki, Si, T) are disjoint in the sense that
there exists no composite (L, U5,V) which is covered by both (Ki, Si, Tj) and
the least common cover of all the (Kj, Sj, Tj) for j #- i. For let (L, U, V)
b,esuch a composite. Suppose that et - :c&/T. Then ei =- aSi#T' and -

o eej :SJTj for all j #7 i. In L we have 1ZJ ->Ua3V

= - and 0- Yuv
anldthis contradictioniproves our assertion.
If B is anl ideal in K and k1-> k denotes the natural homomorphism
between K and K = K - B, it is clear that (K, S, P) where as as, AT AfT
P
is a composite of covered by (K, X, T) and, as we have remarked, any com-
posite covered by (K, S, T7) can be obtaiiied in this way. If C is an ideal
containing B, it is readily seen that the composite (K - B,S ,T ) is a cover
of (K - C, S, P). Now since any ideal Bi of KI is contained in the radical

8 Cf. Theorem 3, Chapter 4 of 1.

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAL FIELDS. 23

ANi,this shows that any composite covered by (Ki, Si, Ti) is a cover of the
composite (Ki - Vi, Si, Ti). Thus it is impossible to find disjoint com-
posites which are covered by (Ki, Si, Ti).
From the decomposition K K1 . = e
Ks we obtain K - B =
K] + + Pi. Also B B e .e DBs where Bi B A Ki. Hence
if kl + k2 + + Es +s- 0 kl + k2 + scS bi+ b2 + + bs where
the bi EBi. Then 7li= bi is in BI and ki 0. Thus k= K] Kse e
Each Ki is indecomposable since it is the homomorphic image of the indecom-
posable algebra Ki. We suppose that kR # 0 if i < r and that 00 if
i > r. We now denote the mapping k -> f by H, the mapping To--> Ii by
Lei and we set Si =SHEJi, T i= THEiYj. Then (Ki, Si, Pi) is a composite of
P with itself. Aloreover,these composites are obtained from (K, S, P) in the
same way that the (Ki, Si, Tj) are obtained from (K, S, T). We assert now
that (Ki, Si, Pi) is coverec[by (Ki, Si, Ti). For suppose that YSt/3' 0. -

Then if k = S/T:s l, -i 0. Now in the decomposition k= 1 + + kI,

ki. is the coset of ki. Hence ki 0. Thus ySi/T-
= (>S/T)HEt 0 =

anid this proves our assertion. In particular we see that if (K, S, P) is

iitdecomposable, i. e. if r 1, then (k, S, P) is covered by one of the corm-
=

pousites (Ki, Si, Ti).

Now suppose that (K, S, T) is the least common cover of the composites
(K'j, S'j, T'j), j =1, , s' which are disjoint and indecomposable in the
senisethat they are niot least common covers of disjoint composites. Then each
aigebra K'j is indecomposable and since (K, S, T) is a cover of (K'j, S"j,T'j)
one of the (Ki, Si, Ti) is a cover of (K'j, S'`, T'j). Since the (Ki, Si, Ti)
are disjoint, onily one of these, say (I-, Sj, Tj-), is a cover for (K'j, S'j, T'j).
Since no two composites covered by (Kj, Sj, Tj) are disjoint, it follows that
if j k
I, then i k. If we re-arrange the (Kj, Sj, Tj), we may suppose that
j j. VWewish to show that (Kj, Sj, Tj) and (K'j, S'j, T'j) are equivalent
and so we assume that this is not the case. Then there is a relation yXS'j,jT'j
= 0 such that ).Sj,LTJ I 0. If ej- and XSiaSjyTi/TJ
ej =S/TT #S/Tj
00. Since ejek=0 if j k,ISk/Sk
# 0 and hence :XSkXSk&T1 /Tk -.
=

Thus we may suppose at the start that 0XS7?Tk for k _ j. It follows that
=

'~S'kXJ'k 0 andl hence that AST 0 a-nd this contradicts the fact that
=

'XSjJTj 00. Now it is clear that s' s. For otherwise there exists a
=

(Ki, Si, Ti) which is covered by the least common cover of the (Kj, Sj, 'j)
for j 7& i. We have therefore proved the following:

THEOREM 17. Any composite (K, S, T) is expressible in one and only

one way as the least comn-mon
cover of disjoint indeconiposable composites.

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
24 N. JACOBSON.

We shall call the indecomposable composites (K., Si, Ti) the indecom-
posable components of (K, S, T). The argument given above proves

THEOREM 18. If (L, U, V) is covered by (K, S, T) then each indecom-

posable component of (L, U, V) is covered by an indecomposable component
of (K,S,T).

Suppose now that (K, S, T) is a Galois composite. Then we have s,een

that any composite covered by (K, S, T) is non-singular. This applies in
particular to the inidecomposablecomponients (Ki, Si, Ti). Consider now the
inverses (KI, T., Si). It is clear that these are indecomposable and disjoint
and that their least common cover is (fK,T, S). Since (K, S, T) (Ki T, S) =

it follows that the set of inverses (Kis,T,, Si) coincides with the set (Ki, Si, Ti).
We observe next that since (K. S, T) X (K, S, T) < (Kf, S, T) and sinice
(K, S, T) X (K, S, T) is a cover of every product (Ki, Si, Ti) X (Kjb Sj, Tj)
the indecomposable componenitsof these products are covered by suitable oni-es
of the (Kjc,S7.,T1g). We niote finally that since (K, S, T) is a cover of
(Pu, U, U) one of the (KA1, Si, Ti) is a cover of this composite. Conversely
suppose that (K, S, T) is any composite whose indecomposable components
(Ki, Si, Ti) satisfyv

1. Each (K1, Si, Ti) is non-singular,

2. One of the (Ki, Si, Ti), say (Kr, S,, T,), is a cover of the identity com-
posite (PU, U, U),
3. For each (Ki, Si, Ti) the inverse (Ki, T,, Si) is an indecomposable
component,
4. The indecomposable componenits of' (Ki, Si, Ti) X (Kj, Sj, Tj) are
covered by the (Kk, Sk, Tk) .

Then since (K, S, T) X (K, S, T) is the least common cover of the products
(K,, Si, T,) X (Kj, Sj, Tj), (K, S, T) is a cover of (K, S, T) X (K, S, T).
It is evident also that (K, S, T) ? (PU, U, U) anid that (K, S, T) is
symmetric.

THEOREM 19. Conditions 1.-4. on the indecomposable components of a

composite are necessary and sufficient that the composite be a Galois composite.

12. Separable fields. We shall call a composite (K, S, T) simple if K

is a field. In this case the only composite covered by (K, S, T) is (K, S, T)
itself. (K, S, T) is semi-simple if its indecomposable components are simple.
If the components of (K, S, T) are (Ki, Sj, Tj), i =1, *, s, it follows that

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAL FIELDS. 25

any composite covered by (K, S, T) is the least common cover of certain of

the (K1, Sj, Tj).
We shall prove first the following

THEOREM 20. Let P be a field, b a subfield such that (P: 4) q < oo

antd let r K, S, T) be the Galois composite whose field of fixed elements
(ff

=r c1. Then a necessary and sufficient condition that P be separable over F

is that r be semi-simple.

Sufficiency. Let acbe any element of P and set Y &J(0P) where p is the =

characteristic of (D. Consider the Galois composite A = (L, U, V) of P having

= I(A as its field of fixed elements. Since A is covered by P, it is semi-
simple. Now (aU)) (P s)U= (e P)V (&TV)P. Hence ( au_-aV)P o and
so au c v. Thus e =4 c(csP) and hence a is separable.
Necessity. Suppose that P is separable over 1. Then there exists a
primitive element a in P such that P (D(a). Let (L, UT,V) be an indecom-
=

posable composite and let pV (X) be the minimum polynomial of aU over the
field pv. Since L is an indecomposable algebra Yv(x) 1lv (x)e where 1fvt(k) =

is irreducible. For if kV(X) = 1IVv(X) 1f2v(X) where (frV(X), /2V(x)) 1,

,//" (aU) is a zero-divisor -7e0 in L and since any zero-divisor of a completely
p-rimary algebra is contained in the radical qij (au) f - 0. This implies that
v1V(A)f is divisible by /, V(k)q,9(V) contrary to (qj1V(k), 1fr2V(A)) 1. Now
let ,(A) be the minimum polynomial of o:over '1. Then pYv(X)- cV(k)I,iV(k)
_lv (A) e1rV (A) . It follows from this that (,u(A), i'(Ak) 1 if e > 1. Thus
e = 1 and aU satisfies an irreducible polynomial over pv. Hence L - PV(aU)
is a field. We have therefore proved that any indecomposable composite of P
is irreducible an-dhence any composite is semi-simple.
We now recall the definition of a hypergroup (with an identity) H as a
system in which a product of pairs equal to a subset of H is defined. If A and
B are subsets of H, we define AB to be the set of elements contained in all
products cub,a in A aiid b in B. It is assumed that 1) the product is
associative: the set (ab)c a(bc), 2) there exists an identity 1 in H such
that al =a a for all a in H and 3) for each a there is an inverse b such
that ab and ba contain 1.
Suppose now that H is a set of non-sinigularsimple composites of P havilng
the following properties: 1) If rI and r2 belong to H, then rl X r2 is the
least common cover of certain ri in H. 2) H contains the identity composite.
3) H contains the ilnverse r-P of any r in H. Then if we define rI.r2 to be
the set of simple composites contained in H and covered by rI' X r2, it is easy
to see that H is a hypergroup. We shall therefore call a set of non-singular

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
26 N. JACOBSON.

simple composites with the properties 1), 2) and 3) a hypergroup of simnple

composites.
Now let H be a finite hypergroup of simple composites and let r = (K, S, T)
be the least common cover of all the ri in 1i. If (Ks, Si, Tj) are the indecom-
posable components of r, it is evident that each ri is covered by one of the
(K,, Si, Ti) no two FT are covered by the same (K,, Si, Ti) ancd no two
(K,, Si, Ti) cover the same ri. It follows that the rT coincide with the
(Ki, Si, T). Hence by Theorem 19, (K, S, T) is a Galois composite and
since its indecomposable components are simple, (K, S, T) is semi-si-mple.
Let 'CIH be the field of fixed elements relative to the ri. Then IH Irp and =

P
hence is finite and separable over 4H. Conversely suppose that P is finite
and separable over 4F and let r be the Galois composite such that 4p = (.
Then we have seen that r is semi-simple and it follows from Theorem 19 that
the indecomposable componientsrT of r form a filite hypergroup II of simple
composites. Evidently BH @.= We have therefore proved

THEOREM 21. Let P be a fixed field and H a finite hypergroup of simnple

composites of P. Then if 4ZI denotes the field of fixed elements under all the
composites in Ni, P is finite and separable over BH. The correspontdence
HI - I? is (1 - 1) between the finite hypergroups of simple composites of P
acndthe subfields ';DH over which P is finite and separable. H1 ? H2 if cand
only if @iiH < DFI2@

13. Normal fields. AWeshall call a composite (L, U, 1V)one-dimensional

if (L: PV) 1. Evidently this implies that (L, U, V) is simple. With this
-

definition we have

THEOREM22. A necessary and sufficient condition that a field P filnite

over a subfieldd be nornmalover D is that every simple composite (L, U, IT),
whose field of fixed elements contains 1I, is one-dimensional.

Suppose that P is normal over J an-dthat (L, U, 1) is a simple composite

-hose field of fixed elements contains 1. Let a be an element of P and let
(A) be its minimum polynomial over 1. Sin-ceP is normal, pA(X) = (A ,)
* (x - a.1), a aC, in P[A]. Hence tx1Y(X) H(X - Iv). Since
,lV(aU) 0, 1I(au- a.) - 0 and since L is a field aU - a.V for a suitable i.
Thus pu ? pv and (L: PV) _ 1.
Next suppose that for every simple (L, U, 17) whose field of fixed elements
contains D, (L: PV) = 1. Consider the Galois composite r- (K, S, T) such

9 Cf. Kaloujnirie, Icc. Cit. 3.

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAL FIELDS. 27

that Ir - (D. Let K = K, 0 . *? K8 where the K' are indecomposable.

If 1Vis the radical of K, N =-- N, 3 *. Ns, Ni the radical of Ki. We define
the Si, Ti as in 11 and consider the simple composite (Ki - Ni, S, T)
where Si, TAare obtained by first applying Si, Ti and then the natural homo-
morphism between Ki and Ki - ATN. By our assumption for any a in P,
-XS = for a suitable /3. Hence (cS8 0 for some ni. We
/3-,Ti)ni
assume that ni is minimal and we may assume also that the notation has been
chosen so that P,3 Pr, 81' / r,+l ***
3rL+r2 = 82 # and
-
that mn1 n, ? A2 > nr,; M2 - nr,+l nrl+2 2 * - nr,+r2;

Considerthe polynomialpT(X) =(A c,T)miL. (A tT) m,. We wish to

prove that IT (A) is the minimum polynomial over pT satisfied by as. We
note first that (aSi -ST8)?n1 . . . (aSi - tTi) mt contains the factor
(aSi _38T.)ni 0. Since (K, 5, T) is the least common cover of the
(Ki, Si, Ti) this implies that pTQ(aS) 0. Now let vT(X) be the minimum
-

s r-1 areIealin
polynomial of a over T,a
Pr. We recall that if 1, , e linearly in-
dependent over 1, then 1, aS,* , (aS) r-1 are linearly independent over pT.
This implies that vT(A) has coeffiients in I - s. Hence v(A) is the mini-
mum polynomial of a over (D. Since vTQxS) = 0, vTt(c(St) - . This implies
that vTi(A) is divisible by (A /Ti)ni and hence that vT(A) is divisible by
-

(x /3PT)ni. It follows that vT(A) is divisible by yT(A) and since vT(A) is

minimal, /T (AS) - VT(A). From the factorization T (A) jH(A _ T) mj we
obtain !L(A) = H (A -j) "Ii. Thus the minimum polynomial of a over (D fac-
tors into linear factors in P [A] and so P is normal over (.
Let (L, U, Vf/)be any non-sinigularone dimensional composite of P. Then
- pv. Since (L: PE) - 1 also, L = pu. If we apply the isomorphism
V'1 to L, we see that (L, U, V) is equivalent to the composite (P, A, 1) where
A - UV' is an automorphism in P and 1 is the identity automorphism; Con-
versely if A is any automorphism, (P, A, 1.) is a non-singular composite. It is
clear that (P, A, 1) (P, 1, A-'). Hence the inverse of (P, A, 1) is
=

(P, A-', 1) = (P, 1, A). It follows directly from the definition of the product
that (P, A, 1) X (P, B, 1) = (P, A, 1) X (P, 1, B-1) - (P, A, B-1)
c (P, AB, 1). Thus the totality of non-singular one-dimensional composites
of P is a group under multiplication isomorphic to the group of auto-
morphisms of P.
Suppose now that H7is a finite hypergroup of one dimensional composites.
T1"henit is clear that IH is a group and by Theorems 21 and 22, P is finite,
separable and normal over (DH. Conversely if P is finite, separable and normal
over D, then the indecomposable components of the Galois composite r such
that (Dr 1 form a group II under multiplication. If (P, A, 1) is one of
=

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
28 N. JACOBSON.

the simple components of r then A is an automorphism of P that leaves the

elemenltsof I] fixed. On the other hand if A is such an automorphism of P,
(P, A, 1) is a composite leaving the elements of 1 fixed. Hence (P, A, 1)
is covered by r and therefore (P, A, 1) is one of the indecomposable com-
ponents of P. This completes the proof of the classical correspondence:

THEOREM 23. If P is an arbitrary field, there is a (1 - 1) correspondelce

betwee?nthe finite gr-oups H of automorp-hismsin P and the subfields 1H over
which P is finite, separable and normal, namely, iH is the set of fixed elements
under the autornorphisms of HI and H is the complete set of automorphisms
leaving the elements of -J fixed.

We recall that (K: pT) = E(Ki: pTt). If P is separable and normal

over pr, each (K: pTt) -1 and hence (P: 4Jr) (K: PT) is the number
-

of components (Ki,sj, Tj). It follows that (P: Ir) is the order of the
Galois group of P over 1rp

14. Simple extensions. Suppose that P 1 (a) a simple extension of

=

f and let 11(A) be the minimum polynomial of a over (. Let r (K, S, T) =

be the Galois composite whose field of fixed elements (Dr ?(. Then the
miinimum polynomial of as over pT is /,J(A) and the degree of /T(k) is the
dimensionality q of K over pT. If A = (L, U, V) is any composite such
that c1P ?-, r is a cover of A and hence L is generated by aU over pv. Thus
if vV(A) is the minimum polynomial of aU over pv and r is the degree of
vV(A) then the elements I, au,. .
(.U)y-1 form a basis for L over pv. Since
p2vQ(U) - 0, vV(X) is a factor of /v((A) and hence the polynomial v(X) in
P[A] is a factor of 11(A). Thus we have associated with the composite A the
factor v(X) of Iu(A) in P[A]. The composite z, ? z2 if and only if the
associated factor v, (A) is divisible by V9(X). We shall show next that every
factor v(X) of /(k) in P[A] arises in this way from some composite. For if
pu(A)-v(A)v (A) and B is the ideal in K generated by vT(S) then
(K - B, S, P) is the required composite. The correspondence between com-
posites A such that (1 ? 4) and the factors v(A) of /(A) in P[A] is therefore
(1 - 1). It is readily seen that the indecomposable components of r corre-
'spond to the prime power factors 7ri(A)el of tt (A) = 7r1(A)el. 7rs(A)es i
P[A] A

Let v (A) XA1kn- #Anz-1 * P.m be a factor of / (A) and let

(L, U, V) be the corresponldingcomposite. Then if we turn L into a double
P-module by defining ax = ux - xcu and xx = cvx =xcv, it may be verified
that the matrix of aU relative to the basis I, aU, * , (aU)m-i is

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions
 AN EXTENSION OF GALOIS THEORY TO NON-NORMAL FIELDS. 29

(22) L .

If 7r(k) = n- 1n-1 is a second factor of 11(k), it determines

* -

in the same way a composite and a self-represeiltation in which a is represented

by the matrix
v 82

The elements ,Pi are polynlomials in a with coefficients in I). If we replace

the Pi3in aF by the corresponding polynomials in aG, we obtain the matrix
2F'XG representing a in the product representation. If p (X) is the minimum
polynomial of aFXG, p (X) is a factor of / (X) and the composite associated
with p (X) is the product of the composites associated with v(A) and the com-
posite associated with 7r(k).
If we apply the theory of a single linear transformation we see that any
self-representation a o- aE is decomposable into " cyclic " self-represenltations
Ei where aEl has the form (22).
THE JOHNS HoPxIi s UNIVERSITY.

This content downloaded from 128.235.251.160 on Sun, 7 Dec 2014 21:32:51 PM

All use subject to JSTOR Terms and Conditions