the levitzki radical in varieties of algebras

8
Math. Ann. 194,27--34 (1971) @ by Springer-Verlag1971 The Levitzki Radical in Varieties of Algebras*'** TIM ANDERSON Introduction Recently Zhevlakov [4] has shown that for Jordan rings free of 2-torsion, the property of being locally nilpotent is a radical property (in the sense of Kurosh-Amitsur [1]), while Tsai [3] has added an analogue of the Babi6 Theorem for these rings. These results on the Levitzki radical are also valid for alternative rings. Indeed, more generally, Zwier [5] has shown that if the defining identities for a variety of rings obey certain combinatorial laws, common to the Jordan and alternative cases, then there is a satisfactory theory of the Levitzki radical in the variety. In this paper we consider the general problem: What are necessary and sufficient conditions for the existence of the Levitzki radical in a variety of rings? We give a partial solution to this problem by formulating a condition on the universal enveloping tings defined for a variety. As in the case of Zwier's work, the theory of the Levitzki radical for alternative rings follows as a special case of a more general result. However, our approach differs from Zwier's in that we study the defining identities of a variety from an implicit point of view. 1. Preliminary Definitions and Discussion Throughout this paper, Q will be a fixed but arbitrary commutative associative ring with unity and the term "algebra" will always mean "Q-algebra". We recall that a Q-algebra is a non-associative ring A which is also a unital left Q-module for the ring Q and such that ct(xy) = (~x)y = x(Qty) for all x, y ~ A and ~tE ~. In particular, if Q is taken to be the ring of integers, then any ring is a Q-algebra. Thus, our results, while stated for Q-algebras, will be equally valid for rings. The term "ideal" will always mean "two-sided ideal" and we shall write I< A to denote the fact that I is an ideal of A. Finally, whenever we use the expression "linear", we shall mean "~-linear". Let A n denote the linear span of all products of n elements of an algebra A. The algebra A is said to be nilpotent if An= 0 for some n. More generally, A is called locally nilpotent if every finitely generated subalgebra of A is nilpotent. It is convenient to describe the nilpotence of an algebra A in terms of its right and left multiplications R a and Lo respectively, where R a :x-~xa and * This research was partially supported by the National Research Council of Canada. ** The author gratefully acknowledges the use of facilities so kindly extended him by the Universityof Barcelonaand the Universityof Birmingham as this paper was beingwritten.

Upload: tim-anderson

Post on 10-Jul-2016

219 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: The Levitzki radical in varieties of algebras

Math. Ann. 194, 27--34 (1971) @ by Springer-Verlag 1971

The Levitzki Radical in Varieties of Algebras*'** TIM ANDERSON

Introduction

Recently Zhevlakov [4] has shown that for Jordan rings free of 2-torsion, the property of being locally nilpotent is a radical property (in the sense of Kurosh-Amitsur [1]), while Tsai [3] has added an analogue of the Babi6 Theorem for these rings. These results on the Levitzki radical are also valid for alternative rings. Indeed, more generally, Zwier [5] has shown that if the defining identities for a variety of rings obey certain combinatorial laws, common to the Jordan and alternative cases, then there is a satisfactory theory of the Levitzki radical in the variety.

In this paper we consider the general problem: What are necessary and sufficient conditions for the existence of the Levitzki radical in a variety of rings?

We give a partial solution to this problem by formulating a condition on the universal enveloping tings defined for a variety. As in the case of Zwier's work, the theory of the Levitzki radical for alternative rings follows as a special case of a more general result. However, our approach differs from Zwier's in that we study the defining identities of a variety from an implicit point of view.

1. Preliminary Definitions and Discussion

Throughout this paper, Q will be a fixed but arbitrary commutative associative ring with unity and the term "algebra" will always mean "Q-algebra". We recall that a Q-algebra is a non-associative ring A which is also a unital left Q-module for the ring Q and such that ct(xy) = (~x)y = x(Qty) for all x, y ~ A and ~t E ~. In particular, if Q is taken to be the ring of integers, then any ring is a Q-algebra. Thus, our results, while stated for Q-algebras, will be equally valid for rings. The term "ideal" will always mean "two-sided ideal" and we shall write I < A to denote the fact that I is an ideal of A. Finally, whenever we use the expression "linear", we shall mean "~-linear".

Let A n denote the linear span of all products of n elements of an algebra A. The algebra A is said to be nilpotent if An= 0 for some n. More generally, A is called locally nilpotent if every finitely generated subalgebra of A is nilpotent.

It is convenient to describe the nilpotence of an algebra A in terms of its right and left multiplications R a and Lo respectively, where R a : x - ~ x a and

* This research was partially supported by the National Research Council of Canada. ** The author gratefully acknowledges the use of facilities so kindly extended him by the

University of Barcelona and the University of Birmingham as this paper was being written.

Page 2: The Levitzki radical in varieties of algebras

28 T. Anderson:

L , , : x - , a x for a, x E A. By induction it may be shown that any product of 2 r dements of A can be written in the form (x) Ya~ ... Y~, for some x, at . . . . . ar~ A, where Y denotes ambiguously R or L. Thus A is nilpotent if and only if there exists an integer r such that Y,l --" Y ,̀~ = 0 for all al . . . . . a, e A.

A weaker notion than nilpotence is that of solvability. For an algebra A we define inductively A ~") by A t°) = A and A ~") = (A ~"- 1))2 for n > 0. While A ~1) is an ideal of A, in general A t") is only a subalgebra of A. If A t") = 0 for some integer n then A is said to be solvable. It is easy to see that A 2r ____ A ~'), so that nilpotent algebras are always solvable.

A technical difficulty which will arise in our work lies in the fact that a variety may be described by several sets of identities. For this reason we need to recall some basic facts about identities.

Let X = {xt, ..., x, . . . . } be a countably infinite set and #{X} be the free non-associative algebra based on the set X. The algebra q~{X} may be charac- terized by the property that any map 0 : X - , A , where A is an algebra, can be uniquely extended to a homomorphism if: #{X} - ,A. An algebra A is said to satisfy the identity p ifp e #{X} and for every homomorphism r/: #{X} - ,A, (p)q =0. In view of the above mapping property for #{X}, this is equivalent to saying that p(at . . . . , a , )= 0 for all a t , . . . , a, e A if p = p ( x t . . . . . x,). Given a subset I of #{X}, the variety C(I ) defined by I consists of those algebras which satisfy all the identities in 1.

An ideal of ~{X} which is invariant under all the endomorphisms of the algebra #{X} is called a T-ideal. Evidently, if I = #{X} and I ' is the T-ideal of # {X} generated by I then ~¢~'(I) = ~(I ' ) . The importance of expanding I to I' lies in the following fact: Ifp is an identity for each A e C(I ) then p e I'. To see this, it is sufficient to note that # { X } / I ' is contained in V(I).

Most studies of algebras take place within varieties defined by homogeneous identities, and in this paper it will always be assumed that the defining identities are homogeneous m all variables. Moreover, in order to make certain degree arguments later, we shall have to require that the identities in the T-ideal I' generated by I are essentially homogeneous as well. Therefore, we now make the additional assumption that the linearizations of the identities of I are contained in I'. Then the structure of ¢~{X} as a graded algebra (with respect to the degree function) carries over to the algebra # { X } / I ' , and we have the following result: If p(x l . . . . . x ,) ~ I ' and p = Y, pj where pj is homogeneous of degree j in xi then pje I ' for allj.

I fA ~ ~ ( I ) and M is a #-space which admits bilinear compositions ma and am(EM) for a e A , rn~ M then we can form the split null extension A @ M by defining (a + m) (b + n) = ab + an + mb for a, b e A, m, n e M, If the algebra A q ~ M is contained in C( I ) then M is called a bimodule (more precisely, an I-bimodule) for A. The notion of bimodule is equivalent to that of representa- tion. That is, given a #-space M and a pair of linear maps a - , S`,, a---, T~ of A into Hom(M,M), bilinear compositions ma and am can be defined by ma = (ra)So and am = (m) T,, for m e M, a e A. If, then, M is a bimodule, the pair of maps a- ,S , , , a - , T`, is called a representation of A in Horn(M, M) (in fact,

Page 3: The Levitzki radical in varieties of algebras

The Levitzki Radical in Varieties of Algebras 29

we may speak of a representation of A in any associative algebra X since in any case, X is isomorphic to a subalgebra of Hom(X, X)).

For A E Y(I) we may form ll(A), the universal enveloping algebra for the representations of A. The associative algebra R(A) has the property that there is a (canonical) representation a--, S*, a--, To* of A in II(A) such that for any representation a ~ S , , a - , Ta of A in an associative algebra X there exists a unique homomorphism ~p : ~(A)---,X such that S = S*~p and T = T*~. This property will be referred to as the universal mapping property for II(A).

Another property we shall need is that II(A) is generated by the elements S*, To*, a e A. To verify this, as well as the existence and uniqueness of R(A), we refer the reader to Jacobson [2]. However, we caution the reader that our definition of ~I(A) differs slightly from Jacobson's in that our II(A) does not contain a unity element.

2. Main Results

(2.1) Lemma. Suppose the variety 3e'(I) has the following property: (J[) Every solvable B ~ ~ ( I ) is locally nilpotent. Then the free algebra A in ~ ( I ) generated by the set {y, xl . . . . . x,} has the

Jbllowing property: (2.2) For each integer k >= 1 there is an integer f (k ) such that any product

of the form (y) Y,~... Y,,k~, where {a~ . . . . . aftk) } C= {X 1 . . . . . X,} and Y = R or L, can be written as

(Y) Y,~ ... Y~,,,,, = Y' ~(Y) Y~,,, ... Yc,,,,,,,

where the ~i ~ ~, the cj, i' s are monomials in x 1 . . . . . x,, and for each i, at least one of cl.i, ..., Cm,,i is of degree at least 2 k-1 in the xfs.

Proof. Consider the algebra A/Tk, where T k is the ideal of A generated by a (k). As T k D__ A (k), A/Tk is solvable and finitely generated; hence A/Tk is nilpotent by condition (J/). Thus, there is an integer f (k ) such that (a)Ya,... Ya,k, ~ Tk for all a, al, . . . , a fiR ) ~ A, where Y= R or L, Evidently then, (a)Ya~ ..: Ya,k~ -- Z u'iv' i + Z (uivi)P i, where u' i, v'i, ui, vi are in A tk-1) and the P{s are products of Yb~, 's, where the elements bj, i are in A and Y = R or L. In this relation put a = y and choose {al . . . . . aftk) } C {X~ . . . . . X.}. Then the left hand side of the resulting relation is of degree one in y, hence so too is the right hand side. Using now the fact that A(k-I)C-_A 2k-i, w e obtain (2.2).

(2.3) Theorem. Let ~/'(I) be a variety satisfying the condition (Jg). Then U(B) is nilpotent for each finitely generated solvable algebra B in ~IP(I).

Proof 1. Since B is finitely generated and solvable, B 2 k - ~ - 0 for some k, because of (Jr'); hence there exist b~ . . . . . b, e B such that every element of B is a linear combination of b 1 . . . . , b.. The algebra II(B) is generated by the elements S*,, T{,*, where b---,S'~, b---, T* is the canonical representation of B in

The author is thankful to the referee for suggesting improvements in the following proof.

Page 4: The Levitzki radical in varieties of algebras

30 T. Anderson

II(B). For the bimodule M belonging to the canonical representation, the split null extension B ~ M is in ~( / ) .

From a fundamental property of free algebras [2, p. 26], the relations (2.2) are identities for all the algebras in ~r'(I), and in particular, for B @ M. Choosing, in (2.2), y E M and {xl, ..., x,} = {bl . . . . . b,}, and using the fact that B 2~-1 =0, we obtain the result that Yo, ... Yo:~k~ = 0 for {a 1 . . . . . a:~k)} ~_ {bl, . . . , b,}, where Y = S* or T*. Therefore II(B) :~k) = 0 and II(B) is nilpotent.

A variety ~ ( I ) for which there exists an integer s ~ 2 such that A ~ ~ ( I ) and J < A imply J ' < A has been called an s-variety by Zwier I-5]. Let us note that for a 2-variety ~(I ) , A ~ ~ ( I ) and J < A imply j k < A for all k ~ 2. To see this, consider the free algebra F in 4/:(1) generated by {xl, x2, x3} and let P be the ideal o f F generated by x 1 and x 2. Then x~x 2 ~p2 and (Xxx2)x3 E p2 because C(I) is a 2-variety. Then from the homogeneity conditions on I, it follows that (xl x2)x3 = f ( x l , x2, x3), where f ( x l , x2, x3) is a linear combina- tion of the terms (x3)Ix, Yx~ and (x3)Y~ Y~,, where Y = R or L. Similarly, x3(x lx2)=g(xt , x2,x3), where g(xl ,x2,x3) is composed of the same terms above. Then using these two identities and induction, it can be shown that J < A ~ ~ ( I ) implies j k < A for all k ~ 2.

We now consider the following condition for 2-varieties: (jlr) H(A) is nilpotent for each finitely generated trivial A ~ C(I).

(2.4) Lemma. Let ~v'(I) be a 2-variety satisfying (jtr). Suppose A is a finitely generated algebra in 3v'(I). Then for each integer k > 1 there exists an integer re(k) such that ((A2) k) Eat.. . Yo,,~ ~- (A2) k+ 1 for all a l , . . . , am(k) ~ A, where Y is ambiguously L or R.

Proof. The algebra A/A 2 is trivial and finitely generated; hence II(A/A 2) is nilpotent by (,4:). Because ~V(I) is a 2-variety, (A2) k and (A2) k+l are ideals of A. Moreover, (A2) k+ 1 contains (A2) k . A 2 and A 2 . (A2)~. Hence (A2)k/(A2) k+ 1 is a bimodule for the algebra A/A 2 and we may represent A/A 2 in the algebra of linear transformations of the space (A2~'/(A2) k÷ 1.

Specifically, for a e A, we let Sa2 +o : (A2) k+ 1 + x ~ (A2) k+ 1 + xR , and TA~+o : (A2) k+l + x ~ ( A 2 ) k+l + xLo, where x e (A2) k. Then the pair A 2 +a~Sa2+o, A2+a~Ta~+o is the representation of A/A 2 provided by the bimodule (A 2)k/(A 2) k+ 1.

Using the universal mapping property for II(A/A2), we conclude that there is a homomorphism ~p:I I (A/A2)~X such that S = S * w and T = T*~p where S* and T* denote the canonical representation of A/A 2 in II(A/A 2) and X is the associative algebra generated by S,t2+a, Ta~+,, a t A. Therefore, tp maps II(A/A 2) onto X and X is nilpotent. This implies there is an integer m(k) such that ((A2)~)Yal ... Y~,,~k,_~(A2~ +1 for all al, ...,am(k)~A , where Y = R or L.

The idea of the next lemma is due to Zhevlakov [4].

(2.5) l.emma. Suppose for each finitely generated algebra A ~ ~v'(I) there exists an integer m such that A s ~_ A 2 A 2. Then if B is a finitely generated algebra in q:(1), so is B ~) finitely generated for all n.

Page 5: The Levitzki radical in varieties of algebras

The Levitzki Radical in Varieties of Algebras 31

Proof. Evidently it is sufficient to prove B 2 is finitely generated, and without loss of generality we may assume B to be the free algebra in ~ ( I ) generated by x 1 . . . . , Xr

Let Z be the set of all monomials of degree r in the xi's, where 2 < r < m and m is an integer such that B m ~_ B 2 B 2. We claim that Z generates B 2, and this will prove the lemma since Z is a finite set.

Suppose y is a monomial of degree greater than two which is not in the subalgebra generated by Z. Necessarily, y has degree at least m, whence y e B 2 B 2. Therefore y = Y.,~tuivi, where the ~i's are in ~ and the ui's and vi's are monomials of degree at least two. Since B is a free algebra and the defining identities of I are homogeneous, degy = degu~ +degv~ for all i. However, by induction, we may assume that the u~'s and v~'s are in the subalgebra generated by Z, whence so too is y in that subalgebra. Thus the lemma is proved.

(2.6) Lemma. Let ~ ( I ) be a 2-variety satisfying (.At). Then (1) each finitely generated solvable algebra in ~v'(I) is nilpotent, and (2) if B is a finitely generated algebra in 4/'(I) so is B ~") finitely generated for all n.

Proof. Let A be a finitely generated algebra in ~e~(I). Applying Lemma 2.4 we can choose an integer re(l) such that (A2)ya, ... Ynm~,~A2A 2 for all a 1 . . . . . amtl~eA, where Y = R or L. Thus for all x, a o , a 1, ..., amtl) e A , (X) YaoYal . . . Yam~I)CA2A2, where Y = R or L. If we put m = 2 m(1)+1 then we have A"C=A2A 2. l~ow Lemma 2.5 gives the result (2).

To prove (1) we use induction on the integer t for which A t') = 0 for a solvable algebra A, the result being trivial for t = 0 or 1. Suppose A is an arbitrary finitely generated solvable algebra in "//'(I). Then from (2) A 2 is also finitely generated and solvable, and applying the induction hypothesis to A 2 we see that (A2) k÷l = 0 for some k. Put r = r e ( l )+ . . . + re(k), where the m(i)'s are chosen as in Lemma 2.4. Then for a~ . . . . . a r ~ A, we have by iteration of Lemma 2.4 that (A2) yal ... Ya~_(A2)k+l=O. Hence for all x ,a o . . . . , a reA , (x) Y,o ... Y~ = 0 and A is nilpotent.

(2.7) Theorem. Let ~e~(I) be a 2-variety. Then the following statements are equivalent:

(i) Local nilpotence is a radical property in "It(I). (ii) Every solvable B ~ q/~(I) is locally nilpotent.

(iii) U(B) is nilpotent for every finitely generated solvable B ~ "t/'(I). (iv) ~(B) is nilpotent for every finitely generated trivial B ~ q/'(I). Moreover, when local nilpotence is a radical property in ~l/'(I), every semi-

simple algebra in "//'(I) is a subdirect sum of prime semi-simple algebras in ~(1) (where "prime" is defined as for associative rings).

Proof. Even when "//(I) is not a 2-variety, if local nilpotence is a radical property in q/'(I) every solvable B e q/'(I) must be locally nilpotent. Indeed this follows easily by induction on the integer t for which B tt~ = 0 and the fact that B/B 2 and B 2 locally nilpotent imply B locally nilpotent, when local nilpotence is a radical property. Thus, (i) implies (ii), and (ii) implies (iii) because of Theo- rem 2.3. Trivially, (iii) implies (iv). On the other hand, (iv) implies Lemma 2.6,

Page 6: The Levitzki radical in varieties of algebras

32 T. Anderson:

and having that lemma, one can simply prove (i) as in the case of the well- known proof for associative rings. See, for example, [1] or [4].

The proof of the remaining part of the theorem, which is the Babi6 Theorem, can again be proved in exactly the same way it is proved for associative rings. See [3].

(2.8) CorolLary. Local nilpotence is a radical property in the variety of alternative algebras.

Proof. It is easy to see that alternative algebras form a 2-variety and that the defining identities have the homogeneity properties of Section 1. Thus, we only have to show they satisfy the condition (Jff).

For an alternative algebra A we have the following identities for the right and left multiplications: LaLQ=La2, RQRa=Ra~, LaLb + LbLa=Lab+ba, RaR b -1- RbR a = Rab+b a, LaR b -- RbL,~ + RaR b - Rob = O, and LaR b + Lab -- RbL a - L b L , , = O for a , b ~ A .

Suppose further A is generated by a I . . . . , a n and that A 2 =0. Then ll(A) is generated by the 2n elements Yo, where Y= S* or T* and a ~ S * , a ~ Ta* is the canonical representation of A in ll(A).

Since A 2 = 0, we have the relations (1) S~ISQ,* * -- T.*a, T,*,, = 0 for all i, and the commutivity relations (2) * * - * * * * - T* for i , j , Sa, S a j - -SajS~,, T~, T ~ j - - T~ ,, (3) T'~,Saj-SajT~,* * * * +Sa, * for all i,j, and (4) T~,* Saj*- Saj* T2,* - T~ T'a, = o for all i,j.

Consider any product P of 2n + I of the generators So*, T~*. It will be of the form e = Yal ... Yu2.÷,, where {ul, . . . , U2n+l} ~ {al . . . . . an} and Y = S* or T*. At least n + 1 of the Y's must be S* or T*. Assume, for example, at least n + 1 of the Y's are S*. Then P has the form

P . . . . s : ,

. . . * * . . . . for some j. Using the relations (2) and (3), we may assume P = SajSaj But then P = 0 because of (1). Equally, if at least n + 1 of the Y's are T*, we may show P = 0 using (2), (4), and (1). Thus II(A) 2"+' =0.

3. Open Questions

A disagreeable feature of Theorem 2.7 is the presence of the assumption that ~f(I) is a 2-variety. For example, the theorem does not apply to the important case of Jordan algebras because they form a 3-variety. However, we believe similar results should be valid for arbitrary varieties - at least for s-varieties.

In proving the existence of the Levitzki radical in ~/r(I), what has to be shown is that the notions of solvability and nilpotence coincide for the finitely generated algebras in ~(I) . In this context it should be noted that the converse of Theorem 2.3 is true. Indeed, if II(B) is nilpotent for all finitely generated solvable B e ~ ( I ) then in particular the regular representation of B is nilpotent,

Page 7: The Levitzki radical in varieties of algebras

The Levitzki Radical in Varieties of Algebras 33

whence B itself is nilpotent. However, in practice it would be very difficult to verify this condition on the universal enveloping algebras of a given variety.

It may be advisable to study the lower radical determined by the trivial algebras in ~ ( I ) in order to arrive at other, more suitable, conditions for the existence of the Levitzki radical in ~( I ) . That is, we put ~t'o = {A ~ ~ ( I ) IA 2 = 0} and define inductively ,/ / , to be the class of all algebras A e i//(1) with the property that each non-zero homomorphic image of A contains a non-zero ideal in J / , _ 1 for 0 < n < ~ . It is easy to see that if A ~") = 0 then A e J t , . Thus if it can be shown that the algebras in Jr', are locally nilpotent for 0 < n < oo, then the Levitzki radical will exist in ~( I ) .

We conclude this paper with a condition which guarantees the Boer radical algebras, that is, the algebras in ~//1, to be locally nilpotent.

(3.1) Lemma. I f A is Boer radical so is any subalgebra B of A Boer radical.

Proof. If B/W is a non-zero homomorphic image of B then we can select an ideal K of A which is maximal with respect to the property that K n B c= 14I. Evidently, K4:A, whence A/K contains a non-zero ideal P/K such that (P/K)2=O. If we put C = W + P n B then B/W will contain the non-zero trivial ideal C/W. Thus B is Boer radical.

(3.2) Lemma. I f B is finitely generated but not nilpotent then B contains an ideal W such that B/W is not nilpotent but B/C is nilpotent for any ideal C of B properly containing 141.

Proof. Let X = {J I J < B and B/J is not nitpotent}, noting that ~t" is non- empty as 0 e X. Let ,/1 __c J2 =c... be a chain in ~ and put J = U J,- We claim

n = l

that J e X. If not, then (B/J) k = 0 for some k. Suppose B is generated by xl . . . . , x~. Then J contains all monomials of degree at least k in the xi's and in particular all monomials of degree s, where k-<.< s < 2k. As there are finitely many such monomials, it follows that some Jm contains all the monomials of degree s, k < s < 2k. In fact, we may show by induction that Jm contains all the monomials of degree at least k. Indeed, if y is a monomial of degree > 2k then as degy > 2, y = ab for some monomials a and b. At least one of a or b has degree at least k, hence is in Jm by the induction hypothesis. As Jm is an ideal, so then is y in Jm" But now we have shown that (B/Jm) k = 0, contrary to the fact that J , is in £r. Therefore, J e 5f.

Having shown that the set ~r is closed when taking chains, we may now use Zorn's lemma to assert that ~r contains a maximal element W, which clearly satisfies the demands of the lemma.

(3.3) Theorem. Let ~( I ) be a variety satisfying the condition (•) ll(A) is nilpotent for each finitely generated nilpotent A e ~lf(I). Then

the Boer radical algebras in ~e~(l) are locally nilpotent.

Proof. Let A be Boer radical and B be a finitely generated subalgebra of A. Assume that B is not nilpotent. According to the previous lemrna, B contains an ideal W which is maximal with respect to the property that B/W is not 3 Math. Ann. 194

Page 8: The Levitzki radical in varieties of algebras

34 T. Anderson: The Levitzki Radical in Varieties of Algebras

nilpotent. Certainly, B / W is non-zero, hence from Lemma 3.1, B / W contains a non-zero ideal C / W such that (C/W) 2 =0. From the choice of IV, B/C is nilpotent, as well as finitely generated. Thus II(B/C) is nilpotent. Furthermore, as C and W are ideals of B and C 2 ~_ IV, C / W is a bimodule for the algebra B/C, and as in Lemma 2.4, we may represent B/C in the algebra of linear trans- formations of the space C/W. Using the nilpotence of II(B/C) and the universal mapping property, we conclude there is an integer m such that (C) Ya~'" Yam C= W for all a 1 . . . . . a m ~ B, where Y = R or L. Moreover, as B/C is nitpotent, there is an integer k such that (B) Ybl ... Ybk ~- C for all bl, ..., b k e B. Altogether, for any c I . . . . . Ck+ m in B, (B) Ycl ... Yck+,, = ((B) Yc~ ... Y J Yc~+x "" Yc~+,,

(C) Yck÷~ .-. Y~k÷,,-~ W. Thus, B / W is nilpotent, contrary to the choice of IV. The algebra A was therefore locally nilpotent.

Remark. We note that the condition (~) is really a statement about the structure of the finite dimensional algebras in ~ ( I ) since any finitely generated nilpotent algebra has a finite basis.

References

1. Divinsky, N. : Rings and radicals. Toronto: University of Toronto Press, 1965. 2. Jacobson, N.: Structure and representations of Jordan algebras. Amer. Math. Soc. Colloq.

Publ. 34, Providence (t969). 3. Tsai, C.: Levitzki radical for Jordan rings. Proc. Amer. Math. Soc. 24, 119--123 (1970). 4. Zhevlakov: Solvability and nilpotence of Jordan rings. Algebra i Logika Sem. 5, 37--58 (1966). 5. Zwier, P. : Prime ideals in a class of narings, to appear.

Dr. T. Anderson Department of Mathematics University of British Columbia Vancouver, 8, B.C., Canada

(Received February 15, 1971)