The dimension subgroup problem

The dimension subgroup problem

JOURNAL OF ALGEBRA 21, 216-23 1 ( 1972) The Dimension Subgroup SANDLING ROBERT Department of Matb.ematics, The University, Communicated Recei...

881KB Sizes 48 Downloads 97 Views

JOURNAL

OF ALGEBRA

21, 216-23

1 ( 1972)

The Dimension

Subgroup SANDLING

ROBERT Department

of Matb.ematics,

The University,

Communicated Received

Problem

Manchester,

U.K.

MI3

9PL

by D. Rees

August

4, 1970

Whether the terms of the lower central series of a group are associated with the powers of the augmentation ideal of its integral group ring is a question which has been open for over thirty years. This question, the dimension subgroup problem, has achieved notoriety for the number of false proofs it has elicited. In this paper, we record general observations on the problem, verify the conjecture for several new classes of groups and develop generalizations which indicate new lines of research. In a subsequent paper [20], we examine the subgroups attached to group rings over arbitrary coefficient domains. The general properties of dimension subgroups are set forth in the first section. Several equivalent formulations of the conjecture are given. The subgroups are calculated in a number of special cases. A minimal counterexample is a finite p-group with cyclic center. In the second section, we take the approach that such reductions make available. The conjecture is established for Abelian-by-cyclic groups and for split extensions of Abelian groups by groups satisfying the conjecture; it follows inductively that the Sylow subgroups of the symmetric groups satisfy the problem and so every p-group embeds in a p-group for which the conjecture is verified. This analysis provides an unexpected bridge to the work of Passi who conjectured a strong result about the properties of factor sets of extensions of divisible groups byp-groups and showed how it would establish the dimension subgroup conjecture. Here another way of using Passi’s conjecture to solve the problem is introduced and applied in a minimal counterexample to obtain new results. The third section places the problem in a wider context: If G, is associated with I”, then G, is associated with even larger ideals. It may be that these ideals will have to be identified before the dimension subgroup problem can be resolved affirmatively. Some such ideals are discovered by expanding the methods of Passi and Moran. The results obtained are used to establish the 216 0

1972 by Academic

Press,

Inc.

THE

DIMENSION

SUBGROUP

PROBLEM

217

fact that the lower central series is associated with the Lie powers of the augmentation ideal for groups of class 5, for metabelian groups and for p-groups whose commutator subgroups have class strictly less than p. Several results concerning the intersection of the powers of the augmentation ideal are stated in the last section.

1. INTRODUCTION Let G be an arbitrary group and R a commutative ring with unit. Form the group ring RG as the free R-module on the set G with multiplication given as: Write an element y of RG as a sum C y(x)x, where Y(X) is in R and IV is in G; then the product $5 is defined by

summed over ally and x in G with yx = X. RG is functorial in R and G. The kernel of the augmentation RG to R induced by collapsing G to 1 is called the augmentation ideal and will be denoted as I or I(G) or I(R, G). Explicitly, I consists of all y in RG for which C y(s) = 0. If N is a normal subgroup of G, the projection G to G/N induces RG to RGjN whose kernel is the two-sided idealI(N)RG. Conversely, we may associate a subgroup of G with an ideal J of RG by considering all x in G for which x - 1 is in J; that is, the subgroup G f~ I + J. If J is two-sided, this subgroup is normal. T(N)RG is the smallest two-sided ideal J for which N = G n 1 + J. The problem of identifying the subgroup attached to an ideal is a difficult one. This chapter explores the problem for some familiar ideals. The subgroups attached to the powers of the augmentation ideal are known as the dimension subgroups; we denote them as i n = i,(G) = i,(R, G) = G n 1 + P(R, G).

i,(R, G) is also functorial of G. A ring homomorphism so that

in R and G so that it is a fully invariant subgroup R to S induces an inclusion i,:(R, G) < i,(S, G), GCZ, G) < WC

G)

for all R, where Z is the ring of integers. These facts, readily calculated, are apparent from the expression of i,(R, G) as the kernel of the natural monoid map G to RG/P defined by sending x to the coset of X. The descending series {in} in G is of the type studied intensively in Lazard’s thesis [12]. Call a descending series (H,} of subgroups of G a Lazard series if

218

SANDLING

LEMMA

Proof. [x, y] -

1.1.

The dimension subgroups form a Lazard series.

Take x in i, and y in i, . Then 1 = x-‘y-‘(xy

-- yx) = x-ly-l((x

-

l)(y

-

1) -

(y -

1)(x -

1))

which is in P+nz. As a corollary, we see that i, contains G, , the n-th term of the lower central series. Magnus [14] showed that, for G free, i,(Z, G) is G, . The conjecture that this remains true for all G is known as the dimension subgroup problem; it has had a stormy history [2; 4; 5; 131. Magnus’ result implies that the conjecture is equivalent to the proposition that the functor i,(Z, ) preserves epimorphisms. In this direction is the following lemma of Jennings [ 1 I] : LEMMA 1.2. If N is a ?wrmal subgroupof G contained in i,(R, G) and C?= G/N, then i,(R, e) = i,(R, G).

Proof. We already know i,(R, G) < i,(R, G) by functoriality. For the opposite inclusion, it suffices to observe that 1(N) RG is in I”(R, G) so that the isomorphism between Re and RG/I(N) RG induces one between I”(R, c) and In(R, G)/I(N) RG. Functoriality also shows that i,(R, ) respects (unlimited) direct products. The proof of this will be clear from the proof of the following related remark. LEMMA 1.3. If G = H&I splits over its normal subgroupM, i,(G) = &(H)(M n &(G)).

Proof. By functoriality, induces i,(G) ,( i,(H). This can be refined to PROPOSITION

centralixesM/M’,

i,(H) < i,(G) while the projection

G to G = H

If G = HA4 splits over its normal subgroupM and G then i,(G) = i,(H) i,(M).

1.4.

Proof. It suffices to show that ITi(M) = I”(G) n RM. We first establish that I”(G) = CifjCn Ii(H) P(M), where i, j > 0 and IO(X) = RX. The identity hm - 1 = (h - 1) + h(m - 1) shows this to be the case for n = 1. For n > 1, use induction and the fact that

I(M) I(H) C I(H) I(M) + 12(M),

THE

DIMENSION

a consequence of the hypothesis of the identity:

SUBGROUP

that ill’

219

PROBLEM

= [M, G~(soI([M,

G]) c I”(~~~)) and

1.5 (m -

l)(h -

1) = (k -

l)(ffz -

1) + h??+ -

= (12 -

l)(Wz -

1) + (h -

+ (h Lastly, in P-(U),

if y is in P(G) i, + j, = n. r(m)

l)(m -

I), X = [11z7hj

1)(X -

1) + (m -- 1)(X -

1)(X -- 1) + (X -

n RM, y =C

= C y(Ii??z) summing

&.

where

1)

1).

CQ is in I;k(Hj

and /z$; is

over 12 in H

But, if i, > 0, C a,(h) = 0, so we have y =x(& K with j, = n, which is in P(M).

a,(h)j ,& summed

over all

Observe that, for G as in 1.4, G,? = H,M,l so that this lemma verifies the dimension subgroup conjecture for G if it is verified for Hand M. An example of such a G is M extended by a group H of inner automorphisms. We have seen that i,(R, G) is the kernel of a monoid map G to RGjI”. If A is an R-algebra, 2l a two-sided ideal of A, C? = -2/W, then 1 + % forms a group of units in A; if 9 = RG and ‘3 = I, the above is a group homomorphism G to 1 + VI. LEhIhIA 1.6. i,(R, 1 + 5!I) = 1. Consequently, i,(R? G) is the iTltersectio:on of the kernels of all homomorphism G to 1 f PI, takefz over all R-algebras A and ideals 5%.

Proof. Let H = 1 + 21. Since H C B, A is an RH-algebra by right multiplication. But Af”(R, H) -= 0 since I”(R, H) is generated by products Y==JJ,~~<~(~ +~~~-i),ol~in\U, w h’lie,i --‘f-’aisin_~-,,~=3i~Il~~~n3il=0. If Gisin i,(R, H), then, in & 0 = l(1 + c: - 1) = Oc. We have seen that i,(R, G) is such a kernel. Let 1%’ be the kernel of a homomorphism G to 1 $- n_l for some 5%-__ and d. Let G = G/N. Then ___i,(R, G) < i,(R, G) but G is a subgroup of 1 + 91 so i,(R,

G) < i,#(R, 1 + ‘Xj = 1.

This gives another formulation of the dimension subgroup problem: Given a group G, there is a ring -4 and an ideal 5% with G/G., a subgroup of 1 -f %/?I”,. The free group case can be established in this manner:

220

SANDLING

THEOREM

1.7.

If G is a freegroup,

Proof. We may free associative ring by all elements of G/G, is a subgroup

LEMMA

G) = G, .

assume that G is finitely generated (see 2.1). Let A be the on the same number of elements and (2I the ideal generated weight > 1, in Hall’s notation [7]. Hall then shows that of 1 + ‘%/21n.

2. INTEGRAL

p-group

i,(Z,

COEFFICIENTS

2.1. A minimal counterexarrqde to i,%(Z, G) = G, G of class less than n whose center is cyclic.

is a jnite

Proof. Take x in i,(Z, G) but not G, ; s - I is a sum of products (x1 - 1) ..- (R, - l), v* i in G; letting H be the subgroup generated by the elements of G appearing in this sum, we see that x is in i&Z, H) but not H, . We may assume that G is finitely generated. Since i,(Z, G)/Gn < i,(Z, G/G,), we may assume that G, = 1; so class G is less than n. If &(Z, P) = 1 f or all finite p-groups P of class less than n, i,(Z, G) < N for every normal subgroup N whose quotient G/N is a finite p-group. But it is a theorem of Hirsch [17, p. SO] that the intersection of such subgroups is trivial. Hence, we may assume that G is a finite p-group of class less than 12. By minima&y, i,(Z, G) < N for every nontrivial normal subgroup N of G. But these subgroups intersect trivially unless G has cyclic center. This reduction shows that the dimension subgroup problem is essentially a problem about finite p-groups. As an example of the use of 2.1, we give a proof of LEMMA

2.2.

&(Z, G) = G’foy

all G.

Proof. We may assume G is Abelian. Since ZG is free Abelian, we can define a map ZG to G by sending x to x. Under this map, (x - l)(y - 1) goes to xyx-iy-1 = 1, so that I2 is in the kernel. Since x - 1 goes to X, no x - 1 can be in I2 unless x = 1. We shall generalize this method of proof to solve other cases of the problem. Many diverse methods have been used to establish the following cases: Free groups, Magnus [14]; groups of class 2, G. Higman [6, 181, Hoare [8]; p-groups of class 3, p > 2, Passi [IS], Moran [IS]; p-groups of class less than p, Moran [16]; p-groups of exponent p, Jennings [ll]. We establish the case of Abelian-by-cyclic groups and also verify the conjecture for a class of split extensions large enough to show that every p-group is contained in a p-group which satisfies the conjecture. A new and easy proof for the class 2 case appears in [20]. Let A be a normal Abelian subgroup of an arbitrary group G and let T be

THE

DIMENSION

SUBGROUP

221

PROBLEM

a set of left coset representatives. Define a map p from ZG to A by sending x = ta to 4. ,4 typical generator of I( is of the form (ta - l)(a’ - 1) which goes to I ; thus, I(G) l(Lq) is in the kernel. Working modulo this ideal, we see that I”(G) = P(T) + I([& (n - l)G]), where [A, mG] is inductively defined as [[rl, (m - l)G], G] and where I(T) denotes the subgroup of ZG generated by all t - 1, t in T, for: A typical generator of I”(G) is y = JJIGiG,(t,ai - 1). But tiai -

I = (ti -

1) + (ai -

1); so y = J&Gi>rt((ti -

1) + (aE -

so that y = n rw
2.3.

If

G is Abelian-by-cyclic,

i,(Z,

1)) 1.5

G) = G, for all n.

Proof. We may assume G,< = 1. Let /J be a normal Abelian subgroup with G = G/A cyclic; select t in G so that i is a generator of G and let T consist of powers of t. If a - 1 is in In(G), an analysis as above shows a - 1 = (t - l)or, 01in I(G) ( we assume n > 1). Passing to ZG, 0 = (i - 1)E so that ol is in the intersection of I(G) and its annihilator, which is trivial, so that E = 0, whence 01is in 1(/l) ZG. This shows that a - 1 is in ker p although, by definition, it is sent to a. Hence, a = 1. The foregoing analysis may be elaborated for metabelian groups but does not seem conclusive except for class 2 groups and a few others. The next application of the map p is to the case in which G splits over -4 ; in this case, T may be chosen as a subgroup of G so that I”(T) is contained in I(T) which is in the kernel of p. Then P(G) = I([& (n - 1jG]), so we have 2.4. If G splits over a normal Abelian subgroup G) = i,(Z, T)[A, (n - I)G], where T is a complement to A. If

THEOREM

i,(Z,

i,(Z,

A,

then

2’) = T, , i,,(Z, G) = G, .

Proof. By 1.3, it suffices to calculate ;1 n i,(Z, G) which contains (A, (12 - l)G] and, by the above remarks, is contained in [A, (72 - l)G]. Theorem 2.4 allows us to embed groups in groups which satisfy the dimension subgroup problem. This is in general a pointless endeavour since any group embeds in a simple group which satisfies the dimension subgroup problem by default. However, for p-groups, the concept has more substance as the dimension subgroup conjecture is equivalent to the proposition that every p-group embeds in a p-group of the same class which satisfies the conjecture. The wreath product structure of the Sylow subgroups of the symmetric groups [lo, p. 3791 sh ows them to satisfy the conjecture. Let p” be the

222

SANDLING

order of the smallest set on which the p-group G acts faithfully; equivalently, pa is the index of the largest core-free subgroup of G. Then G is a subgroup of the symmetric group of degree pd, which has class pd-l (see [21] for a new proof). Hence, LEMPL~ 2.5.

If n = pd-l, then i,n+l(Z, G) = 1.

Jennings’ result [ll] p rovides another bound on the extent to which the conjecture may fail. Let G,,, = G, JJ G$, the product taken over all i, i with ipi 3 k; let d& be the rank of G,,,/G,,,,, and n =(p - 1)x kd, . Then i,+,(Z, G) = 1. Kaloujnine’s embedding of a group G, given as an extension of a normal subgroup -4 by its quotient H, in the wreath product A - H [IO, p, 991 provides a more interesting application of 2.4. Taking A Abelian, we have by induction that H is contained in a group H* satisfying the conjecture. But G < A N H < A - H*, and A - H* satisfies the dimension subgroup problem by 2.4; so we have again suceeded in embedding ap-group in another which satisfies the conjecture. This embedding holds out another possibility. For G a minimal counterexample, we may take A to be cyclic and central and H to be of class less than class G. The usual embedding G ,( A - H depends upon the choice of factor set f from H x H to A; we may hope to solve the problem by choosing f so that G is contained in a split subgroup of class equal to class G. The largest such subgroup in which G could be so embedded can be calculated by viewing 4 N H as the group ring Z,,H, pe the order of A, extended by H in its right regular representation; it is the right annihilator of I”(Z,, ) H) extended by H. The proposition that f can be chosen to embed Gin this subgroup is, up to possible extension of G by central product over ,-2 (see 3.11), equivalent to Passi’s conjecture [IS] that any factor set H x H to Q/Z is cohomologous to another which, when extended by linearity to ZH x ZH to Q/Z, kills the subgroup fn(Z, H) x ZH, n = 1 + class H. Passi verified his conjecture for class H = 1 and for class H = 2, P > 2; he used this to prove cases of the dimension subgroup problem in a manner quite different from the above.

3. THE

LIE

POWERS

OF THE

AUGMENTATION

IDEAL

If A and B are subsets of ZG, define the Lie bracket (A, B) as the subgroup of ZG generated by all (01,P) = @ - t% LY.in A, /3 in B. Even if A and B are ideals of ZG, (A, B) need not be an ideal.

THE

If

LEnfraA 3.1‘

DIMENSION

SUBGROUP

223

PROBLEM

H and K aye subgroups of G, (I(H),I(K))ZG

= I([H,

K])ZG.

Proof. As right ideals, the left is generated by all (h (iz, k), h in H, k in K while the right by all [Iz, k] - 1. But [h, k] -

1, K -

1) =

1 = (h-l, k-l) AK.

Define the Lie powers of I as II = I and, inductively, I, = (In-r , I)ZG. Denote G n 1 + I, as i(,, = i&Z, G). That C, < it,, is a Corollary of 3.1. On the other hand, I, is contained in P so that i(,, < i,? . Thus the dimension subgroup conjecture implies the conjecture that G, = 5’~~) . To prove this, we may again assume that G is a finite p-group of class < n with cyclic center. The methods used in establishing cases of the dimension subgroup problem may be extended to solve this problem in roughly twice as many cases (because class G is roughly twice class G’). We shall establish the following cases of G, = i(,,(Z, Gj: THEOREM

3.2.

G arbitrary

THEOREM

3.3.

G p-group

and II < 6. with class G’ < p alad n arbitraq).

COROLLARY

3.4.

G p-group

and n < 2p.

COROLLARY

3.5.

G metabelian

and n arbitrary.

The iast two are immediate consequences of 3.3, although 3.5 can be obtained readily without 3.3. These cases provide some evidence for the dimension subgroup problem; their proofs provide a larger context in which to understand this problem. The conjecture is G, = G n 1 + In; we shall see that G, = G n 1 + J, for ideals J generally strictly greater than I”. The key to these results is a description of I, by sums of products of the augmentation ideals of the terms of the lower central series of G. This requires several preliminary observations. The first serves as a model and is of interest for the integral group ring probIem: LEMMA

(I(N)ZG,

3.6.

i’f N

and ill are normal

I(M)ZG)ZG

= I([N, t

Furthermore,

subgroups of 6,

M])ZG

I(N)I([M

[N, M] may be characterized G n 1 + (I(N)ZG,

+ I(N)I(M)I(G’)ZG G])ZG

+ I([N,

as I(M)ZG)

ZG.

G]) I(M)ZG

PYOC$ The left is generated I(M), x and y in G. But

as right

ideal by all (olx, py), a: in I(N),

/3 in

which is seen to be in the right by use of 3.1 and its corollary that I(H)I(K)ZG is in I(K)I(H)ZG + I([H, K])ZG. The opposite inclusion is established term by term: the first by 3.1; the second, being generated by all c@(x, y) since Ia = I(G) ZG by 3.1, will follow from the above equation once the last two terms are settled; the third follows from the equation LY(/~,S) = (01, Px) + (p x, LYX))X-~ and the fact that M is normal; the fourth by symmetry since, mod I([N, :‘ki’]) ZG, I([N, GJ) I(M)ZG may be replaced by I(M) I([N, GJ) ZG. To prove the last assertion, we may assume that [N, M] = 1. But then we can show that 1 = G n 1 + I(N)I(M)ZG which suffices. For this we may assume that G = NM by the following principle: LEMNA 3.7. If H is a subgroupof G and J a right ideal of ZIiT containedin I(H), thtil G n 1 + JZG = H n 1 $- J. To prove this, let T be a set of coset representatives for H in G; then JZG is the direct sum of all Jt, t in T. If G = AT&I, A = N n M is central and I(N) I(M) is already an ideal of ZG. It suffices to choose coset representatives for -4 so that the associated map p from ZG to A has I(N) I(M) in its kernel. Let 5’ be a set of representatives for A in N and T a set for A in M, both containing 1; then the set ST is a set of representatives for A in G. Taking sa in N and ta’ in M, we see that (~a - I)(ta’ - I) goesto aa’a-la’-’ = 1 under p. The next lemma is technical, its proof little more than that of 3.1: LEMMA

(I-I VA

Iv2 are normal subgroups of G, is W-N ZG containedin the sumof all pro&cts 3.8.

If

1 < i < h,

then

I1 Wi) I(P’,i >W> g WJ ~6’. / iti Proof.

Taking

(I-J (%l -

xi in Ni and y in M, we have

l), y -

1) = c Jj (“Yi j ici

I)&

-

l)( y -

1) j-J (Xi i>j

1).

We come now to the main result of this section: THEOREM 3.9. In = I(G,,)ZG -+- C nj I(G,JZG, n > rzj > 1, for which C (nj - I) = n - 1.

sum

over

all

.a1 ,

THE

DIMENSION

SUBGROUP

PROBLEM

225

Proof. We show that the right is contained in the left by showing that JJ I( G, ;) ZG is in IPi , where 1 fj~K,n~-ni~landC(nj-lI)=n--~; we allow By = 1 to simplify the induction, which is on k and then on nk . For k = I, this is I(G,,J ZG is in I, . For k > 1 and Q = 1, we conclude by induction on k, We assume that k > 1 and nk > 1. By 3.1, n Z(GTijj ZG is generated by all OL(~,X) where 01is in Is1 ij P is in (~~~jir,I(G,(),I)I(G,~-~) which, by 3.8, is in

which is in I, by induction on nL . To show that the left is in the right, we induct on n, the statement being clear for n = 2. Assuming equality for n - 1, it suffices to show that, for n-1 2nj> l,~(“j-l)=n-2,(njI(G,)ZG,I) is in I,. By the identity (LYX,yj == (~(2,y) + (01,X)J7, we seethat this Lie bracket is contained in n WJ J

I(G’) ZG -t (fl I(Gnjj, i) ZG.

The first product is in the right by definition, the secondby use of 3.8. I, admits other expressionswhich we state but shall not use. The first is clear from 3.9 while the secondrequires a little proof which we omit. COROLLARY

3.10.

sum oce7 all ni , n > nj > l,.for

which C (nj -

1) == n -- 1.

In = I(G,) ZG + x nItj(Gj)

ZG,

We can now prove the results 3.2 to 3.5. SinceI, = I(G’)ZC, Lemma 3.6 shows directIy that i(,,(Z, G) = G, for all G and n < 3. The other results follow indirectly by translating the question of I,4 in ZG to a question about ideals of ZG’, a process which leads to suggestive generalizations of the dimension subgroup conjecture. The following proof of the metabelian case will serve asmodel:

226

SANDLING

Proof of 3.5. Let G be metabelian. To prove i(,,(Z, G) = G, , we may assume G, = 1. We assume n > 2. By 3.7 and 3.9, it&Z, G) = G’ n 1 + J where J = 2 nj I(G,i,i)ZG’, sum over all n, , n > nj > 2, for which C (n; - 1) = n - 1. But 1z > 2; so J is contained in I*(G); so i(,,(Z,

G) < z,(Z, G’) = 1

since G’ is Abelian. This has the corollary that i&Z, G) = G, for all G and 71 < 4. To stretch this to the whole of 3.2, we again use 3.7 and 3.9 as above and observe that, upon substitution of G’ for H, it suffices to prove: THEOREM

3.11. For an arbitrary group H euitlzcenterZ(H), H3 = H n I + P(H) + I(H) I@(H)).

Proof. The usual reductions show that we may assume H to be a finite p-group of class 2 with cyclic center. Furthermore, we may assume that H can be expressed as an extension of its center Z(H) (write Z(H) = A) by an Abelian quotient E defined by a factor set f from f7 x n to A, where f is a bilinear map; since f is a factor set, bilinearity is equivalent to I”(Z, H^) x ZH being killed by f extended to a map ZR x ZR to A: Embedding A in Q/Z, we extend f to a cocycle R x a to Q/Z. As remarked at the end of Section 2, Passi [18] has proved that f is cohomologous to a factor set g from f7 x R to Q/Z f or which 1a(Z, R) x ZR is killed by the extended g. But i7 is finite; so we may view g as a factor set g x H to some cyclicp-group A* containing A. Since g and f are cohomologous, the group H* defined byg is the central product of A * and H with the central subgroup A amalgamated. But then 13(H) + I(H) I(Z(H)) is in 13(H*) + I(H*) I(Z(H*)); so we may assume H = H* to prove 3.11; that is, we may assume f is bilinear. f is defined by a set T of coset representatives for A in H; as before, T defines a map p from ZH to A. We conclude by showing that ker p contains 13(H) + WWW)). Since Z(H) = A, a typical generator of I(H)I(Z(H)) is (ta - l)(a’ - 1) where t in T, R, a’ in A. But this goes to aa’a-W-I = 1 under p. A typical generator of 13(H) is (tl a1 - 1)(&a, - 1)(&a, - 1) which p sends to f (tlt2 , f3) f (tl , $-If (& , ia)-’ by the definition of a factor set. But this is 1 by the assumed bilinearity off. We now come to the proof of 3.3, which relies upon a clever application by Moran [16] of a substantial theorem of Lazard. We first prove a special case which provides a model for the general case and whose proof is self-

THE

DIMENSION

SUBGROUP

contained. This is the case of class 2-by-Abelian 3.7 and 3.9 as before, it suffices to show: LEhmn,r-a 3.12.

Let H be a p-group,

227

PROBLEM

p-groups,

p odd.

Using

p odd. Then

H3 = Hn 1 +13(Hj+C(xthe sum taker1 OYW all pairs x, y of comn&g

l)(y-

l)ZH,

elements.

Proof. -4s usual, we may assume H3 = 1. Since p is odd, we may define the operation x f- y = x~[x, y]plja on N; that L = H under the operation + forms an Abelian group is an observation which apparently originated with Hopkins [9] and which is easy to verify. Define a map p from ZH to L by sending x to s for all x in H. If x and y commute, (X - l)(y - 1) is sent to ~y~-ryy-~ = 1; so it suffices to show that I3 is also in the kernel of p= Taking .x~y, z in H, we see that (x -

l)(y

-

l)(a -

1) = (.Xyx - yz) + (.a - XZ) f

is sent by p to x[x, y]lP[.v, ,z]ip + +[z, ~]ila + seen to be I. The Abelian group L above admits another namely, (x, y) = [x, y], under which it becomes portion of a theory of Lazard, the inversion formula, which asserts [12, p. 178 and preceding]:

((-.vy

+ X) + y)

[y, ~t.]i:~ in L which

is readily

operation derived from H, a Lie ring. This is a smal! of the Campbell-Hausdorff

THEOREM 3.13. If G is a p-group of class < p, the set G may be considered as a Lie &g L of class = class G under the operations

,Y + y = xy[x, y]-‘:‘[x, (.x, y) = [x, y][x, y, y]‘/”

y, y-lP[y,

x, .]-l/l2

..’ I

... .

If L is p-Lie ring of class
Furthermore, if L is so obtained from a p-group then His isomorphic to G. The Campbell-Hausdorff over Q by means of formal

G and H is so obtained.from

H of

L,

formula is defined in the context of free Lie rings power series for log and exp, as

xy = log ((exp X)(erp y))

228 which

SANDLING Dynkin

[3] expanded

xy = I((k

as

I)“-l/k) c

~(xa~~yal~ . . . .+yQy~

((q,!),

where the sum is over all aij > 0, 1 < i < 2, 1 < j < k, such urj + uzj > 0 for all j, and where

$YYl ..~rn> = l/N...

that

(Yl ,Y%), *.. ,Y,>.

It applies to p-Lie rings of class

p when, by assumption, it is 0. Lazard’s theorem allows us to consider anyp-group H of class


3.14.

wherethe sumis over all aij > 0, 1
O

Moran then shows that, if c = class H, le+l is in the kernel of the map ,U from ZH to L defined by sending x to x, x in H. We give his exposition here, the introduction of Lazard series and Lemma 3.16 being new. THEOREM 3.15.

Let H beup-group and n < p. Then

wherethe-first sumis O-WY all Lazard series{Ki} in Hand the secondsumis over all setsof integersii 3 1,for which h& < H, wherec = C ij . Proof. We may assume H, = 1 so that class H < p and H may be viewed as a Lie ring L as above. Thus the map p from ZH to L is available. It suffices to show that nlGjGm I(Kij) is in ker p for all Lazard series {Ki} and integers

THE

DIMENSION

SUBGROUP

PROBLEM

229

ij as described. This is generated as ,4belian group by all products (1 - x1) ~.. (1 - s,,,), where xi is in Kij . y = (l - x1) ... (1 - xi,,) = c (- l)Sw, ) sum over all subsets S of (I,..., ?ri), where, if S consists cf = (-l)d and xs = xslx,+ ... xl6 a integers si , 1 < s1 < ... < sd < m, (-1)” Under ,tiI y goes to C (-l)s~(xs). Let T be a proper subset of {i,..., nz); choose integers nij > 0, 1 < i < UZ, 1
does not involve the element x, for s not in T. ~(a~~) appears in the expression for I given by 3.14 if and only if T is a subset of S; but it appears wit?1 the coefficient (-l)s so that, in the expression for p(y), ~(~ij) appears with the coefficient C (- l)s, sum over all subsets S which contain T. Since T is proper, this coefficient is 0. To show ,I&) = 0, it remains to dispose of those terms which involve xi for all i. It suffices to show: LEn,rnIa 3.16. Let {Ki) be a Lazard series in a p-group H qf class
Proqf. We use reverse induction on the following lists of integers (s, m, , nr,-, ).~.) q) under dictionary ordering where s = x ii and m,, is the number of ij > U, to show that (...(q , x2) ,..., x2) = 0 for all choices of xj in Kf. ~ Ifs 3 cd, d = class H, this is so as either k > d or there is some ii > ‘,. To proceed with the induction we use the fact that [x,~] in H is (x, J?) + terms of weight 23, which may be readily calculated from the first few terms of the Campbell-Hausdorff formula. But then (...@l I x2>, “Q) )...) XR) =: (. ..( [x1 ) x2], X-J,.. ~) XJ + terms involving at least three xi’s, i 1 1 or 2. The induction hypothesis shows immediately that the latter terms since s increases. The first term is eliminated because [x1 , x2] is in

[Kil , “i,l

< K, ,

are 0

n = il + i, ;

so that PK, increases. With 3.15 proved, we can see that the dimension subgroup problem is true for p-groups of class and integers &=l,l
230

SANDLING

Returning

to the proof of 3.3, we see by 3.7 and 3.9 that, assuming i&Z,

G) = G’ n 1 + 1 nl(Gnj) j

G,n = 1,

ZG’.

But we are assuming that class G’ < p, so that 3.15 applies to H = G’. Let Kg = Gi, ; then [Ki , KJ < K.i+j+l ; SO {KJ is a Lazard series and i&Z, G) = G’ n 1 + x JJIj I(K,,)ZG’, sum over integers is , n > ij > 1, for which C ij = II - 1. Let c = 1 + class G’ which is
4. THE TRANSFINITE If the of all G, I = I(Z, thus, by

ZG’ = i(,)(Z,

G).

POWERS OF THE AUGMENTATION

IDEAL

dimension subgroup conjecture is true, then G, , the intersection , is the subgroup attached to the ideal IW, the intersection of all P, G). Buckley [l] sh owed that this was the case for finite groups; the usual reductions we have

THEOREM 4.1.

Let G be Jinitely generated.

Then

i, =Gnl+P=G,. If the dimension subgroup conjecture is false, a search for counterexamples to i, = G, among infinite groups might well be more promising than a search for a minimal counterexample to ifi = G,, among finite p-groups. A proof of 4.1 for finite groups may be obtained from the following decomposition of IO,. LEMMA 4.2. Iw = I(GJ ZG + C I(P) I(Q) ZG, the sum taken p-subgroups P and q-subgroups Q where p = q and [P, Q] = 1. This lemma and methods of Gruenberg (see [19]). of Ltn = G r\ 1 +IUP

and Roseblade

lead to a calculation

THEOREM 4.3. Let M = G, and IV, be a Sylow p-subgroup S, a Sylow p-subgroup of G for all p. Then G+n.

over all

of N

and

= N n [IV, , S, , $1 n [N, , n%l,

where the jirst product is taken over all triples of district primes p, q, P and the second over all primes p. Furthermore, IwlIuI” is isomorphic to iJiw+a .

THE

We wish couragement,

DIMENSION

SUBGROUP

to thank James Roseblade the N.S.F. for its support

PROBLEM

and John Thompson for and Cambridge University

231

their help and enfor Its hospitalitp.

REFERENCES 1. J. BUCKLEY, On the D-series of a finite group, Proc. LAmr. Math. SW. 18 (1967), 185-186. 2. P. COHN, Generalization of a theorem of l&gnus, AK. L~WZ&JFT A~&. SOC. (3) 2 (1952), 297-310. 3. E. B. DYNKIN, Normed Lie algebras and analytic groups, L:spehi Mat. A\ea~rk RIath. Sot. Translations 97, (X.S.) 5 (1950), 135-186 (Russian); Amer. Providence, R.I., 1953. 4. R. H. Fox, Free differential calculus I, Ann. of Math. 57 (1953), 547-560. 5. 0. GROAN, Zusammenhang zwischen Potenzbildung und Kom~n~~~iltorbildung, J. Reim A-ngem. Math. 182 (1940), 158-177. 6. K. W. GRUENBERG, ‘Some cohomological topics in group theory,” Queen Mary College Math. Notes, London, 1968; Lecture Notes in biath. 143, Springer, Berlin, 1970. 7. P. HALL: “Nilpotent Groups,” Cand. Math. Congress, Univ. of -Alberta, 1957; Queen Mary College Math. Notes, London, 1970. 8. =2. H. M. HOARE, Group rings and lower central series, J. Lo&on illirrh. Sot. (2) I (1969), 37-40. 9. C. HOPKINS, Metabelian groups of order p”, p > 2, Tt'G?lS, &her. ~htth. SOC. 37 (1935), 161-195. 10. B. HUPPERT, “Endliche Gruppen I,” Springer, Berlin, 1967. 11. S. A. JENNINGS, The structure of the group ring of a p-group over a modular field, I’rans. Anzer. Math. Sot. 50 (1941), 175-185. 12. M. LAZARD, Sur les groupes nilpotents et les anneaclx de Lie, Ac7r. &ok A~owz. Sup. 71 (1954), 101-190. 13. G. Loser, On dimension subgroups, Trans. Amer. Akztlz. Sot. 97 (1960), 474-486. 14. TV. h’I.%Nus, .tiber Beziehungen zwischen hiiheren Kommuratoren, j. Reine rlngezu. Math. 177 (1937), 105-115. 15. S. MORAN, t0 appear. 16. S. &~OR~h’, Dimension subgroups module n, Proc. Cnmb. Phil. Sot. 68 (I 97O), 579-582. 17. H. NEUAMNN, “Varieties of Groups,” Springer, Berlin, 1967. 18. I. B. S. PASSI, Dimension subgroups, J. AZgebrn 9 (1968), 152-182. 19. R. S.ANDLING, The modular group rings of p-groups, Thesis, Univ. of Chicago, Chicago, Ill., 1969. 20. R. SAXDLING, Dimension subgroups over arbitrary coefticient rings, J. Algebra 21 (19?2), 250-3-G. 21. R. SWDLING, Subgroups dual to dimension subgroups, Proc. Cawzb. Phil. Sot. 71 (1972), 33-38.