, , November 27, 2006

The first motivating result for this paper is the following famous theorem of J. Thompson [17] (see also [5] ): a finite group

G

is solvable if and only if every 2-generated subgroup of

G

is solvable.

Our main aim is to prove the following extension of Thompson's theorem.

Theorem 1.1. Let

G

be a finite group, and let

R (G)

be the solvable radical of

G

(namely the maximal solvable normal subgroup of

G

). Then

R (G)

coincides with the set of all elements

y \in G

with the following property:

for any

x \in G

the subgroup generated by

x

and

y

is solvable.

Our proof of Theorem 1.1 invokes the Classification of finite simple groups, and uses the so called “one and a half generation” of almost simple groups, proved by Guralnick and Kantor [7] using probabilistic arguments. Thus our result may be regarded as yet another demonstration of the power of probabilistic and counting methods in group theory (see the survey paper [15] for further background).

Theorem 1.1 can be extended to some classes of infinite groups (see Theorems 4.1 and 4.4 below). It has an obvious Lie-algebraic counterpart (Theorem 2.1 ). We call a criterion given in Theorem 1.1 the Thompson-like characterization of the solvable radical.

Acknowledgements. Kunyavskiĭ and Plotkin were partially supported by the Ministry of Absorption (Israel), the Israel Science Foundation founded by the Israel Academy of Sciences — Center of Excellence Program, the Minerva Foundation through the Emmy Noether Research Institute of Mathematics, and the EU networks HPRN-CT-2002-00287 and INTAS 00-566. Shalev was partially supported by the Israel Science Foundation founded by the Israel Academy of Sciences .

The authors are very grateful to B. Plotkin for inspiring suggestions and discussions.

2 Lie algebras

We start with a Lie-algebraic counterpart of Theorem 1.1 which will give us important hints to its proof.

Theorem 2.1. Let

L

be a finite dimensional Lie algebra defined over a field

k

of characteristic zero, and let

R (L)

be the solvable radical of

L

(namely the maximal solvable ideal of

L

). Then

R (L)

coincides with the set of elements

y \in L

with the following property: for any

x \in L

the subalgebra generated by

x

and

y

is solvable.

Proof. If $y \in R (L)$ , then for any $x \in L$ the subalgebra generated by $x$ and $y$ contains a solvable ideal with one-dimensional quotient and is therefore solvable. We shall give two proofs for the converse inclusion.
1st proof. Suppose $y / \in R (L)$ . We have to prove that there is $x \in L$ such that the subalgebra generated by $x$ and $y$ is not solvable. After factoring out $R (L)$ , we are reduced to proving this in the case where $L$ is semisimple. Clearly, it is enough to consider the case where $L$ is simple. In that case the result follows from [8] where it is proved that for any nonzero element $y$ of a simple Lie algebra $L$ there is $x$ such that $x$ and $y$ generate $L$ .
2nd proof. Suppose $y$ has the property stated in the theorem. We have to prove that $y \in R (L)$ . Consider the sequence of words $v_{n} (x, y)$ defined inductively by the following rule:
$v_{1} (x, y) = x, v_{n + 1} (x, y) = [v_{n} (x, y), [x, y]], . . .$ This sequence can be used for characterization of $R (L)$ : according to [2,Theorem3.7] , $y \in R (L)$ if and only if for any $x$ there exists $n$ such that $v_{n} (x, y) = 0$ .
Let now $x$ be an arbitrary element of $L$ . Since the subalgebra generated by $x$ and $y$ is solvable, it satisfies the identity $v_{n} (x, y) \equiv 0$ for some $n$ [2,Theorem3.4] , and we are done.
□

Remark 2.2. In view of results of Section 4 , it seems plausible that Theorem 2.1 can be extended to some classes of infinite dimensional Lie algebras (in particular, to Lie algebras with polynomial identity).

3 Finite groups

In this section we prove Theorem 1.1 . For brevity, let us introduce the following notion.

Definition 3.1. Let

G

be a group. We say that

y \in G

is a radical element if for any

x \in G

the subgroup generated by

x

and

y

is solvable. Denote by

S (G)

the set of radical elements of

G

Note that in any group

G

we have

R (G) \subseteq S (G)

. Indeed, if

y \in R (G)

, then for any

x

the subgroup generated by

x

and

y

contains a solvable normal subgroup with cyclic quotient and is therefore solvable.

With this terminology, Theorem 1.1 says that if

G

is finite, then

R (G) = S (G)

Proof. It suffices to prove that S ( G ) ⊆ R ( G ) . One could try to mimic one of two proofs of Theorem 2.1 . Thus Theorem 1.1 would immediately follow from the following
Conjecture 3.2. [2,Conj.2.12] For any finite group $G$ there exists a sequence $u_{n} (x, y)$ such that $R (G)$ coincides with the set of elements $y$ having the following property: for any $x \in G$ there is $n$ such that $u_{n} (x, y) = 1$ .
However, [2] contains only partial results towards Conjecture 3.2 . Fortunately, the “one and a half generation” theorem of Guralnick and Kantor [7,CorollaryofTh. Ionp. 745] allows us to imitate the first proof of Theorem 2.1
.
First, we prove an auxiliary result. For a finite group T and a positive integer n denote by T n the direct product of n copies of T . Let σ be an automorphism of T . For any group T and x ∈ T , σ ∈ Aut T , denote by 〈 x 〈 σ 〉 〉 the minimal subgroup containing x and invariant under the automorphism σ .
Lemma 3.3. Suppose T is a nonabelian simple group, and let σ ∈ Aut ( T n )
.
If $σ \neq 1$ , then there exists $\bar{x} \in T^{n}$ such that the subgroup $〈 {\bar{x}}^{〈 σ 〉} 〉$ is not solvable.

Proof. It is well known (see, for example, [14,3.3.20] ) that $Aut (T^{n}) = Aut (T) w r S_{n} = (Aut T)^{n} ⋋ S_{n}$ , where $S_{n}$ is the symmetric group. Then one can write $σ = ((σ_{1}, . . ., σ_{n}), s)$ , where $σ_{i} \in Aut T$ , $s \in S_{n}$ .
Suppose $s \neq 1$ . In this case we prove that there is $\bar{x} \in T^{n}$ such that already the subgroup generated by $\bar{x}$ and ${\bar{x}}^{σ}$ is not solvable.
We may assume that $s (k) = 1$ for some $k > 1$ . Then $(\bar{x})^{σ} = (x_{1}, . . ., x_{n})^{σ} = (x_{1}^{σ_{1}}, . . ., x_{n}^{σ_{n}})^{s} = (x_{k}^{σ_{k}}, . . .)$ . Now we can use the above cited “one and a half generation” theorem by Guralnick and Kantor. Choose $\bar{x}$ so that $x_{1} \neq 1$ . Since $T$ is simple, according to [7] there is $b \in T$ such that $b$ and $x_{1}$ generate $T$ . One can choose $\bar{x}$ so that $x_{k}^{σ_{k}} = b$ (put $x_{k} = b^{σ_{k}^{- 1}}$ ). With such a choice of $x_{1}$ and $x_{k}$ , the element $\bar{x} = (x_{1}, . . ., x_{k}, . . .)$ is as required: the group $〈 \bar{x}, {\bar{x}}^{σ} 〉$ cannot be solvable since the first components of $\bar{x}$ and ${\bar{x}}^{σ}$ are $x_{1}$ and $b$ , and they generate the whole group $T$ .
Let now $s = 1$ . Consider the semidirect product $T_{1}$ of the first copy of $T$ with the automorphism $σ_{1}$ . This group is almost simple and once again we can use [7] in order to find for $σ_{1} \in T_{1}$ an element $τ \in T_{1}$ such that $σ_{1}$ and $τ$ generate the subgroup which contains the socle of $T_{1}$ and thus is not solvable. Since $τ = x_{1} σ_{1}^{m}$ for some integer $m$ and some $x_{1} \in T$ , we conclude that the subgroup $〈 x_{1}, σ_{1} 〉 = 〈 τ, σ_{1} 〉$ , where $〈, 〉$ stands for the subgroup generated by elements in brackets, is not solvable. Hence the group $〈 x_{1}^{〈 σ_{1} 〉} 〉$ is not solvable. If we take $\bar{x} = (x_{1}, 1, \dots, 1)$ , this completes the proof.
□

We are now able to prove that $S (G) \subseteq R (G)$ . It is easy to see that $S (G / R (G)) = S (G) / R (G)$ . Factoring out $R (G)$ , we may assume that $G$ is semisimple (i.e. $R (G) = 1$ ).
We have to prove that $S (G) = 1$ .
According to [14,3.3.16] , any finite semisimple group $G$ contains a unique maximal normal centreless completely reducible subgroup $C R (G)$ (by definition, CR means a direct product of finite nonabelian simple groups) which is called the CR-radical of $G$ . It is known that the centralizer of $C R (G)$ is trivial [14,proofof3.3.18(i)] . We call a product of the isomorphic factors in the decomposition of the $C R$ -radical an isotypic component of $G$ . Thus $C R (G) = H_{1} \times \dots \times H_{t}$ , where $H_{i}$ is an isotypic component.
Any $y \in G$ acts (by conjugation) as an automorphism $\hat{y}$ on $C R (G)$ and, correspondingly, on every isotypic component $H_{i} = T_{i}^{n_{i}}$ where $T_{i}$ is a nonabelian simple group and $n_{i} \geq 1$ .
According to Lemma 3.3 if $\hat{y} \neq 1$ then there exists $\bar{x} \in H_{i}$ such that the subgroup $〈 {\bar{x}}^{〈 \hat{y} 〉} 〉$ is not solvable. However, the subgroup $〈 \bar{x}, y 〉$ is solvable if and only if the subgroup $〈 {\bar{x}}^{〈 \hat{y} 〉} 〉$ is solvable. Thus, if $y \in S (G)$ then $\hat{y} = 1 .$ This means that $y$ belongs to the centralizer $C_{G} (C R (G))$ which is trivial. Hence $S (G) = 1$ .
The theorem is proved. □

Corollary 3.4. Let

G

be a finite group, let

y \in G

, and let

〈 y^{G} 〉

denote the minimal normal subgroup of

G

containing

y

. Then

〈 y^{G} 〉

is solvable if and only if the subgroup

〈 y^{〈 x 〉} 〉

is solvable for all

x \in G

4 Linear groups and PI-groups

Theorem 4.1. Let

K

be a field. If

G \leq GL (n, K)

, then

R (G) = S (G)

Proof. As noted earlier we have $R (G) \subseteq S (G)$ , and therefore it suffices to prove that the subgroup $H$ generated by the set $S (G)$ is solvable (and thus coincides with the radical).
Let $H_{1} = 〈 g_{1}, . . ., g_{s} 〉$ be a finitely generated subgroup of $H$ where all $g_{i}$ 's are radical elements. Then $H_{1}$ is approximated by finite linear groups $G_{α} = H_{1} / N_{α}, \cap N_{α} = 1$ in dimension $n$ [9] . Each $G_{α}$ is finite and is generated by the images of radical elements which are radical as well, and thus $G_{α}$ is solvable. Since all $G_{α}$ 's are linear in dimension $n$ , their derived length is bounded, say, by $k = k (n)$ . Thus the group $H_{1}$ has derived length at most $k$ . Each finitely generated subgroup of $H$ lies in some $H_{1}$ . Thus $H$ is locally solvable. Since $H$ is linear, it is solvable [18] . □

For the case of PI-groups we use some facts from [10] , [12] , [13] .

Definition 4.2. A group

G

is called a PI-group (PI-representable in terms of [10] ) if

G

is a subgroup of the group of invertible elements of an associative PI-algebra over a field.

Linear groups are a particular case of PI-groups. It is known that every PI-group

G

has a unique maximal locally solvable normal subgroup

R (G)

called the locally solvable radical of

G

and that the locally solvable radical of a finitely generated PI-group is solvable [10] . (For arbitrary groups the locally solvable radical may not exist, and for arbitrary PIgroups the locally solvable radical is not necessarily solvable).

PI-groups have the following invariant series:

1 ◃ H_{0} ◃ H ◃ G

where

H_{0}

is a locally nilpotent normal subgroup,

H / H_{0}

is nilpotent and

G / H

is a linear group over a cartesian sum of fields [13] .

We want to show that in a PI-group

G

the locally solvable radical coincides with

S (G)

Let us introduce a useful notion of oversolvable group.

Definition 4.3. (cf. [12] ). A group

G

is called oversolvable if it has an ascending normal series with locally nilpotent factors.

An arbitrary group

G

has the oversolvable radical

\tilde{H P} (G) = \tilde{η} (G)

(that is the unique maximal normal oversolvable subgroup). The quotient group

G / \tilde{η} (G)

is semisimple with respect to the property of being locally nilpotent, i.e.

η (G / \tilde{η} (G)) = 1

where

η (G)

is the locally nilpotent radical of

G

(see [12] for the above facts). If

G

is finite, noetherian, or linear,

\tilde{η} (G)

coincides with the solvable radical

R (G)

[16] .

Theorem 4.4. If

G

is a PI-group, then

R (G) = \tilde{η} (G) = S (G)

Proof. We consider three cases. 1. $G \leq {GL}_{n} (P)$ where $P$ is a field. 2. $G \leq {GL}_{n} (K)$ where $K$ is a cartesian sum of fields. 3. General case.
Case 1. If $G \leq {GL}_{n} (P)$ , where $P$ is a field, then $R (G) = S (G)$ by Theorem 4.1 .
Case 2. Suppose $G \leq {GL}_{n} (K)$ , where $K = \oplus_{s} P_{s}$ is a cartesian sum of fields. Consider the set of congruence subgroups $U_{s}$ of ${GL}_{n} (K)$ such that ${GL}_{n} (K) / U_{s} \sim = {GL}_{n} (P_{s})$ . Since $\cap_{s} U_{s} = 1$ , the group $G$ lies in the cartesian product $\prod_{s} {GL}_{n} (P_{s}) .$ We have $G \subset {GL}_{n} (K) \subset \prod_{s} {GL}_{n} (P_{s}) \to {GL}_{n} (P_{s})$ . Let $H$ be the subgroup in $G$ generated by the set $S (G)$ . It is enough to show that $H$ is solvable. Set $U_{s}^{'} = H \cap U_{s}$ . Then $\cap_{s} U_{s}^{'} = 1$ . Each $H / U_{s}^{'}$ can be viewed as a subgroup in ${GL}_{n} (P_{s})$ and is therefore generated by the radical elements. Thus they are all solvable of bounded derived length. Therefore $H$ is solvable and $R (G) = S (G)$ .
Similar arguments give $R (G) = \tilde{η} (G)$ .
Case 3. Let us first show that $R (G) \subseteq S (G)$ . Let $g \in R (G)$ , $h \in G$ . Consider the subgroup $G_{0} = 〈 g, h 〉$ . We have $g \in R (G) \cap G_{0}$ and, consequently, $g \in R (G_{0})$ . By [10] , the locally solvable radical of a 2-generated group is solvable. So the group $G_{0}$ is solvable as a cyclic extension of a solvable group.
Now we want to show that $R (G) = \tilde{η} (G)$ . Let us first prove that $\tilde{η} (G) \subseteq R (G) .$ We have to prove that the group $\tilde{η} (G)$ is locally solvable. We take a finitely generated subgroup $G_{0}$ in $\tilde{η} (G)$ and show that $G_{0}$ is solvable. Consider the locally solvable radical $R (G_{0})$ . Since $G_{0}$ is finitely generated, the radical $R (G_{0})$ is solvable [10] . So it is enough to prove that the group $G_{0} / R (G_{0})$ is solvable. We use the following result about the structure of PI-groups [13] : in every PI-group $G$ the quotient group $\tilde{η} (G) / η (G)$ is solvable. We have $η (G) \subseteq R (G)$ as a locally nilpotent subgroup. Apply this to $G_{0}$ . Since $G_{0} \subseteq \tilde{η} (G)$ , we have $\tilde{η} (G_{0}) = G_{0}$ .
Since $\tilde{η} (G_{0}) / η (G_{0})$ is solvable and $η (G_{0}) \subseteq R (G_{0})$ , the group $\tilde{η} (G) / R (G_{0})$ is solvable.
Thus $G_{0} / R (G_{0})$ is solvable, and hence so is $G_{0}$ . The inclusion $\tilde{η} (G) \subseteq R (G)$ is proved.
Let us prove the opposite inclusion $R (G) \subseteq \tilde{η} (G)$ . Recall that in every PI-group $G$ there is a normal subgroup $H$ which is an extension of a locally nilpotent group by a nilpotent group and such that $G / H$ lies in ${GL}_{n} (K)$ where $K$ is a cartesian sum of fields.
Then $H$ is oversolvable. Therefore $H \subseteq \tilde{η} (G)$ , and thus $H \subseteq R (G)$ . Consider the group $G / H$ and its subgroup $R (G) / H$ . This is a locally solvable normal subgroup in $G / H$ and thus lies in $R (G / H)$ . The group $G / H$ is linear and hence $R (G / H) = \tilde{η} (G / H)$ . Then $R (G) / H \subseteq \tilde{η} (G) / H = \tilde{η} (G / H)$ . Thus $R (G) \subseteq \tilde{η} (G)$ and $R (G) = \tilde{η} (G)$ .
We are now able to prove that $S (G) \subseteq R (G)$ . Let $g \in S (G)$ . Denote by $\bar{g} \in G / H$ its image under the natural projection. Then $\bar{g} \in R (G / H)$ and thus $\bar{g} \in \tilde{η} (G) / H .$ Then $g \in \tilde{η} (G) = R (G)$ . □

The above theorem has an obvious consequence which can be viewed as a natural generalization of Thompson's theorem:

Corollary 4.5. A PI-group

G

is locally solvable if and only if every two-generated subgroup of

G

is solvable.

Remark 4.6. In linear groups the locally solvable radical is solvable. From the above theorem it follows that in PI-groups the locally solvable radical is solvable modulo the locally nilpotent radical. Indeed,

\tilde{η} (G) / η (G)

is solvable [13] , and

\tilde{η} (G) = R (G)

by Theorem 4.4 .

5 Concluding Remarks

Let us observe that certain important classes of groups and Lie algebras can be explicitly characterized in terms of two-variable identities: one can mention here classical results for finite dimensional nilpotent Lie algebras (Engel) and finite nilpotent groups (Zorn [19] ) and their recently obtained counterparts for finite dimensional solvable Lie algebras [6] and finite (or linear) solvable groups [3] (see also [4] ). Moreover, Engel identities were used by Baer to characterize explicitly the nilpotent radical of an arbitrary finite (and, more generally, noetherian) group [1] . Baer's theorem was extended to the locally nilpotent radical of linear groups and PI-groups [11] , [13] and to the nilpotent and the solvable radical of finite dimensional Lie algebras [2] . These results give a certain hope for characterization of the solvable radical

R (G)

of a finite group

G

in similar, Engel-like terms. However, the corresponding Conjecture 3.2 (see also [2] ) is still far from being proved, and therefore less explicit descriptions of the solvable radical, such as the Thompson-like characterization of Theorem 1.1 , are very useful.

References

R. Baer, Engelsche Elemente Noetherscher Gruppen, Math. Ann. 133 (1957), 256–270.
T. Bandman, M. Borovoi, F. Grunewald, B. Kunyavskiĭ, and E. Plotkin, Engel-like characterization of radicals in finite dimensional Lie algebras and finite groups, preprint math.GR/0411463.
T. Bandman, G.-M. Greuel, F. Grunewald, B. Kunyavskiĭ, G. Pfister, and E. Plotkin, Two-variable identities for finite solvable groups, C.R. Acad. Sci. Paris, Ser. I 337 (2003), 581–586.
J. N. Bray, J. S. Wilson, and R. A. Wilson, A characterization of finite soluble groups by laws in two variables, Bull. London Math. Soc. 37 (2005), 179–186.
P. Flavell, Finite groups in which every two elements generate a soluble group, Invent. Math. 121 (1995), 279–285.
F. Grunewald, B. Kunyavskiĭ, D. Nikolova, and E. Plotkin, Two-variable identities in groups and Lie algebras, Zap. Nauch. Semin. POMI 272 (2000), 161–176; J. Math. Sci. (New York) 116 (2003), 2972–2981.
R. M. Guralnick and W. M. Kantor, The probability of generating a simple group, J. Algebra 234 (2000), 743–792.
T. Ionescu, On the generators of semi-simple Lie algebras, Linear Algebra and its Appl. 15 (1976), 271–292.
A. I. Malcev, On faithful representations of infinite groups of matrices, Mat. Sb. 8 (1940), 405–422; English transl. in Amer. Math. Soc. Transl. Ser. (2) 45 (1965), 1–18.
S. A. Pihtil'kov, On the prime radical of PI-representable groups, Mat. Zametki 72:5 (2002), 739–744; English transl. in Math. Notes 72:5 (2002), 682–686.
V. P. Platonov, Engel elements and the radical in PI-algebras and topological groups, Dokl. Akad. Nauk SSSR 161 (1965), 288–291; English transl. in Sov. Math. Dokl. 6 (1965), 412–415.
B. I. Plotkin, Radical groups, Mat. Sb. N.S. 37(79) (1955), 507–526; English transl. in Amer. Math. Soc. Transl. (2) 17 (1961), 9–28.
B. I. Plotkin, Notes on Engel groups and Engel elements in groups. Some generalizations, Preprint, available at http://arXiv.org/math.GR/0406100.
D. J. S. Robinson, A Course in the Theory of Groups, Springer-Verlag, New York, 1995.
A. Shalev, Probabilistic group theory, Groups St Andrews 1997 in Bath, II, London Math. Soc. Lecture Note Series 261, Cambridge University Press, Cambridge, 1999, pp. 648–678.
D. A. Suprunenko, Matrix Groups, Nauka, Moscow, 1972; English transl. in Transl. Math. Monographs, vol. 45, Amer. Math. Soc., Providence, RI, 1976.
J. Thompson, Non-solvable finite groups all of whose local subgroups are solvable, Bull. Amer. Math. Soc. 74 (1968), 383–437.
H. Zassenhaus, Beweis eines Satzes über diskrete Gruppen, Abh. Math. Sem. Univ. Hamburg 12:3/4 (1938), 289–312.
M. Zorn, Nilpotency of finite groups, Bull. Amer. Math. Soc. 42 (1936), 485–486.

Kunyavskiɪ and Plotkin: Department of Mathematics, Bar-Ilan University, 52900 Ramat Gan, ISRAEL E-mail address : kunyav@macs.biu.ac.il, plotkin@macs.biu.ac.il Shalev: Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, Hebrew University of Jerusalem, 91904 Jerusalem, ISRAEL E-mail address : shalev@math.huji.ac.il

November 27, 2006

Boris Kunyavskiĭ, Eugene Plotkin, and Aner Shalev