ABSTRACT QUOTIENTS OF PROFINITE GROUPS, AFTER NIKOLOV AND SEGAL BENJAMIN KLOPSCH Abstract. In this expanded account of a talk given at the Oberwolfach Ar- 6 1 beitsgemeinschaft “Totally Disconnected Groups”, October 2014, we discuss 0 results of Nikolay Nikolov and Dan Segal on abstract quotients of compact 2 Hausdorff topological groups, paying special attention to the class of finitely v generatedprofinitegroups. Ourprimarysourceis[17]. Sidesteppingalldifficult o andtechnicalproofs,wepresentaselectionofaccessibleargumentstoilluminate N key ideas in the subject. 2 2 1. Introduction ] §1.1. Many concepts and techniques in the theory of finite groups depend in- R trinsically on the assumption that the groups considered are a priori finite. The G theoretical framework based on such methods has led to marvellous achievements, . h including – as a particular highlight – the classification of all finite simple groups. t a Notwithstanding, the same methods are only of limited use in the study of infinite m groups: it remains mysterious how one could possibly pin down the structure of a [ general infinite group in a systematic way. Significantly more can be said if such a 2 group comes equipped with additional information, such as a structure-preserving v action on a notable geometric object. A coherent approach to studying restricted 3 4 classes of infinite groups is found by imposing suitable ‘finiteness conditions’, i.e., 3 conditions that generalise the notion of being finite but are significantly more 0 0 flexible, such as the group being finitely generated or compact with respect to a . natural topology. 1 0 One rather fruitful theme, fusing methods from finite and infinite group theory, 6 consists in studying the interplay between an infinite group Γ and the collection of 1 : all its finite quotients. In passing, we note that the latter, subject to surjections, v naturally form a lattice that is anti-isomorphic to the lattice of finite-index normal i X subgroups of Γ, subject to inclusions. In this context it is reasonable to focus on r a Γ being residually finite, i.e., the intersection N of all finite-index normal TNEfΓ subgroups of Γ being trivial. Every residually finite group Γ embeds as a dense subgroup into its profinite completion Γ = lim Γ/N. The latter can be con- b ←−NEfΓ structed as a closed subgroup of the direct product Γ/N of finite discrete QNEfΓ groups and thus inherits the structure of a profinite group, i.e., a topological group that is compact, Hausdorff and totally disconnected. The finite quotients of Γ are in one-to-one correspondence with the continuous finite quotients of Γ. b Profinite groups also arise naturally in a range of other contexts, e.g., as Galois groups of infinite field extensions or as open compact subgroups of Lie groups over non-archimedean local fields, and it is beneficial to develop the theory of profinite groups in some generality. At an advanced stage, one is naturally led to examine more closely the relationship between the underlying abstract group 2010 Mathematics Subject Classification. Primary 20E18;Secondary 20F12,22C05. 1 2 BENJAMIN KLOPSCH structure of a profinite group G and its topology. In the following we employ the adjective ‘abstract’ to emphasise that a subgroup, respectively a quotient, of G is not required to be closed, respectively continuous. For instance, as G is compact, an abstract subgroup H is open in G if and only if it is closed and of finite index in G. A profinite group may or may not have (normal) abstract subgroups of finite index that fail to be closed. Put in another way, G may or may not have non-continuous finite quotients. What can be said about abstract quotients of a profinite group G in general? When do they exist and how ‘unexpected’ can their features possibly be? To the newcomer, even rather basic groups offer some initial surprises. For instance, consider for a prime p the procyclic group Z = lim Z/pkZ of p-adic p ←−k∈N integers under addition. Of course, its proper continuous quotients are just the finite cyclic groups Z /pkZ . As we will see below, the abelian group Z fails to p p p map abstractly onto Z, but does have abstract quotients isomorphic to Q. §1.2. In [17], Nikolov and Segal streamline and generalise their earlier results that led, in particular, to the solution of the following problem raised by Jean-Pierre Serre in the 1970s: are finite-index abstract subgroups of an arbitrary finitely generated profinite group G always open in G? Recall that a profinite group is said to be (topologically) finitely generated if it contains a dense finitely generated abstract subgroup. The problem, which can be found in later editions of Serre’s book on Galois cohomology [22, I.§4.2], was solved by Nikolov and Segal in 2003. Theorem 1.1 (Nikolov, Segal [14]). Let G be a finitely generated profinite group. Then every finite-index abstract subgroup H of G is open in G. Serrehadprovedthisassertioninthespecialcase, whereGisafinitelygenerated pro-p group for some prime p, by a neat and essentially self-contained argument; compareSection2. Theproofofthegeneral theoremisconsiderably moreinvolved and makes substantial use of the classification of finite simple groups; the same is true for several of the main results stated below. The key theorem in [17] concerns normal subgroups in finite groups and estab- lishes results about products of commutators of the following type. There exists a function f: N → N such that, if H E Γ = hy ,...,y i for a finite group Γ, 1 r then every element of the subgroup [H,Γ] = h[h,g] | h ∈ H,g ∈ Γi is a product of f(r)factorsoftheform[h ,y ][h ,y−1]···[h ,y ][h ,y−1]withh ,...,h ∈ H. 1 1 2 1 2r−1 r 2r r 1 2r Under certain additional conditions on H, a similar conclusion holds based on the significantly weaker hypothesis that Γ = Hhy ,...,y i = Γ′hy ,...,y i, where 1 r 1 r Γ′ = [Γ,Γ] denotes the commutator subgroup. A more precise version of the re- sult is stated as Theorem 3.1 in Section 3. By standard compactness arguments, one obtains corresponding assertions for normal subgroups of finitely generated profinite groups; this is also explained in Section 3. Every compact Hausdorff topological group G is an extension of a compact connected group G◦, its identity component, by a profinite group G/G◦. By the Levi–Mal’cev Theorem, e.g., see [10, Theorem 9.24], the connected component G◦ is essentially a product of compact Lie groups and thus relatively tractable. We state two results whose proofs require the new machinery developed in [17] that goes beyond the methods used in [14, 15, 16]. Theorem 1.2 (Nikolov, Segal [17]). Let G be a compact Hausdorff topological group. Then every finitely generated abstract quotient of G is finite. ABSTRACT QUOTIENTS OF PROFINITE GROUPS 3 Theorem 1.3 (Nikolov, Segal [17]). Let G be a compact Hausdorff topological group such that G/G◦ is topologically finitely generated. Then G has a countably infinite abstract quotient if and only if G has an infinite virtually-abelian contin- uous quotient. In a recent paper, Nikolov and Segal generalised their results in [17] to obtain the following structural theorem. Theorem 1.4 (Nikolov, Segal [18]). Let G be a compact Hausdorff group such that G/G◦ is topologically finitely generated. Then for every closed normal subgroup H E G the abstract subgroup [H,G] = h[h,g] | h ∈ H,g ∈ Gi is closed in G. c There are many interesting open questions regarding the algebraic properties of profinite groups; the introductory survey [12] provides more information as well as a range of ideas and suggestions for further investigation. In Section 4 we look at one possible direction for applying and generalising the results of Nikolov and Segal, namely in comparing the abstract and the continuous cohomology of a finitely generated profinite group. 2. Serre’s problem on finite abstract quotients of profinite groups §2.1. A profinite group G is said to be strongly complete if every finite-index abstract subgroup is open in G. Equivalently, G is strongly complete if every finite quotient of G is a continuous quotient. Clearly, such a group is uniquely determined by its underlying abstract group structure as it is its own profinite completion. It is easy to see that there are non-finitely generated profinite groups that fail to be strongly complete. For instance, G = lim Cn ∼= Cℵ0, the direct product ←−n∈N p p of a countably infinite number of copies of a cyclic group C of prime order p, has p 22ℵ0 subgroups of index p, but only countably many of these are open. This can be seen by regarding the abstract group G as a vector space of dimension 2ℵ0 over a field with p elements. Implicitly, this simple example uses the axiom of choice. Without taking a stand on the generalised continuum hypothesis, it is slightly more tricky to produce an example of two profinite groups G and G that are 1 2 non-isomorphic as topological groups, but nevertheless abstractly isomorphic. As anillustration, wediscuss aconcrete instance. In[8], JonathanA.Kiehlmann clas- sifies more generally countably-based abelian profinite groups, up to continuous isomorphism and up to abstract isomorphism. Proposition 2.1. Let p be a prime. The pro-p group G = ∞ C is abstractly Qi=1 pi isomorphic to G×Z , but there is no continuous isomorphism between the groups p G and G×Z . p Proof. The torsion elements form a dense subset in the group G, but they do not in G×Z . Hence the two groups cannot be isomorphic as topological groups. p It remains to show that G is abstractly isomorphic to G×Z . Let U ⊆ P(N) = p {T | T ⊆ N} be a non-principal ultrafilter. This means: U is closed under finite intersections and under taking supersets; moreover, whenever a disjoint union T ⊔T of two sets belongs to U, precisely one of T and T belongs to U; finally, U 1 2 1 2 does not contain any one-element sets. Then U contains all co-finite subsets of N, and, in fact, one justifies the existence of U by enlarging the filter consisting of all co-finite subsets to an ultrafilter; this process relies on the axiom of choice. 4 BENJAMIN KLOPSCH Informally speaking, we use the ultrafilter U as a tool to form a consistent limit. Indeed, with the aid of U, we define ∞ ψ: G ∼= YZ/piZ −→ Zp ∼= l←im−Z/pjZ, i=1 j 2 x = (x +pZ,x +p Z,...) 7→ lim(xψ ), 1 2 ←− j j where xψ = a+pjZ ∈ Z/pjZ if U (a) = {k ≥ j | x ≡ a} ∈ U. j j k pj Observe that U ∋ {k | k ≥ j} = U (0)⊔···⊔U (pj−1) to see that the definition of j j ψ is valid. It is a routine matter to check that ψ is a (non-continuous) homomor- j phism from G onto Z . Furthermore, G decomposes as an abstract group into a p ∼ direct product G = kerψ×H, where H = h(1,1,...)i = Z is the closed subgroup p ∼ generated by (1,1,...). Consequently, kerψ = G/H as an abstract group. Finally, we observe that ∞ ∞ ϑ: G ∼= YZ/piZ −→ G ∼= YZ/piZ, (xi +piZ)i 7→ (xi −xi+1 +piZ)i i=1 i=1 is a continuous, surjective homomorphism with kerϑ = H. This shows that ∼ ∼ G = G/H ×H = G×Z p as abstract groups. (cid:3) Theorem 1.1 states that finitely generated profinite groups are strongly com- plete: there areno unexpected finite quotients ofsuch groups. What aboutinfinite abstract quotients of finitely generated profinite groups? In view of Theorem 1.1, the following proposition applies to all finitely generated profinite groups. Proposition 2.2. Let G be a strongly complete profinite group and N E G a normal abstract subgroup. If G/N is residually finite, then N E G and G/N is a c profinite group with respect to the quotient topology. Proof. The group N is the intersection of finite-index abstract subgroups of G. Since G is strongly complete, each of these is open and hence closed in G. (cid:3) Corollary 2.3. A strongly complete profinite group does not admit any countably infinite residually finite abstract quotients. We remark that a strongly complete profinite group can have countably infinite abstractquotients. Indeed, weconcludethissectionwithasimpleargument, show- ing that the procyclic group Z has abstract quotients isomorphic to Q. Again, p this makes implicitly use of the axiom of choice. Proposition 2.4. The procyclic group Z maps non-continuously onto Q. p Proof. Clearly, Z is a spanning set for the Q-vector space Q . Thus there exists p p an ordered Q-basis x = 1, x , ..., x , ... for Q consisting of p-adic integers; 0 1 ω p here ω denotes the first infinite ordinal number. Set y = x = 1, y = x −p−i for 1 ≤ i < ω and y = x for λ ≥ ω. 0 0 i i λ λ The Q-vector space Q decomposes into a direct sum Q⊕W = Q , where W has p p Q-basis y ,...,y ,..., inducing a natural surjection η: Q → Q with kernel W. 1 ω p By construction, Z +W = Q so that Z /(Z ∩W) ∼= (Z +W)/W ∼= Q. (cid:3) p p p p p ABSTRACT QUOTIENTS OF PROFINITE GROUPS 5 To some extent this basic example is rather typical; cf. Theorems 1.2 and 1.3. §2.2. We now turn our attention to finitely generated profinite groups. Around 1970 Serre discovered that finitely generated pro-p groups are strongly complete, and he asked whether this is true for arbitrary finitely generated profinite groups. Here and in the following, p denotes a prime. Theorem 2.5 (Serre). Finitely generated pro-p groups are strongly complete. It is instructive to recall a proof of this special case of Theorem 1.1, which we base on two auxiliary lemmata. Lemma 2.6. Let G be a finitely generated pro-p group and H ≤ G a finite-index f subgroup. Then |G : H| is a power of p. Proof. Indeed, replacing H by its core in G, viz., the subgroup Hg E G, we Tg∈G f may assume that H is normal in G. Thus G/H is a finite group of order m, say, and the set X = G{m} = {gm | g ∈ G} of all mth powers in G is contained in H. Being the image of the compact space G under the continuous map x 7→ xm, the set X is closed in the Hausdorff space G. Since G is a pro-p group, it has a base of neighbourhoods of 1 consisting of open normal subgroups N E G of p-power index. Hence G/N is a finite p-group o for every N E G. Let g ∈ G. Writing m = prm with p ∤ m, we conclude that o gpr ∈ XN for every N E G. Since X is closed ineG, this yieelds o gpr ∈ \ XN = X ⊆ H. NEoG Consequently, everyelementofG/H hasp-powerorder, and|G : H| = m = pr. (cid:3) For any subset X of a topological group G we denote by X the topological closure of X in G. Lemma 2.7. Let G be a finitely generated pro-p group. Then the abstract com- mutator subgroup [G,G] is closed in G, i.e., [G,G] = [G,G]. Proof. We use the following fact about finitely generated nilpotent groups that can easily be proved by induction on the nilpotency class: (†) if Γ = hγ ,...,γ i is a nilpotent group then 1 d [Γ,Γ] = {[x ,γ ]···[x ,γ ] | x ,...,x ∈ Γ}. 1 1 d d 1 d Suppose that G is topologically generated by a ,...,a , i.e., G = ha ,...a i. 1 d 1 d Being the image of the compact space G × ··· × G under the continuous map (x ,...,x ) 7→ [x ,a ]...[x ,a ], the set 1 d 1 1 d d X = {[x ,a ]...[x ,a ] | x ,...,x ∈ G} ⊆ [G,G] 1 1 d d 1 d is closed in the Hausdorff space G. Using (†), this yields [G,G] = \ [G,G]N = \ XN = X ⊆ [G,G], NEoG NEoG and thus [G,G] = [G,G]. (cid:3) The Frattini subgroup Φ(G) of a profinite group G is the intersection of all its maximal open subgroups. If G is a finitely generated pro-p group, then Φ(G) = Gp[G,G] is open in G. Here Gp = hgp | g ∈ Gi denotes the abstract subgroup generated by the set G{p} = {gp | g ∈ G} of all pth powers. 6 BENJAMIN KLOPSCH Corollary 2.8. Let G be a finitely generated pro-p group. Then Φ(G) is equal to Gp[G,G] = G{p}[G,G] = {xpy | x ∈ G and y ∈ [G,G]}. Proof. Indeed, Gp[G,G] = G{p}[G,G],becauseG/[G,G]isabelian. ByLemma2.7, [G,G] is closed in G. Being the image of the compact space G×[G,G] under the continuous map(x,y) 7→ xpy, thesetG{p}[G,G]isclosedintheHausdorff spaceG. Thus Φ(G) = Gp[G,G] = G{p}[G,G]. (cid:3) Proof of Theorem 2.5. Let H ≤ G. Replacing H by its core in G, we may assume f that H E G. Using Lemma 2.6, we argue by induction on |G : H| = pr. If r = 0 f then H = G is open in G. Now suppose that r ≥ 1. FromCorollary2.8we deduce thatM = HΦ(G)isaproper opensubgroupofG. Since M is a finitely generated pro-p group and H E M with |M : H| < |G : H|, f induction yields H ≤ M ≤ G. Thus H is open in G. (cid:3) o o For completeness, we record another consequence of the proof of Lemma 2.7 that can be regarded as a special case of Corollary 3.4 below. Corollary 2.9. Let G be a finitely generated pro-p group and N E G a normal abstract subgroup. If N[G,G] = G then N = G. Proof. Suppose that N[G,G] = G. Then NΦ(G) = G and N contains a set of topological generators a ,...,a of G. Arguing as in the proof of Lemma 2.7 we 1 d obtain [G,G] ⊆ N and thus N = N[G,G] = G. (cid:3) §2.3. There were several generalisations of Serre’s result to other classes of finitely generated profinite groups, most notably by Brian Hartley [6] dealing with poly- pronilpotent groups, by Consuelo Mart´ınez and Efim Zelmanov [9] and indepen- dentlybyJanSaxlandJohnS.Wilson[19], eachteamdealingwithdirectproducts of finite simple groups, and finally by Segal [20] dealing with prosoluble groups. The result of Mart´ınez and Zelmanov, respectively Saxl and Wilson, on pow- ers in non-abelian finite simple groups relies on the classification of finite simple groups and leads to a slightly more general theorem. A profinite group G is called semisimple if it is the direct product of non-abelian finite simple groups. Theorem 2.10 (Mart´ınez and Zelmanov [9]; Saxl and Wilson [19]). Let G = S be a semisimple profinite group, where each S is non-abelian finite simple. Qi∈N i i Then G is strongly complete if and only if there are only finitely many groups S i of each isomorphism type. In particular, whenever a semisimple profinite group G as in the theorem is finitely generated, there are only finitely many factors of each isomorphism type andthegroupisstronglycomplete. WedonotrecallthefullproofofTheorem2.10, but we explain how the implication ‘⇐’ can be derived from the following key ingredient proved in [9, 19]: (‡) for every n ∈ N there exists k ∈ N such that for every non- abelian finite simple group S whose exponent does not divide n, S = {xn···xn | x ,...,x ∈ S}. 1 k 1 k Supposed that G = S as in Theorem 2.10, with only finitely many groups Qi∈N i S of each isomorphism type, and let H ≤ G. Replacing H by its core we may i f assume that H E G. Put n = |G : H| and choose k ∈ N as in (‡). Then H f ABSTRACT QUOTIENTS OF PROFINITE GROUPS 7 contains X = {xn···xn | x ,...,x ∈ G} ⊇ S , where j ∈ N is such that 1 k 1 k Qi≥j i the exponent of S does not divide n for i ≥ j, and hence H is open in G. The i existence of the index j is guaranteed by the classification of finite simple groups. Observe that implicitly we have established the following corollary. Corollary 2.11. Let G be a semisimple profinite group and q ∈ N. Then Gq = hgq | g ∈ Gi, the abstract subgroup generated by all qth powers, is closed in G. §2.4. The groups [G,G] and Gq featuring in the discussion above are examples of verbal subgroups of G. We briefly summarise the approach of Nikolov and Segal that led to the original proof of Theorem 1.1 in [14]; the theorem is derived from a ‘uniformity result’ concerning verbal subgroups of finite groups. Let d ∈ N, and let w = w(X ,...,X ) be a group word, i.e., an element of 1 r the free group on r generators X ,...,X . A w-value in a group G is an element 1 r of the form w(x ,x ,...,x ) or w(x ,x ,...,x )−1 with x ,x ,...,x ∈ G. The 1 2 r 1 2 r 1 2 r verbal subgroup w(G) is the subgroup generated (algebraically, whether or not G is a topological group) by all w-values in G. The word w is d-locally finite if every d-generator group H satisfying w(H) = 1 is finite. Finally, a simple commutator oflengthn ≥ 2isawordoftheform[X ,...,X ], where [X ,X ] = X−1X−1X X 1 n 1 2 1 2 1 2 and[X ,...,X ] = [[X ,...,X ],X ] for n > 2. It is well-known that the verbal 1 n 1 n−1 n subgroup corresponding to the simple commutator of length n is the nth term of the lower central series. Suppose that w is d-locally finite or that w is a simple commutator. In [14], Nikolov and Segal prove that there exists f = f(w,d) ∈ N such that in every d-generator finite group G every element of the verbal subgroup w(G) is equal to a product of f w-values in G. This ‘quantitative’ statement about (families of) finite groups translates into the following ‘qualitative’ statement about profinite groups. If w is d-locally finite, then in every d-generator profinite group G, the verbal subgroup w(G) is open in G. From this one easily deduces Theorem 1.1. Similarly, considering simple commutators Nikolov and Segal prove that each term of the lower central series of a finitely generated profinite group G is closed. By a variation of the same method and by appealing to Zelmanov’s celebrated solution to the restricted Burnside problem, one establishes the following result. Theorem 2.12 (Nikolov, Segal [16]). Let G be a finitely generated profinite group. Then the subgroup Gq = hgq | g ∈ Gi, the abstract subgroup generated by qth powers, is open in G for every q ∈ N. The approach in [14] is based on quite technical results concerning products of ‘twisted commutators’ in finite quasisimple groups that are established in [15]. The proofs in [16] make use of the full machinery in [14]. Ultimately these results all rely on the classification of finite simple groups. Remark 2.13. An independent justification of Theorem 2.12 would immediately yield a new proof of Theorem 1.1. Furthermore, Andrei Jaikin-Zapirain has shown that the use of the positive solution of the restricted Burnside problem in proving Theorem 2.12 is to some extent inevitable; see [7, Section 5.1]. We refer to the survey article [25] for a more thorough discussion of the back- ground to Serre’s problem and further information on finite-index subgroups and verbal subgroups in profinite groups. A comprehensive account of verbal width in groups is given in [21]. 8 BENJAMIN KLOPSCH 3. Nikolov and Segal’s results on finite and profinite groups §3.1. In this section we discuss the new approach in [17]. The key theorem concerns normal subgroups in finite groups. A finite group H is almost-simple if S E H ≤ Aut(S) for some non-abelian finite simple group S. For a finite group Γ, let d(Γ) denote the minimal number of generators of Γ, write Γ′ = [Γ,Γ] for the commutator subgroup of Γ and set Γ0 = \{T E Γ | Γ/T almost-simple} = \{CΓ(M) | M a non-abelian simple chief factor}, where the intersection over an empty set is naturally interpreted as Γ. To see that the two descriptions of Γ agree, recall that the chief factors of Γ arise as the 0 minimal normal subgroups of arbitrary quotients of Γ. Further, we remark that Γ/Γ is semisimple-by-(soluble of derived length at most 3), because the outer 0 automorphism group of any simple group is soluble of derived length at most 3. This strong form of the Schreier conjecture is a consequence of the classification of finite simple groups. For X ⊆ Γ and f ∈ N we write X∗f = {x ···x | x ,...,x ∈ X}. 1 f 1 f Theorem 3.1 (Nikolov, Segal [17]). Let Γ be a finite group and {y ,...,y } ⊆ Γ 1 r a symmetric subset, i.e., a subset that is closed under taking inverses. Let H E Γ. (1) If H ⊆ Γ and Hhy ,...,y i = Γ′hy ,...,y i = Γ then 0 1 r 1 r h[h,g] | h ∈ H,g ∈ Γi = {[h ,y ]···[h ,y ] | h ,...,h ∈ H}∗f , 1 1 r r 1 r where f = f(r,d(Γ)) = O(r6d(Γ)6). (2) If Γ = hy ,...,y i then the conclusion in (1) holds without assuming H ⊆ Γ 1 r 0 and with better bounds on f. While the proof of Theorem 3.1 is rather involved, the basic underlying idea is simple to sketch. Suppose that Γ = hg ,...,g i is a finite group and M a non- 1 r central chief factor. Then the set [M,g ] = {[m,g ] | m ∈ M} must be ‘relatively i i r large’ for at least one generator g . Hence [M,g ] is ‘relatively large’. In order i Qi=1 i to transform this observation into a rigorous proof one employs a combinatorial principle, discovered by Timothy Gowers in the context of product-free sets of quasirandom groups and adapted by Nikolay Nikolov and La´szlo´ Pyber to obtain product decompositions in finite simple groups; cf. [5, 13]. Informally speaking, to show that a finite group is equal to a product of some of its subsets, it suffices to know that the cardinalities of these subsets are ‘sufficiently large’. A precise statement of the result used in the proof of Theorem 3.1 is the following. Theorem 3.2 ([1, Corollary 2.6]). Let Γ be a finite group, and let ℓ(Γ) denote the minimal dimension of a non-trivial R-linear representation of Γ. If X ,...,X ⊆ Γ, for t ≥ 3, satisfy 1 t t Y|Xi| ≥ |Γ|tℓ(Γ)2−t, i=1 then X ···X = {x ···x | x ∈ X for 1 ≤ i ≤ t} = Γ. 1 t 1 t i i Ashort, butinformativesummaryoftheproofofTheorem3.1,basedonproduct decompositions, can be found in [12, §10]. ABSTRACT QUOTIENTS OF PROFINITE GROUPS 9 §3.2. By standard compactness arguments, Theorem 3.1 yields a corresponding result for normal subgroups of finitely generated profinite groups. For a profinite groupG, let d(G) denote the minimal number of topological generators of G, write G′ = [G,G] for the abstract commutator subgroup of G and set G0 = \{T Eo G | G/T almost-simple}, where the intersection over an empty set is naturally interpreted as G. As in the finite case, G/G is semisimple-by-(soluble of derived length at most 3). For 0 X ⊆ G and f ∈ N we write X∗f = {x ···x | x ,...,x ∈ X} as before, and the 1 f 1 f topological closure of X in G is denoted by X. Theorem 3.3 (Nikolov, Segal [17]). Let G be a profinite group and {y ,...,y } ⊆ 1 r G a symmetric subset. Let H E G be a closed normal subgroup. c (1) If H ⊆ G and Hhy ,...,y i = G′hy ,...,y i = G then 0 1 r 1 r h[h,g] | h ∈ H,g ∈ Gi = {[h ,y ]···[h ,y ] | h ,...,h ∈ H}∗f , 1 1 r r 1 r where f = f(r,d(G)) = O(r6d(G)6). (2) If y ,...,y topologically generate G then the conclusion in (1) holds without 1 r assuming H ⊆ G and better bounds on f. 0 Proof. We indicate how to prove (1). The inclusion ‘⊇’ is clear. The inclusion ‘⊆’ holds modulo every open normal subgroup N E G, by Theorem 3.1. Consider o the set on the right-hand side, call it Y. Being the image of the compact space H ×···×H, with rf factors, under a continuous map, the set Y is closed in the Hausdorff space G. Hence h[h,g] | h ∈ H,g ∈ Gi ⊆ \ YN = Y. (cid:3) NEoG In particular, the theorem shows that, if G is a finitely generated profinite group and H E G a closed normal subgroup, then the abstract subgroup c [H,G] = h[h,g] | h ∈ H,g ∈ Gi is closed. Thus G′ and more generally all terms γ (G) of the abstract lower cen- i tral series of G are closed; these consequences were already established in [14]. Theorem 1.4, stated in the introduction, generalises these results to more general compact Hausdorff groups. Furthermore, one obtains from Theorem 3.3 the following tool for studying abstract normal subgroups of a finitely generated profinite group G, reducing certain problems more or less to the abelian profinite group G/G′ or the profinite group G/G which is semisimple-by-(soluble of derived length at most 3). 0 Corollary 3.4 (Nikolov, Segal [17]). Let G be a finitely generated profinite group and N E G a normal abstract subgroup. If NG′ = NG = G then N = G. 0 Proof. Suppose that NG′ = NG = G, and let d = d(G) be the minimal number 0 of topological generators of G. Then there exist y ,...,y ∈ N such that 1 2d G hy ,...,y i = G′hy ,...,y i = G. 0 1 2d 1 2d Applying Theorem 3.3, with H = G , we obtain 0 [G ,G] ⊆ h[G ,y ]∪[G ,y−1] | 1 ≤ i ≤ 2di ⊆ N, 0 0 i 0 i hence G = NG′ = N[NG ,G] = N. (cid:3) 0 10 BENJAMIN KLOPSCH Using Corollaries 3.4 and 2.11, it is not difficult to derive Theorem 1.1. Proof of Theorem 1.1. Let H ≤ G be a finite-index subgroup of the finitely gen- f erated profinite group G. Then its core N = Hg E G is contained in H, Tg∈G f and it suffices to prove that N is open in G. The topological closure N is open in G; in particular, N is a finitely generated profinite group and without loss of generality we may assume that N = G. Assume for a contradiction that N (cid:8) G. Using Corollary 3.4, we deduce that (1) NG′ (cid:8) G or NG (cid:8) G. 0 Setting q = |G : N|, we know that the abstract subgroup Gq = hgq | g ∈ Gi generated by all qth powers is contained in N. By Theorem 3.3, the abstract commutator subgroup G′ = [G,G] is closed, thus the subgroup G′Gq is closed in G. Hence G/G′Gq, being a finitely generated abelian profinite group of finite exponent, is finite and discrete. As NG′/G′Gq is dense in G/G′Gq, we deduce that NG′ = NG′Gq = G. It suffices to prove that NG = G in order to obtain a contradiction to (1). 0 Factoring out by G , we may assume without loss of generality that G = 1. Then 0 0 G has a semisimple subgroup T E G such that G/T is soluble. From G = NG′ c we see that G/NT is perfect and soluble, and we deduce that (2) NT = G. Corollary 2.11 shows that Tq ≤ T. Factoring out by Tq (which is contained c in N) we may assume without loss of generality that Tq = 1. The definition of G 0 shows that T is a product of non-abelian finite simple groups, each normal in G, of exponent dividing q. Using the classification of finite simple groups, one sees that the finite simple factors of T have uniformly bounded order. Thus G/C (T) G is finite. As T ∩C (T) = 1, we conclude that T is finite, thus T ∩N ≤ G. This G c shows that T = [T,G] = [T,N] ≤ [T,N] ≤ T ∩N, and (2) gives N = NT = G. (cid:3) §3.3. The methods developed in [17] lead to new consequences for abstract quo- tientsoffinitelygeneratedprofinitegroupsand, moregenerally, compactHausdorff topologicalgroups. Intheintroductionwestatedthreesuchresults: Theorems1.2, 1.3 and 1.4. We finish by indicating how Corollary 3.4 can be used to see that, for profinite groups, the assertion in Theorem 1.2 reduces to the following special case. Theorem 3.5. A finitely generated semisimple profinite group does not have countably infinite abstract images. Nikolov and Segal prove Theorem 3.5 via a complete description of the maximal normal abstract subgroups of a strongly complete semisimple profinite group; see [17, Theorem 5.12]. Their argument relies, among other things, on work of Martin Liebeck and Aner Shalev [11] on the diameters of non-abelian finite simple groups. Proof of ‘Theorem 3.5 implies Theorem 1.2 for profinite groups’. Fora contradic- tion, assume that Γ = G/N is an infinite finitely generated abstract image of a profinite group G. Replacing G by the closed subgroup generated by any finite set mapping onto a generating set of Γ, we may assume that G is topologically