KERNELS OF MAPS BETWEEN CLASSIFYING SPACES by Dietrich Notbohm Abstract. For homomorphisms between groups, one can divide out the kerne* *l to get an injection. Here, we develop a notion of kernels for maps between c* *lassifying spaces of compact Lie groups. We show that the kernel is a normal subgrou* *p in a modified sense and prove a generalization of a theorem of Quillen, namely* *, a map f:BG-! BH^pis injective, iff the induced map in mod-p cohomology is finit* *e. More- over, for compact connected Lie groups, every map f: BG -! BH^pfactors ov* *er a quotient of G in a modified sense and this factorisation is an injection. 1. Introduction. For a homomorphism ae: G - ! H between groups, we know that the kernel ker(ae) of ae is a normal subgroup of G, which gives rise to an exact sequence ker(ae) -! G -! G=ker(ae) : The induced homomorphism __ae:G=ker(ae) -! H is an injection. In this paper we will develop a analoguos concept for maps bet* *ween classifying spaces. To investigate the topological situation we pass to the p-adic completion. * *We also allow a more general target. Let G be a compact connected Lie group. A spa* *ce X is called BG-local if the evaluation induces an equivalence map(BG; X) ' X between the mapping space and X. The space X is called almost BG-local if the evaluation induces an equivalence map(BG; X)const' X between the component of the constant map const: BG- ! X and X. This is equivalent to the condition that the loop space X is BG-local. The situation, we are interested in, is the following f: BG -! X^pis a map, where G is a compact Lie group, where (S) X^pis p-complete and almost BZ=p-local and where H*(X; Fp) is of finite type. When talking about the kernel of a map f: BG- ! X^p as in (S), one has to look for elements g 2 G of order a power of p, such that f|B is homotopic to the constant map. Here, denotes the subgroup generated by g. This leads to Typeset by AM S-T* *EX 1 2 a definition of the prekernel, due to Ishiguro [7, 8], and of the kernel, which* * we explain now. For every compact Lie group G there exists a maximal p-toral subgroup SpG, unique up to conjugation, and every p-toral subgroup is subconjugated to SpG [9* *]. The group SpG is called the p-toral Sylow group of G. If G is finite, SpG is the usual p-Sylow group. For a compact Lie group G, we denote by TG the maximal torus, by NTG the normalizer of TG , and by WG the Weyl group. Then, SpG is the counterimage of SpWG in NTG . We also denote SpG by NpTG to indicate that SpG is the p-toral Sylow subgroup of NTG , too. TG is the component of the unit of NpTG = SpG. We define a subgroup Sp1 G SpG by the commutative diagram Tp1 --! Sp1 G --! ss0(SpG) ?? ? ? y ?y ?y TG --! SpG --! ss0(SpG) ; where Tp1 TG denotes the subgroup generated by all elements of order a power of p. For a map f: BG- ! X^p, we define the prekernel of f as preker(f) := {g 2 Sp1 G:f|B ' const} ; and the kernel of f as ker(f) := cl(preker(f)) ; where cl( ) denotes the closure in SpG or G. 1.1 Theorem. Let f: BG -! X^pbe a map as in (S). (1) preker(f) is a subgroup of Sp1 G. (2) ker(f) is a p-toral subgroup of SpG. (3) f|Bker(f) is nullhomotopic. As the proof in the next section shows, the theorem is true without the aasu* *ption that H*(X^p; Fp) is of finite type. These results lead to the following definitions: A map f: BG -! X^p as in (* *S) is called injective if ker(f) is the trivial group, and, following [4] or [13],* * monic or finite if H*(BG; Fp) is finitely generated over H*(X^p; Fp). In [13] is pro* *ved that the kernel of a homomorphism ae: G- ! H between two compact Lie groups is finite and of order coprime to p if and only if the induced map Bae:BG- ! BH^pis monic. The following statement also generalizes a result of Dwyer and Wilkerson [4, Proposition 4.4] 1.2 Theorem. Let f: BG- ! X^p be a map as in (S). Then, f is injective, if and only if f is monic. Let O(G) be the orbit category of G, and let Op(G) be the full subcategory, whose objects are homogenous spaces G=P , where P is a p-toral subgroup of SpG. Usually, Op(G) is defined to be the full subcategory of the spaces G=P , where P is any p-toral subgroups of G. Our definition is more convenient for our purpo* *se and gives a homotopy equivalent category. This follows because, up to conjugati* *on, every p-toral subgroup is contained in SpG [9]. For a subgroup K SpG and a p-toral subgroup P SpG, we define KP := K \ P . 3 1.3 Lemma. For a p-toral subgroup SpG the following conditions are equiv- alent: (1) For every x 2 of order a power of p and for every g 2 G, we have gxg-1 2 SpG if and only if gxg-1 2 . (2) For every pair P; P 0 SpG of p-toral subgroups and for every g 2 G, such that gP g-1 P 0, we have gP g-1 P0. Proof. Sp1 is a dense subset and contains only elements of order a power of p. If satisfies the first condition and if gP g-1 P 0, then g(Sp1 \ P )g-1 Sp1 \ P 0. Because P 0is closed, this is also true for , which is condition (* *2). Every element of order a power of p generates a finite p-group. Hence, (2) impl* *ies (1) obviously. A p-toral subgroup SpG is called Op(G)-normal, if satisfies one of the conditions of Lemma 1.3. 1.4 Proposition. Let f: BG- ! X^p be a map as in (S). Then, ker(f) SpG is Op(G)-normal. Proof. In [7] is shown that for two subgroups ; 0 SpG, which are conjugated in G, the restriction f|B is nullhomotopic if and only if f|B0 is nullhomoto* *pic. Hence, for every x 2 ker(f), conjugated elements are also contained in ker(f). For a map f: BG -! X^p as in (S), the kernel ker(f) SpG is not a normal subgroup of G in general. But, for G connected, ker(f) is the right invariant, * *which tells us, on which part the map f: BG- ! X^p is trivial. To avoid discussions a* *nd arguments about homotopy colimits, we study Op(G)-normal subgroups of NpTG . This investigation allows to prove the following theorem: 1.5 Theorem. Let G be a compact connected Lie group and let f: BG- ! X^pbe a map as in (S). Then, there exists a compact Lie group H and a commutative diagram f BG --! X^p ? fl q?y flfl __f BH^p --! X^p __ such that f: BH^p-! X^pis injective. Using a weak form of Theorem 1.1 (1), Ishiguro proved a similiar result for simple Lie groups. [7, 8]. We remark that we have to take the completion of BH. Moreover, as the proof shows, the group H in the theorem is not connected in general. This statement says that, at least, we can divide out by the kernel to q make the map injective. Moreover, the homotopy fiber of BG^p-! BH^p is closely related to ker(f). One might think of this homotopy fiber as being the kernel * *of f and of q as a surjection. In general, the homotopy fiber of q might not be t* *he completion of the classifying space of a compact Lie group. 4 The paper is organized as follows: In the next section we proof Theorem 1.1,* * the third section contains a proof of Theorem 1.2, and in the last section Op(G)-no* *rmal subgroups are discussed to prove Theorem 1.5. Completion are always meant in the sense of [2]. It is pleasure to thank K. Ishiguro for several discussions on this subject. 2. Prekernels and kernels. We start with the following observation: 2.1 Lemma. Let X^p be a p-complete almost BZ=p-local space. Then, X^p is almost BG-local for every compact Lie group G. Proof. A p-complete space X^pis always BZ=p0-local for any prime p0 6= p. If X^p is also BZ=p-local, then, by [11, x9], follows that X^p is Bss-local for any lo* *cally finite group ss. For every compact Lie group G, there exists a mod-p equivalence Bss- ! BG, where ss is locally finite [6]. So, X^p is BG-local. Because for a* * p- complete space X^p, the loop space (X^p) is also p-complete, the same arguments apply to show that every p-complete almost BZ=p-local space is almost BG-local for every compact Lie group G. Let ss be a finite group. For x 2 ss, we define (x) to be the smallest subgr* *oup of ss, which is normal in ss and contains x. This is welldefined because the inter* *section of two normal subgroups is also normal. 2.2 Lemma. If ss is a finite p-group and noncyclic, then, for every x 2 ss, (x* *) ss is a proper subgroup and (x) = . For a set S of elements of ss, we denote by the subgroup generated by the elements of S. Proof. The center of ss is nontrivial and contains Z=p as subgroup. Hence, the* *re q __ __ exists a central extension Z=p!- ss!- ss:= ss=Z=p. If ss ~=Z=pk is cyclic,* * all central group extensions are given by abelian groups. Thus , as a noncyclic gro* *up, ss ~=Z=p Z=pk and obviously satisfies the statement. If __ssis noncyclic, we can use an induction over the order of ss. Let __x:* *= q(x). Then, q((x)) (__x) 6= __ssby induction hypothesis. This shows that (x) 6= ss. To prove the second part of the statement, we observe that the group 0(x) := , generated by all the conjugates of x, is normal in ss. Hence, (x) 0(x). On the other hand, yxy-1 2 (x) for all y 2 ss, which shows that 0(x) (x). To prove Theorem 1.1 we need the following result, which may be found in [12* *], or [7]. q 2.3 Lemma. Let K!- G!- H be an exact sequence of groups, and let X be an almost BK-local space. Then, S q*: map(BH; X) --! f|BK 'constmap(BG; X)f is an equivalence. 5 2.4 Proposition. Let f: Bss -! X^pbe a map, where ss is a finite p-group and * *X^p a p-complete and almost BZ=p-local space. Let {x1; . .;.xr} be a set of generat* *ors. If f|B ' const for all i, then f is homotopically trivial. Proof. We pove the statement by an induction over the order of ss. If ss ~=Z=pk* * is cyclic, there is nothing to show because one of the elements must generate ss. * *If ss is noncyclic, by Lemma 2.1, there exists an exact sequence (x1) -! ss -! __ss:= ss=(x1) : (x1) is generated by elements of the form yx1y-1 , and f|B ' const. The order of (x1) is smaller than the order of ss. By induction hypothesis, f|B(x1) ' const. By Lemma 2.1, X^pSis B(x1)-local, and Lemma 2.2 establishes an equivalence map(B__ss; X^p) ' g|B(x1)'constmap(Bss; X^p)g. In particuliar* *, f __ __ factors over a map f: Bss-! X^p. The quotient __ssis generated by the elements __xi:= q(xi). The exact sequen* *ce (x1) \ -! -! <__xi> __ and another application of Lemma 2.2 show that f |B<__xi>'_const. Thus we can again apply the induction hypothesis, which shows that f ' const. This finishes the proof. Now, we are prepared to prove Theorem 1.1. Proof of Theorem 1.1. Let x; y 2 preker(f). We want to show that xy 2 preker(f) or, more generally, that f|B' const. Sp1 G is a locally finite p group. In particuliar, there exists a sequence 1 2 . . .r . .S.p1G S of finite p groups such that Sp1 G = rr [5]. Therefore, is a finite p-* *group, and, by the last proposition, f|B' const.SThis proves (1). Let 0r:= r \ preker(f). Then, preker(f) = r0r. By Proposition 2.3, f|B0r is nullhomotopic. Because X^p is almost BZ=p-local and hence almost B0r-local (Lemma 2.1), lim-1ss1(map(B0r; X^p)const ~= lim-1ss1(X^p) vanishes. The Milnor sequence for calculating the homotopy groups of inverse homotopy limits proves that f|Bpreker(f)is nullhomotopic. Lemma 2.5 below shows that Bpreker(f)- ! Bker(f) is a mod-p equivalence. For the p-complete space X^p, the map [B(ker(f); X^p] -! [B(preker(f); X^p] be- tween homotopy classes of maps is a bijection [2]. This implies that f|Bker(f)* * is nullhomotopic and proves part (3). ker(f) is the closure of preker(f) in SpG. Thus, the group of the components of ker(f) is a finite p-group, and ker(f) is a p-toral group which is part (2).* * This finishes the proof. 6 2.5 Lemma. Let f: BG- ! X^pbe a map as in (S). Then,the map B(preker(f)) -! B(ker(f)) is a mod-p equivalence. Proof. Let T (f) denote the component of the unit of ker(f), T1 (f) the interse* *ction of T (f) and Sp1 G, and let ss := ss0(ker(f). These groups fit into the commuta* *tive diagram T1 (f) --! preker(f) --! ss ?? ? fl y ?y flfl T (f) --! ker(f) --! ss : Both rows are exact. As a locally finite abelian p-group, T1 (f) ~= (Z=p1 )r x A, where A is a fi* *nite abelian p-group. Because the closure of T1 (f) is T (f), A is trivial, and T (* *f) ~= (S1)r. So, BT1 (f)- ! BT (f) is a mod-p equivalence. The Serre spectral sequence for mod-p cohomology for the fibrations in the diagram BT1 (f) --! Bpreker(f) --! Bss ?? ? fl y ?y flfl BT (f) --! Bker(f) --! Bss proves the statement. 3. Injective and monic maps. In this section we proof Theorem 1.2. Let f: BG- ! X^pbe a map as in (S). Let Ap(G) denote the Quillen category. The objects are given by elementary abelian * *p- subgroups and the morphisms by conjugation in G [13]. To get a finite category,* * we take only one object for every isomorphism class of objects, i.e. for every con* *jugacy class of a group. The Quillen map OE:H*(BG; Fp) -! lim- H*(BV ; Fp) V 2Ap(G) is H*(BG; Fp)-linear and an F -isomorphism; i.e. kernel and cokernel are nilpot* *ent [13, Theorem 7.2]. Let Y B := im(OE O f* ) lim- H*(BV ; Fp) H*(BV ; Fp) V 2Ap(G) V 2Ap(G) be the image of H*(X^p; Fp). 7 Proof of Theorem 1.2. First, we assume that f is injective and show that f is monic. For any elementary abelian p-subgroup V G, the restriction f|BV is also injective. By [4, Proposition 4.4], which is the analogous statement of Theorem 1.2 for elementary abelian p-groups, this implies that H*(BV ; Fp) is a finitel* *y gen- eratedQmodule over H*(X^p; Fp) and over B. Because Ap(G) is a finite category, V 2Ap(G) H*(BV ; Fp) is a finitely generated module over B and a finitely gen* *er- ated algebra over Fp. Therefore, B is also a finitely generated algebra over Fp* * [1, Proposition 7.8] and hence noetherian. This implies that lim- H*(BV ; Fp), as Q V 2Ap(G) a submodule of V 2Ap(G)H*(BV ; Fp), is a finitely generated module over B and over H*(X^p; Fp). If f is not monic, i.e the finitely generated Fp-algebra H*(BG; Fp) is not a* * finitely generated module over H*(X^p; Fp), there exists an element y 2 H*(BG; Fp) such that {yi: i 2 N} is a set of linearly independent elements over H*(X; Fp). By thePabove considerations, for r 2 N big enough, there exists a relation OE(yr) = r-1 i * ^ r P r-1 i i=0xiOE(y ) with xi 2 H (Xp ; Fp). That is that y - i=0 xiy is in the ke* *rnel of OE and hence nilpotent. Thus, for s 2 N big enough, r-1X r-1X 0 = (yr - xiyi)s = yrs - x0jyj i=0 j=0 for suitable x0i2 H*(X^p; Fp). This is a contradiction and proves that f is mon* *ic. Now, we assume that f is monic. Let Z=p G be a subgroup of G. Up to conjugation, Z=p is contained in SpG. By [13], H*(BZ=p; Fp) is finitley generat* *ed over H*(BG; Fp), and therefore, also over H*(X^p; Fp). That is that the map f BZ=p!- BG!- X^p is homotopically nontrivial. This implies that ker(f) = {1} and that f is injective. 4. Op(G)-normal subgroups. For every compact connected Lie group G, there exists a finite covering ff K --! eG --! G ; where Ge ~=GsQx T is a product of a simply connected Lie group Gs and a torus T . Gs ~= Gi is a product of simply connected simple Lie groups. K is a fin* *ite central subgroup of eG. The group eGis unique up to isomorphisms. Such coverings we call universal finite. ff 4.1 Lemma. Let K!- Ge!- G be an exact sequence of compact Lie groups, K finite and eGand G connected. Let f: BG!- X^pbe a map as in (S). (1) The sequence SpK!- preker(f O Bff)!- preker(f) is exact. (2) ker(f OBff)!- ker(f) is an epimorphism, and ker(f OBff) = Spff-1 ker(f* *). 8 Proof. Obviously, ff-1 (preker(f)) \ Sp1 eGis contained in preker(f O Bff). In * *parti- culiar, SpK preker(f O Bff).SLet := preker(f O Bff)=SpK preker(f). Then, by Lemma 2.3, map(B; X^p) ' g|BSpK 'constmap(Bpreker(f O Bff); X^p)g. This implies that f|B is homotopically trivial, and hence, that preker(f), which establishes the desired sequence of (1). To prove the second statement we first observe that in this case epimorphism are maintained under taking closures. The second part in (2) follows from the f* *acts that, as a p-toral group, ker(f O Bff) Spff-1 ker(f), and that f|BSpff-1ker(f)* *is homotopically trivial. ff 4.2 Lemma. Let K!- Ge!- G be an exact sequence of compact Lie groups, K finite and Ge and G connected. Let P SpG be a p-toral subgroup. Then, P is Op(G)-normal if and only if Spff-1 (P ) SpGe is Op(Ge)-normal. Proof. Let Qe := ff-1 (P ) and Pe := SpQe. The composition Pe!- Qe!- P is * *an epimorphism. This follows, because Pe and Qe have identical components of the unit, and because, passing to the components, the composition ss0(Pe) = Spss0(Q* *e)!- ss0(P ) is an epimorphism of finite groups. K eGis a central subgroup. The multiplication : (K \ eQ) x eP-! Qe fits into the pull back diagram (K \ eQ) x eP --! Qe ?? ? y ?yff eP -ff-!P : Thus, is an epimorphism, and Pe eQ is a normal subgroup. That is that Pe is the only p-toral Sylow subgroup of eQ. Every element x 2 P of order a power of p has a lift "x2 eP, also of order a* * power of p. Let "g2 eG and g := ff("g). Then, "g"x"g-12 SpGe if and only if gxg-1 2* * SpG, and, because eP eQis the only p-toral Sylow subgroup, "g"x"g-12 ePif and only if gxg-1 2 P . This proves the statement. Lemma 4.1 and Lemma 4.2 reduce the calculation of kernels and Op(G)-normal subgroups, G connected, to the case of products of simply connected Lie groups * *and tori. Let eG~= Gs x T be such a product. In order to describe Op(Ge)-normal sub- groups, we associate for every prime p a subgroup H(G; p) to each simply connec* *ted simple Lie group G . We define 8 > SU(2) o Z=2 if G = G2 and p = 3 : G else . Q Q We define H(Gs; p) := H(Gi; p) for a product Gs = Gi of simply connected simple Lie groups. Then, BH(Gs; p),! BG is a mod-p equivalence. If p = 3 and G = G2, this follows from the isomorphism H*(BG2; Z=p) ~=H*(BSU(2); Z=p)Z=2, and if (p; |WGs |) = 1 from the isomorphism H*(BGs; Z=p) ~=H*(BTGs ; Z=p)WGs . 9 Q 4.3 Proposition. Let eG= Gix T be a product of simply connected simple Lie groups Gi and a torus T . Let NpTGe be a Op(Ge)-normal subgroup. Then, we can split Gs = G0x G00such that ~=NpTG0 x ^ and ^ TG00x T . Moreover, ^ is normal in H(G00; p) x T and the image of ^ in G00is finite. We postpone the proof. This result enables us to prove Theorem 1.5. Proof of Theorem 1.5. Let f: BG -! X^pbe a map as in (S), and let K - ! Ge -! G be a universal finite covering, where Ge ~= Gs x T . By the last proposition a* *nd Proposition 1.3, Gs ~=G0x G00and ker(f O Bff) ~=NpTG0 x ^. Now, we define H = (H(G00; p)xT )=^ . The classical kernel of the projection G0xH(G00; p)xT - ! H * *is given by G0x ^, which contains K by construction. We get a commutative diagram Bi fOBff B(G0x H(G00; p) x T ) --! BGe ---! X^p ? ? fl Bf^f?y Bff?y flfl B_i f B(G0x H(G00; p) x T )=K --! BG --! X^p ? ? fl ^q?y q?y flfl __f BH^p ______BH^p --! X^p : _ Bi and Bi are mod-p equivalences. BH^pis p-complete, because ss1(BH) is a finite group [2]. This establishes the map q: BG- ! BH^p. The quotient (G0x ^)=K is a normal subgroup of (G0x H(G00; p) x T_)=K, and X^pis_almost_B(G0x ^)=K-local (Lemma 2.1)._Therefore, the map_f O Bi factors_over f:BH^p-! X^p(Lemma 2.3). Moreover, f O q ' f, because f O ^q' f O Bi. This proves the first half of Theo* *rem 1.5. fiO^ff __ By_Lemma_4.1 , ker(f O Bff O Bi) --! ker(f ) is an epimorphism. This shows that f is injective. In the rest of this section, we prove Proposition 4.3. Proof of Proposition 4.3. The subgroup T = \ TGe is invariant under the Weyl group action. In particuliar, WGe acts on the component e of the unit of , and H2(Be; Q) is a WGe-submodule of M H2(BTGe; Q) ~= H2(BGi; Q) H2(BT ; Q) : The first summands are irreducible. Thus, T \ TGi = TGi or the intersection is trivial. Let G0 be the product of all factors Gi with TGi , and G00the product of the other factors of eG. Let x 2 NpTG0 but x 62 TG0. Then x is conjugated to an ele* *ment 10 in TG0 and therefore, x 2 . This implies that NpTG0 . Moreover, NpTG0 is a normal subgroup of . In the commutative diagram i NpTG0 --! --! ^ := =NpTG0 flfl ? ? fl ?y ?y j NpTG0 --! NpTG0 x NpTG00x T --! NpTG00x T both rows are central extensions. The map j has a section, which establishes al* *so a section for i. Therefore, the upper sequence is the trivial extension, ~=NpTG0* * x ^, and ^ NpTG00x T . This is the first part of the statement. Because e NpTG0 x T , the image of ^ in G00is a finite group and Op(G00)- normal. We have to investigate finite Op(G)-normal subgroups of simply connected simple Lie groups. The following proposition finishes the proof. 4.4 Proposition. Let G be a simply connected simple Lie group, and NpTG a Op(G)-normal finite p-subgroup. (1) If p divides |WG |, and if G 6= G2 or p 6= 3, then is central in G. (2) If G = G2 and p = 3, then is central in SU(3) and hence normal in H(G2; 3). (3) If (p; |WG |) = 1, then is normal in H(G; p). Proof. If x 2 then txt-1 x-1 2 for all t 2 TG . Because is finite, all the commutators are trivial. Thus, x centralizes TG . Because CG (TG ) = TG , = T TG . If (p; |WG |) = 1, is normal in NTG = H(G; p). If p divides |WG |, we have to prove that is central in G or, for G = G2 and p = 3, central in SU(3). Let x 2 \ Z(G) and xpk = 1. Using this element, we construct an element \ TG , which gives a contradiction. This is done by a case by case checking and very m* *uch in the flavour of the proof of [7, Theorem 2']. Before we start, we make two observations. First, if (1) is true for two gr* *oups q __ G and H, then it is obviously true for the product G x H and, Second, if G!- _G_ is a covering of connected Lie groups, then (1) is true for G if and only if G satisfies condition (1). To see this, we consider a finite Op(G)-normal p-subgr* *oup NpTG . Then, 0 := <; SpZ(G)> is also a finite Op(G)-normal_p-subgroup. Because 0 = Sp(q-1 (q(0))), the group q(0)_is Op(G )-normal if and only if 0 is Op(G)-normal (Lemma 4.2), and q(fl0) G is central if and only if 0 G is central. Let G = SU(n), n p. Up to conjugation, x can be represented by a di- agonal matrix D = D(a1; . .;.an), where ai is a pk-th root of unity, and a1 6= a2. The element y = D(a2; a1; a3; . .;.an) is conjugate to x and thus, xy-1 = D(a1a-12; a-11a2; 1; . .;.1) 2 . Because a1a-12 6= 1, conjugates of xy-1 gene* *rate a subgroup of TG , which contains the maximal elementary abelian p-subgroup VG of TG . Every element of order p is conjugate to an element in VG . This give* *s a contradiction. For G = U(n), n p, the same proof works. 11 Let G = Sp(n), n p. Then, U(n) and SU(2)n are subgroups of maximal rank. Therefore, TSp(n) is central in U(n) and SU(2)n. But Z(U(n)) \ Z(SU(2)n) = Z=2 = Z(Sp(n)). By the above observation, the case of G = Spin(n), n 2p and n 5, can be reduced to the case of SO(n). Let n = 2k or n = 2k + 1. Then, U(k) SO(2k) SO(2k + 1) is a subgroup of maximal rank, and is central in U(k). For n 5, the only WSO(2k) or WSO(2k+1)-invariant subgroup of S1 = Z(U(k)) is Z=2. For SO(2k), Z=2 U(k) SO(2k) is central, and for SO(2k + 1), Z=2 U(k) SO(2k + 1) is conjugate to a subgroup in NTSO(2k+1) \ TSO(2k+1), which proves t* *he statement for SO(n). Let G be an exceptional Lie group and p a divisor of |WG |. In this case, we choose subgroups of maximal rank, given by the following list: G2 : p = 2; 3 H = SU(3) F4 : p = 2 H = SU(3) xZ=3 SU(3) p = 3 Spin(9) E6 : p = 2; 5 H = SU(3)3=Z=3 p = 3 H = SU(2) xZ=2 SU(6) E7 : p = 3; 5; 7 H = SU(8)=Z=2 p = 2 H = SU(3) xZ=3 SU(6) E8 : p = 3; 5; 7 H = SU(2) xZ=2 E7 p = 2 H = SU(9)=Z=3 : Beside the case G = G2 and p = 3, the inclusion H G always induces an isomorphism SpZ(H) ~= SpZ(G) between the p-Sylow subgroups of the centers. The data may be obtained from [9], where one can find a complete list of maximal subgroups of maximal rank of the exeptional Lie groups, and from [14]. Now, we can argue as follows: Let NpTG be an Op(G)-normal subgroup. Then, by induction over the rank, by the above observations, and by the already calculated cases, is central in H and hence, central in G. For G = G2 and p = 3, the argument only shows that is central in SU(3). That is = Z=3 or is the trivial group. In both cases, is normal in SU(3) o Z=2 = H(G2; 3). This finishes the last open case and the proof of the proposition. References. [1] M.F. Atiyah and I.G. MacDonald. Introduction to commutative algebra, Addiso* *n- Wesley (1969). [2] A. Bousfield and D. Kan. Homotopy limits, completion, and localisations, SL* *NM 304_, Springer Verlag (1972). [3] W.G. Dwyer and C.W. Wilkerson. Mapping spaces of nullhomotopic maps, Asterisque 191_(1990), pp. 97-108. [4] W.G. Dwyer and C.W. Wilkerson. A cohomological decomposition theorem, Preprint. [5] M. Feshbach. The Segal conjecture for compact Lie groups, Top. 26_(1987), p* *p. 1-20. [6] E. Friedlander and G.Mislin. Locally finite approximation s of compact Lie groups, I, Inv. Math. 83_(1986), pp. 425-436. [7] K.Ishiguro. Classifying spaces of compact simple Lie groups and p-tori, Pr* *oc. 12 of 1990 Barcelona Conf., SLNM 1509_, pp. 210-226. [8] K.Ishiguro. Retracts of classifying spaces of compact connected Lie groups, Preprint. [9] S. Jackowski, J.McClure, and R.Oliver. Homotopy classification of self maps* * of BG via G actions, Part I and II, Ann. Math. 135 (1992), 183-226, 227-270. [10] J. Lannes, Sur les espaces fonctionelles dont la source est la classifiant* * d'un p-groupe abelienelemementaires, preprint 1990. [11] H. Miller. The Sullivan conjecture on maps from classifying spaces, Ann.Ma* *th. 120_(1984), pp. 39-87. [12] D. Notbohm. Maps between classifying spaces and applications, Math. Gott. Heft 20_(1991). [13] D.Quillen. The spectrum of an equivariant cohomology ring I, Ann. Math. 94_ (1971), pp. 549-572. [14] J. Tits. Tabellen zu den einfachen Lie Gruppen und ihren Darstellungen, SL* *NM 40_, Springer Verlag 1967. Mathematisches Institut Bunssenstr. 3-5 3400 G"ottingen Germany