On splitting of the normalizers of maximal tori in E7(q) and E8(q)
Alexey Galt and Alexey Staroletov
Corresponding author, Sobolev Institute of Mathematics, 4 Acad. Koptyug avenue, 630090 Novosibirsk Russia, [email protected]
Abstract
Let G be a finite group of Lie type E7 or E8 over Fq and W be the Weyl group of G. We describe all maximal tori T of G such that T has a complement in its algebraic normalizer N(G,T). Let T correspond to an element w of W. When T does not have a complement, we show that w has a lift to N(G,T) of order ∣w∣ in all considered groups, except the simply-connected group E7(q). In the latter case we describe the elements w that have a lift to N(G,T) of order ∣w∣.
keywords: finite group of Lie type, maximal torus, algebraic normalizer, Weyl group
1 Introduction
Let G be a simple connected linear algebraic group over an algebraic closure Fp of a finite field of positive characteristic p. Consider a Steinberg endomorphism σ and a maximal σ-stable torus T of G. It is well known that all maximal tori are conjugate in G and the quotient NG(T)/T is isomorphic to the Weyl group W of G.
The natural question is to describe groups G such that NG(T) splits over T.
A similar question can be formulated for finite groups of Lie type. More precisely,
let G be a finite group of Lie type, that is Op′(Gσ)⩽G⩽Gσ. Let T=T∩G be a maximal torus in G and N(G,T)=NG(T)∩G be the algebraic normalizer of T. Then the question is to describe groups G and their maximal tori T such that N(G,T) splits over T.
These questions were stated by J. Tits in [19]. In the case of algebraic groups it was solved independently in [1] and in [7, 8, 9, 10]. In the case of finite groups the problem was
studied for the groups of Lie types An,Bn,Cn,Dn, and E6 in [8, 9, 10, 11].
J. Adams and X. He [1] considered a related problem. Namely, it is natural to ask about the orders of lifts of w∈W to NG(T). They noticed that if d is the order of w then the minimal order of a lift of w is either d or 2d, but it can be a subtle question which holds. Clearly, if NG(T) splits over T then the minimal order is equal to d.
In the cases of Lie types E6, E7, and E8 they proved that the normalizer does not split and the minimal order of lifts of w is d if w belongs to so-called regular or elliptic conjugacy classes.
G. Lusztig [12] showed that every involution of the Weyl group of a split reductive group over Fq has a lift n such that the image of n under the Frobenius
map is equal to n−1.
In this paper we consider finite groups G of Lie types E7 and E8 over a finite field Fq of characteristic p. Let W be the Weyl group of type El, where 6⩽l⩽8, and Δ={r1,…,rl} be a fundamental system of a root system El.
We enumerate conjugacy classes and roots of W as in [6]. Denote by wi the element of W corresponding to the reflection in the hyperplane orthogonal to the i-th positive root ri.
For completeness, we first formulate the statement for groups of Lie type E6.
Following [6], we suppose that r14=r2+r4+r5 and r36=r1+2r2+2r3+3r4+2r5+r6. Denote by E6−(q) the finite group 2E6(q) and by E6+(q) the finite group E6(q) (in both cases groups can be adjoint or simply-connected).
As a consequence of [11, Theorem 1.1] we obtain the following.
Theorem 1.1**.**
Let G=E6ε(q) (adjoint or simply-connected), where ε∈{+,−},
and W be the Weyl group of G. Consider a maximal torus T of G corresponding to an element w of W. Then the following statements hold.
T* does not have a complement in N(G,T) if and only if either
q≡−ε1(mod4) and w is conjugate to w3w2w4w14
or q is odd and w is conjugate to one of the following elements:
1, w1, w1w2, w2w3w5, w1w3w4, w1w4w6w36, w1w4w6w3, w1w4w6w3w36;*
there exists a lift of w to N(G,T) of order ∣w∣.
According to [6], if l=7 then r16=r2+r4+r5 and r53=r1+2r2+2r3+3r4+2r5+r6.
Denote by U the subgroup of W generated by w1,w2,w3,w4,w5,w6.
Note that U is isomorphic to the Weyl group of type E6. Given a root system Φ, we denote by Φsc(q) and Φad(q) a simply-connected and adjoint group, respectively.
Theorem 1.2**.**
Let G=E7(q) (adjoint or simply-connected) with the Weyl group W and w0 be the central involution of W. Consider a maximal torus T of G corresponding to an element w of W. Then the following statements hold.
If G=E7ad(q) then w has a lift of order ∣w∣ to N(G,T).
Moreover, T does not have a complement in N(G,T) if and only if q is odd and w or ww0 is conjugate to one of the following:
1, w1, w1w2, w2w3w5, w1w3w4, w1w4w6w53, w1w4w6w3,
w3w2w4w16, w1w4w6w3w53, w3w2w4w16w7.
If G=E7sc(q) then T does not have a complement in N(G,T) if and only if q is odd. Moreover, w has a lift to N(G,T) of order ∣w∣ if and only if either q is even or w satisfies one of the following:
∣w∣* is divisible by 4;*
∣w∣* is odd;*
w* is conjugate in W to an element of U.*
Remark. In fact, it follows from Table LABEL:t:main:E7 that
the order of an element w∈W divides 4 if and only if w or ww0
is conjugate to one of the elements listed in the claim (i).
According to [6], if l=8 then r18=r2+r4+r5, r26=r2+r4+r5+r6, r46=r1+r2+r3+r4+r5+r6+r7, r69=r1+2r2+2r3+3r4+2r5+r6, r74=r2+r3+2r4+2r5+2r6+2r7+r8, and r120=2r1+3r2+4r3+6r4+5r5+4r6+3r7+2r8.
Theorem 1.3**.**
Let G=E8(q) with the Weyl group W and w0 be the central involution of W.
Consider a maximal torus T of G corresponding to an element w of W. Then the following statements hold.
T* does not have a complement in N(G,T) if and only if q is odd and w or ww0 is conjugate to one of the following: 1, w1, w1w2, w3w1, w2w3w5, w1w3w5, w1w3w4, w1w4w6w69, w1w2w3w5, w1w4w6w3,
w3w2w5w4, w3w2w4w18, w1w4w6w3w69, w2w5w3w4w6, w26w5w4w3w2, w1w4w6w3w7, w3w2w4w18w7, w46w3w5w1w4w6, w2w3w5w7, w74w3w2w5w4, w8w1w4w6w3, w1w2w3w6w8w7,
w1w4w3w7w6w8, w4w8w2w5w7w120, w2w3w4w8w7w18, w2w3w4w5w6w8, w2w4w5w6w7w8w120,
w2w3w4w7w120w8w18, or w2w3w4w7w120w18w8w74;*
the element w has a lift to N(G,T) of order ∣w∣.
We illustrate the results of these theorems in Tables LABEL:t:main:E7, LABEL:t:main:E8, and LABEL:t:main with some additional information about the maximal tori.
This paper is organized as follows. In Section 2 we recall notation and basic facts about algebraic groups. In Section 3 we prove auxiliary results and explain how we use MAGMA in the proofs. Section 4–6 are devoted to the proofs of the main results.
2 Notation and preliminary results
By q we always denote some power of a prime p, Fp is an algebraic closure of a finite field Fp of order p. The symmetric group on n elements is denoted by Sn and the cyclic group of order n by Zn. Following [5], we write xy=yxy−1 and [x,y]=yxy−1.
By G we denote a simple connected linear algebraic group over Fp
with the root system Φ of Lie type El, where l∈{6,7,8}. We assume that
Δ={r1,…,rl} is a fundamental system of Φ.
We use notation from [5], in particular, the definitions of elements xr(t),
nr(λ) (r∈Φ, t∈Fp, λ∈Fp∗). In contrast to [5], the Magma convention is that hr(λ)=nr(−1)nr(λ) and we will follow this definition. According to [5], G is generated by xr(t): G=⟨xr(t)\leavevmode ∣\leavevmode r∈Φ,t∈Fp⟩. The group
T=⟨hr(λ)\leavevmode ∣\leavevmode r∈Δ,λ∈Fp∗⟩ is a maximal torus of G and N=⟨T,nr\leavevmode ∣\leavevmode r∈Δ⟩, where nr=nr(1), is the normalizer of T in G [5, §7.1, 7.2]. By W we denote the Weyl group N/T and by π the natural homomorphism from N onto W. Throughout, we denote by σ
the classical Frobenius automorphism defined on the generators as follows:
[TABLE]
In particular, we have nrσ=nr and (hr(λ))σ=hr(λσ).
Define the action of σ on W in the natural way. Elements w1,w2∈W are called σ-conjugate if w1=(w−1)σw2w for some w in W.
The following statement holds for G=Gσ.
Proposition 2.1**.**
[3, Propositions 3.3.1, 3.3.3].
A torus Tg is σ-stable if and only if gσg−1∈N. The map Tg↦π(gσg−1) determines a bijection between the G-classes of σ-stable maximal tori of G and the σ-conjugacy classes of W.
It follows from Proposition 2.1 that the cyclic structure of a torus (Tg)σ in G and of the corresponding tori
in the sections of G is determined only by a σ-conjugacy class of the element π(gσg−1).
Proposition 2.2**.**
[2, Lemma 1.2].
Let n=gσg−1∈N. Then (Tg)σ=(Tσn)g, where n acts on T by conjugation.
Proposition 2.3**.**
[3, Proposition 3.3.6].
Let gσg−1∈N and π(gσg−1)=w. Then
[TABLE]
By Proposition 2.2, we have (Tg)σ=(Tσn)g and (NG(Tg))σ=(Ng)σ=(Nσn)g. Hence,
[TABLE]
Remark 2.4**.**
Let n and w be as in Propositions 2.2 and 2.3, respectively. Suppose that n1=g1σg1−1∈N and π(n1)=w. Since n and n1 act on T in the same way, we have
(Tg1)σ=(Tσn1)g1=(Tσn)g1 and (NG(Tg1))σ=(Ng1)σ=(Nσn1)g1.
Hence,
[TABLE]
Thus, (Tg)σ has a complement in its algebraic normalizer if and only if there exists a complement for Tσn in Nσn1 for some n1 with π(n1)=w. Similarly, if w has a lift to Nσn1 of order ∣w∣
then w has such a lift to N(G,T) as well.
Lemma 2.5**.**
Let w∈W and w0∈Z(W). Let T, T0 be maximal tori corresponding to w, ww0, respectively.
Then T has a complement in its algebraic normalizer if and only if T0 does.
Proof.
Assume that T=Tσn has a complement K in Nσn. Then K≃CW,σ(w) and w0∈CW,σ(w). Let n,n0 be preimages of w,w0 in K, respectively. For any n1∈K⩽Nσn we have [n0,n1]=1. It follows that (n1)σnn0=n1n0=n1 and n1∈Nσnn0. Hence, K⩽Nσnn0.
Since π(n)=w, we have π(K)=π(Tσn⋅K)=π(Nσn)=CW,σ(w). Therefore, π∣K:K→W is an embedding and K∩T=1. In particular, K∩Tσnn0=1.
Since Nσnn0/Tσnn0≃CW,σ(ww0)=CW,σ(w), we have K is a complement for Tσnn0 in Nσnn0.
∎
For simplicity of notation, we write hr for hr(−1). If r=ri then hi stands for hri and ni stands for nri. Every element H of T can be uniquely written in the form i=1∏lhri(λi) and we write briefly H=(λ1,λ2,…,λl).
The group T=⟨nr\leavevmode ∣\leavevmode r∈Δ⟩ is called the Tits group [1]. Denote H=T∩T. It is known
that H=⟨hr\leavevmode ∣\leavevmode r∈Δ⟩ and so H is an elementary abelian group such that T/H≃W.
Remark 2.6**.**
Observe that if p=2 then hr=1 for every r∈Δ, in particular, H=1 and T≃W. Moreover, the restriction of the homomorphism π to T is an isomorphism between T and W. Let T be a maximal torus corresponding to w∈W. Then n=π−1(w) is a lift of w to Nσn of the same order and π−1(CW(w)) is a complement for Tσn in Nσn. Therefore, Theorems 1.1-1.3 hold in this case by Remark 2.4.
Similarly to [5, Theorem 7.2.2], we have:
nsnrns−1=nws(r)(ηs,r),ηs,r=±1,
nshr(λ)ns−1=hws(r)(λ).
We choose values of ηr,s as follows.
Let r∈Φ and r=i=1∑lαiri. The sum of the coefficients i=1∑lαi is called the height of r. According to [15], we fix the following total ordering of positive roots: we write r≺s if either h(r)<h(s) or
h(r)=h(s) and the first nonzero coordinate of s−r is positive.
The table of positive roots with respect to this ordering can be found in [15].
Recall that a pair of positive roots (r,s) is called special if r+s∈Φ and r≺s.
A pair (r,s) is called extraspecial if it is special and for any special pair (r1,s1) such that r+s=r1+s1 one has r≼r1. Let Nr,s be the structure constants of the corresponding simple Lie algebra [5, Section 4.1]. Then the signs of Nr,s may be taken arbitrarily at the extraspecial pairs and then all other structure constants are uniquely determined [5, Proposition 4.2.2]. In our case we choose sgn(Nr,s)=+ for all extraspecial pairs (r,s).
Values of the structure constants for all pairs can be found in [15].
The numbers ηr,s are uniquely determined by the structure constants [5, Proposition 6.4.3].
3 Preliminaries: calculations
We use MAGMA [17] to calculate products of elements in N.
All calculations can be performed using online Magma Calculator [16] as well.
At the moment it uses Magma V2.25-5. We use the following preparatory commands:
L:=LieAlgebra("E7",Rationals());
R:=RootDatum(L);
B:=ChevalleyBasis(L);
By default, L has adjoint type. To switch data to simply-connected groups
one can replace the first line on the following:
[TABLE]
The following command produces the list of extraspecial pairs and signs of the corresponding structure constants:
x,y,h:=ChevalleyBasis(L);IsChevalleyBasis(L,R,x,y,h);
[⟨1,3,1⟩, ⟨1,10,1⟩, ⟨1,15,1⟩, ⟨1,17,1⟩, ⟨1,22,1⟩, ⟨1,24,1⟩,
⟨1,28,1⟩, ⟨1,29,1⟩, ⟨1,31,1⟩, ⟨1,35,1⟩, ⟨1,36,1⟩, ⟨1,40,1⟩,
⟨1,41,1⟩, ⟨1,45,1⟩, ⟨1,49,1⟩, ⟨1,62,1⟩, ⟨2,4,1⟩, ⟨2,10,1⟩, ⟨2,11,1⟩,
⟨2,17,1⟩, ⟨2,18,1⟩, ⟨2,24,1⟩, ⟨2,25,1⟩, ⟨2,31,1⟩, ⟨2,50,1⟩, ⟨2,54,1⟩,
⟨2,57,1⟩, ⟨2,59,1⟩, ⟨3,4,1⟩, ⟨3,11,1⟩, ⟨3,18,1⟩, ⟨3,25,1⟩,
⟨3,32,1⟩, ⟨3,38,1⟩, ⟨3,43,1⟩, ⟨3,44,1⟩, ⟨3,48,1⟩, ⟨3,52,1⟩, ⟨3,61,1⟩,
⟨4,5,1⟩, ⟨4,12,1⟩, ⟨4,19,1⟩, ⟨4,22,1⟩, ⟨4,29,1⟩, ⟨4,36,1⟩, ⟨4,46,1⟩,
⟨4,51,1⟩, ⟨4,55,1⟩, ⟨4,60,1⟩, ⟨5,6,1⟩, ⟨5,13,1⟩, ⟨5,35,1⟩,
⟨5,41,1⟩, ⟨5,57,1⟩, ⟨6,7,1⟩, ⟨6,45,1⟩].
Here, for example, the triple ⟨2,4,1⟩ means that the pair (r2,r4) is extraspecial and Nr2,r4=1.
It is straightforward to verify the defined above ordering gives the same set of extraspecial pairs.
Thus, calculations in MAGMA for N correspond to the ordering and structure constants defined in the previous section. The following commands construct elements ni and hi.
G:=GroupOfLieType(L);
n:=[elt⟨G\leavevmode ∣\leavevmode i⟩:i in [1..NumberOfPositiveRoots(R)]];
h:=[TorusTerm(G,i,−1):i in [1..NumberOfPositiveRoots(R)]];
To obtain the list of matrices of reflections one can use the following command:
w:=[Transpose(i) : i in ReflectionMatrices(R)];
Lemma 3.1**.**
[11, Lemma 1]*
Let g∈G and n=gσg−1∈N. Suppose that H∈T and u∈T. Then*
(i) Hu∈Nσn if and only if H=Hσn[n,u];
(ii) If H∈H then Hu∈Nσn if and only if [n,Hu]=1.
The following result is our main tool for calculations of powers and commutators of elements in N.
Lemma 3.2**.**
Let Φ be a simply laced root system, that is all roots in Φ have the same length.
Let l be the number of fundamental roots in Φ and k=∣Φ∣. Assume that n∈T and π(n)=w. Consider the matrix A=(aij)l×l of w in the basis r1,r2,…,rl and an element H=(λ1,λ2,…,λl) of T. Then the following statements hold.
(i) Hn=(λ1′,λ2′,…,λl′),
where λi′=λ1ai1⋅λ2ai2…⋅λlail with 1⩽i⩽l;
(ii) (Hn)m=(λ1′,λ2′,…,λl′)nm,
where m is a positive integer, λi′=λ1bi1⋅λ2bi2…⋅λlbil with 1⩽i⩽l, and bij are elements of the matrix t=0∑m−1At.
Proof.
The proof is similar to that of [11, Lemma 2], where this assertion was proved for Φ=E6. ∎
Since we often use Lemma 3.2, we illustrate its applying with the following example.
Example. Let Φ=E7, w=w1w2w3, and n=n1n2n3. Then it is easy to see that
w1(r1)=−r1, w1(r3)=r1+r2, w2(r2)=−r2, w2(r4)=r2+r4,
w3(r3)=−r3, w3(r1)=r1+r3, and w3(r4)=r3+r4.
Therefore, in this case the matrix A for w is the following.
[TABLE]
Let H=(λ1,λ2,λ3,λ4,λ5,λ6,λ7)∈T. Then by Lemma 3.2, we can use the rows of A to compute Hn, namely Hn=(λ3−1λ4,λ2−1λ4,λ1λ3−1λ4,λ4,λ5,λ6,λ7).
Now let B=A0+A+A2+A3+A4+A5. Then
[TABLE]
It is easy to see that n6=h2, so
[TABLE]
The following lemma is clear.
Lemma 3.3**.**
Suppose that T and N are subgroups of a group G such that T is abelian and
N≤NG(T). Let N1=H1u1,N2=H2u2, where H1,H2∈T and u1,u2∈N. Then
[TABLE]
In particular, if [u1,u2]=1 then
N1N2=N2N1⇔H1−1H1u2=H2−1H2u1.
4 The proof for type E7
In this section we prove Theorem 1.2. Since the case p=2 was considered in Remark 2.6,
we suppose that q is odd. We assume that G is a finite group of Lie type E7.
The extended Dynkin diagram of type E7 is the following.
-r0-1r12r33r44r53r62r221r7
In this case the Weyl group W is isomorphic to 2×O7(2) (in the notation of [14]). So W has the central involution w0. It is easy to verify that w0=w1w2w5w7w37w55w61. There are in total 60 conjugacy classes in W
and all of them are divided into pairs. Namely, for every w∈W elements w and ww0 belong to
different conjugacy classes. We list 30 of representatives of conjugacy classes in Table LABEL:t:main:E7 and the remaining 30 can be obtained by multiplying on w0. This information can be verified with the aid of GAP [18].
There are exactly 60 conjugacy classes of maximal tori in G. The cyclic structures of maximal tori
in groups of type E7 were described in [6]. We add this information in Table LABEL:t:main:E7. To see the structure of a torus that corresponds to ww0, where w∈W, one can substitute −q instead of q and then multiply by −1 in the structure of a torus that corresponds to w.
We divide the proof into three general cases. First, we consider the simply-connected case. Groups of adjoint type are considered in two subsections. One subsection is devoted to maximal tori that do not have complements in their algebraic normalizers and the other subsection is for the remaining maximal tori.
4.1 Simply-connected groups of type E7
First, we show that any maximal torus T of the group G=E7sc(q) does not have a complement in its algebraic normalizer N(G,T). Denote n0=n1n2n5n7n37n55n61. Using MAGMA, we
verify that n02=h2h5h7, [n0,ni]=1, where i=1,…,7, and hence [n0,hi]=1,
where i=1,…,7. Assume that T has a complement K in N(G,T). Since w0∈CW(w), there exists a preimage N0 of w0 in K such that N02=1. Then N0=Hn0, where H∈T. Since w0 maps r to −r
for every r∈Φ, Lemma 3.2 implies that Hn0=H−1 for every H∈T. Therefore, we have N02=HHn0n02=HH−1h2h5h7=h2h5h7, a contradiction.
Suppose that w is an element from the second column of Table 1. Then there exists a lift of w to N(G,T) of the same order. Examples of the lifts we list in the fourth column of Table 1.
In each case we present an element nw∈T for w∈W such that π(nw)=w.
Using MAGMA, we see that ∣nw∣=∣w∣ and hence nw is a lift of w to Nσnw.
Consider w such that ∣w∣=4k for an integer k.
Since the lift N in Table 1
for w lies in T and n0∈Z(T), we get (Nn0)4k=N4kn04k=(h2h5h7)2k=1. Therefore, Nn0 is a lift of ww0 of the same order.
Let ∣ww0∣=4k+2 for an integer k. Suppose that ww0 has a lift N to its algebraic normalizer of the same order.
Then N=Hnn0=(λ1,λ2,λ3,λ4,λ5,λ6,λ7)nn0 for some λi∈Fp and n∈T. Since n0 commutes with n,
we have (nn0)2i=n2i(h2h5h7)i and (nn0)2i+1=n2i+1n0(h2h5h7)i,
where i is an integer. Then
[TABLE]
Let w∈⟨w1,w2,w3,w4,w5,w6⟩. This is the case for the first 25 representatives in Table 1. By Lemma 3.2, we infer that Hni=(∗,∗,∗,∗,∗,∗,λ7) for every i>0 and n4k+2=(∗,∗,∗,∗,∗,∗,1).
Hence, N4k+2=(∗,∗,∗,∗,∗,∗,−1)=1; a contradiction.
In the remaining two cases (representatives 27 and 29 in Table 1) the order of w is odd, say ∣w∣=2k+1. Then (ww0)2k+1=w0 and ∣ww0∣=4k+2. If ww0 has a preimage N in N(G,T) of the same order then N2k+1 is a preimage of w0 of order two, which is impossible by the argument above.
4.2 Adjoint groups of type E7: non-splitting cases
Let G=E7ad(q).
Throughout this subsection we suppose that T is a maximal σ-stable torus of G and T is a maximal torus of G corresponding to a conjugacy class of w∈W.
We write H=(λ1,λ2,λ3,λ4,λ5,λ6,λ7) for an arbitrary element of T. This notation means that H=i=1∏7hri(λi).
Observe that Z(E7sc(q))=⟨h2h5h7⟩.
For simplicity, we identify hi with their images in G and
assume that h2h5h7=1, that is (1,−1,1,1,−1,1,−1)=(1,1,1,1,1,1,1). Recall that π is the natural homomorphism from N onto W.
The main tool to show that T does not have a complement in N(G,T) is the following assertion.
Lemma 4.1**.**
Let w∈W and CW(w)⩾⟨w2w5,w49,w63⟩≃Z2×Z2×Z2.
Suppose that n∈N such that π(n)=w and n2n5,n49,n63 lie in Nσn. Then Tσn does not have a complement in Nσn.
Proof.
Assume that Tσn has a complement K in Nσn.
Let N1,N2,N3 be preimages of w2w5, w49, and w63 in K, respectively. Then N1=H1n2n5, N2=H2n49, N3=H3n63, where
[TABLE]
are elements of Tσn.
Since w632=1, it is true that N32=1. Computations in MAGMA show
that n632=h3h5h7. By Lemma 3.2,
[TABLE]
Therefore, β12=εβ22 and β52=−εβ13,
where ε∈{1,−1}.
Since [w63,w2w5]=[w63,w49]=1, we infer that [N3,N1]=[N3,N2]=1.
Using MAGMA, we see that [n63,n2n5]=[n63,n49]=1. It follows from Lemma 3.3 that H3−1H3n2n5=H1−1H1n63 and
H3−1H3n49=H2−1H2n63.
By Lemma 3.2,
[TABLE]
[TABLE]
[TABLE]
So we have (1,β2−2β4,1,1,β4β5−2β6,1,1)=(μ1−2,μ1−2,μ1−3,μ1−4,μ1−3,μ1−2,μ1−1).
Whence μ1−3=μ1−4 and so μ1=1. Then β2−2β4=β4β5−2β6=1. Therefore, β52=β6β22.
Since β12=εβ22 and β52=−εβ13, we infer that β6=−β1.
The equality H3−1H3n49=H2−1H2n63 implies that
[TABLE]
Applying β1β6−1=−1 to the latter equality, we find that
[TABLE]
From the equalities of the first coordinates, we see that α12=1. Since on the right side the fifth and the seventh coordinates have different signs, we infer that α1−3=−α1−1; a contradiction with α12=1.
∎
The following lemma finishes the proof of the theorem for cases when T does not have complements.
Lemma 4.2**.**
Let w or w0w is one of the following:
1, w1, w1w2, w2w3w5, w1w3w4, w1w4w6w53, w1w4w6w3,
w3w2w4w16, w1w4w6w3w53, w3w2w4w16w7. Suppose that T
is a maximal torus that corresponds to the conjugacy class of w in W.
Then N(G,T) does not split over T and there exists a lift of w to N(G,T) of order ∣w∣.
Proof.
By Lemma 2.5, we can assume that w is an element from the list in the hypothesis.
Using GAP, we find that in each case there exists w′∈W which is conjugate to w
and such that w2w5, w49, w63∈CW(w′). Examples of such w′ for every w
are listed in Table LABEL:t:nonsplit_E7. The first column of this table contains the number of a torus according to Table LABEL:t:main:E7,
and the third column contains an example of w′ for w. The fourth column contains elements nw′∈T
such that [nw′,n2n5]=[nw′,n49]=[nw′,n63]=1 and π(nw′)=w′.
Then Lemma 3.1 yields n2n5, n49, and n63 belong to Nσnw′ in each case.
Now Lemma 4.1 implies that Nσnw′ does not split over Tσnw′. So N(G,T) does not split over T by Remark 2.4.
Finally, the fifth
column contains another preimage n′ of w′ in T such that ∣w′∣=∣n′∣. The latter equality can be verified in MAGMA. Thus, n′ is a lift of w′ to Nσn′ of the same order.
Then n′n0 is a lift of w′w0 and the lemma is proved.
∎
4.3 Adjoint groups of type E7: splitting cases
In this subsection, we show that maximal tori of G that are not considered in the previous subsection
have complements in their algebraic normalizers.
Recall that w0=w1w2w5w7w37w55w61 is the central involution of W and n0=n1n2n5n7n37n55n61. As above, we identify hi with their images in G,
in particular h2h5h7=1.
We divide the proof into two parts: first, we consider maximal tori that need a special treatment and
then we finish with remaining maximal tori using a common approach.
Lemma 4.3**.**
Suppose that w or ww0 is conjugate to one
of the following: w3w1, w1w3w5, w1w2w3w5, w2w5w3w4w6, w23w5w4w3w2,
w1w4w6w3w7, or w39w3w5w1w4w6. If T is a maximal torus corresponding to the conjugacy class of w
then N(G,T) splits over T.
Proof.
We consider each case separately.
Torus 4. In this case w=w3w1 and CW(w)=⟨ww0⟩×⟨w2,w50,w7,w6,w5⟩≃Z6×S6.
Here we use that
[TABLE]
and w2, w50, w7, w6, and w5 satisfy this set of relations.
Let α∈Fp such that α2=−1 and set
[TABLE]
[TABLE]
Let n=n3n1, N1=nn0, N2=H2n2, N3=H3n50, N4=H4n7, N5=H5n6, and N6=H6n5.
We claim that K=⟨N1,N2,N3,N4,N5,N6⟩ is a complement for Tσn in Nσn.
Since Hi−1Hn0=Hi−2=1 for every i∈{2…6}, Lemma 3.3 implies
that [n0,N2]=[n0,N3]=[n0,N4]=[n0,N5]=[n0,N6]=1. Using MAGMA, we see that N16=1 and [n3n1,n2]=[n3n1,n50]=[n3n1,n7]=[n3n1,n6]=[n3n1,n5]=1, so n2, n50, n7, n6, n5 belong to Nσn.
By Lemma 3.2, we have
Hn=(λ1−1λ3,λ2,λ1−1λ4,λ4,λ5,λ6,λ7).
Therefore, Hin=Hi for i∈{2,…,6}. This implies that Hiσn=(1,(−1)((q−1)/2),1,1,(−1)((q−1)/2),1,(−1)((q−1)/2))Hi=Hi, where i∈{2,…,6}, and hence Hi∈Tσn.
Moreover, Lemma 3.3 yields [n3n1,Ni]=1, where i∈{2,…,6}. So K≃⟨N1⟩×⟨N2,N3,N4,N5,N6⟩.
Now we prove that N22=(N2N3)3=[N2,N4]=[N2,N5]=[N2,N6]=1.
Since n22=h2, Lemma 3.2 implies that (Hn2)2=(λ12,−λ4,λ32,λ42,λ52,λ62,λ72). Therefore, N22=h2h5h7=1.
Calculations in MAGMA show that [n2,n7]=[n2,n6]=[n2,n5]=1. By Lemma 3.3,
to get that [N2,N4]=[N2,N5]=[N2,N6]=1 it is sufficient to verify that H4n2=H4, H5n2=H5, H6n2=H6 and H2=H2n7=H2n6=H2n5.
By Lemma 3.2, we find that
[TABLE]
[TABLE]
Applying these equations to H2, H4, H5, and H6, we infer that N2 commutes with N4, N5 and N6.
Finally, observe that N2N3=H2H3n2n2n50 and H2H3n2=H2(1,−α,1,1,−α,−1,α)=1. So (N2N3)3=(n2n50)3=1, as claimed.
Now we prove that N32=(N3N4)3=1 and [N3,N5]=[N3,N6]=1.
Using MAGMA, we see that n502=h1h2h4h6. Lemma 3.2 implies that
[TABLE]
Therefore, N32=h2h5h7=1. Calculations in MAGMA show that [n50,n6]=[n50,n5]=1.
From the above equations, we see that H3n6=H3 and H3n5=H3.
Lemma 3.2 implies that H−1Hn50=(t,t,t2,t3,t2,t,1), where t=λ2λ4−1λ7.
So H5−1H5n50=H6−1H6n50=1. By Lemma 3.3, we infer that N3 commutes with N5 and N6.
Observe that N3N4=H3H4n50n50n7. We know that H4=H5 and H5n50=H5, so H3H4n50=H3H4=h1h2h4h6h7. Using MAGMA, we see that (h1h2h4h6h7n50n7)3=1 and hence (N3N4)3=1.
Now we prove that N42=(N4N5)3=1 and [N4,N6]=1.
First, observe that N42=(H4n7)2=H4H4n7h7. By the above equation for Hn7, we infer that
H4n7=H4h7 and so N42=1. Since N4N5=H4H5n7n7n6 and H4H5n7=H4H4n7=h7, we have (N4N5)3=(h7n7n6)3=1. Calculations in MAGMA show that [n7,n5]=1, so [N4,N6]=1 is equivalent to
H4−1H4n5=H6−1H6n7. By the above equations for Hn7 and Hn6, we have
H4=H4n5 and H6=H6n7. Therefore, N4 and N6 commute.
Finally, we verify that N52=N62=(N5N6)3=1. Since N52=H5H5n6h6 and
N62=H6H6n5h5, we infer that N52=N62=1. Observe that N5N6=H5H6n6n6n5=H5(−1,α,1,−1,α,1,α)n6n5=h2h5h7n6n5=n6n5, so
(N5N6)3=(n6n5)3=1. Thus, ⟨N2,N3,N4,N5,N6⟩≃S6 and hence K≃CW(w).
Torus 6. In this case w=w1w3w5 is conjugate to w′=w1w2w3 and
[TABLE]
Let n=n1n2n3 and α∈Fp such that α2=−1. Using MAGMA, we see that [n,n5]=[n,n6]=[n,n7]=1 and n5,n6,n7∈Nσn.
Put
[TABLE]
By Lemma 3.2,
[TABLE]
Then H2σn=(−1,α,1,−1,α,−1,α)σ=H2 and H2∈Tσn.
It is easy to see that h6∈Tσn and hence H3,H4∈Tσn.
Put N2=H2n5,N3=H3n6,N4=H4n7.
We claim that ⟨N2,N3,N4⟩≃S4.
By Lemma 3.2, we have
[TABLE]
[TABLE]
Then N22=H2H2n5n52=h2h7h5=1. Similarly, we see that N32=N42=h2h5h7=1.
Moreover, we have
[TABLE]
[TABLE]
Calculations in MAGMA show that (h6n5n6)3=(h6n6n7)3=1.
Hence, (N2N3)3=(N3N4)3=1 and ⟨N2,N3,N4⟩≃S4.
Assume that q≡3(mod4). Let β∈Fp such that βq+1=−1,
[TABLE]
Then αq=−α and H1σn=(1,β−1,1,1,−α,−1,α)σ=(1,−β,1,1,α,−1,−α)=H1. Therefore, we have H1∈Tσn.
Since H0=H1(1,α,1,1,1,1,1) and (1,α,1,1,1,1,1)σn=(1,α−1,1,1,1,1,1,)σ=(1,α,1,1,1,1,1), we also get H0∈Tσn. Put N1=H1n and N0=H0n0.
We claim that K=⟨N1,N0,N2,N3,N4⟩ is a complement for Tσn.
Calculations in MAGMA show that n6=h2. By Lemma 3.2, we have
[TABLE]
Hence, N16=h2h5h7=1.
Using the above equations, we find that H1−1H1n5=H1−1H1n6=H1−1H1n7=1, H1−1H1N0=H1−2=(1,β−2,1,1,−1,1,−1).
Moreover, we see that H2−1H2n=H3−1H3n=H4−1H4n=1 and H0−1H0n=(1,−β−2,1,1,1,1,1).
Hence, by Lemma 3.3 we obtain [N1,N2]=[N1,N3]=[N1,N4]=[N1,N0]=1.
Now we see that H2−2=H3−2=H4−2=(1,−1,1,1,−1,1,−1)=1.
On the other hand, H0−1H0n5=H0−1H0n6=H0−1H0n7=1 and
hence Lemma 3.3 implies that [N0,N2]=[N0,N3]=[N0,N4]=1. So
K≃⟨N1⟩×⟨N0⟩×⟨N2,N3,N4⟩≃Z6×Z2×S4 is a complement for Tσn.
Assume that q≡1(mod4). Let α,δ∈Fp such that α2=−1 and δq−1=−1.
Observe that αq=α and δq=−δ. Put
[TABLE]
Then H1σn=(1,1,1,1,α,−1,−α)σ=H1 and so H1∈Tσn.
Now
[TABLE]
and hence H0∈Tσn.
We claim that K=⟨N1,N2,N3,N4,N0⟩ is a complement for Tσn.
As it was shown above, we have
(Hn)6=(λ42,−λ43,λ44,λ46,λ56,λ66,λ76).
Therefore, we find that N16=(1,−1,1,1,−1,1,−1)=1.
We see above that H2−1H2n=H3−1H3n=H4−1H4n=1.
Since H−1Hn=(λ1−1λ3−1λ4,λ2−2λ4,λ1λ3−2λ4,1,1,1,1), then
[TABLE]
Recall that
[TABLE]
Then H1−1H1n5=H1−1H1n6=H1−1H1n7=1 and
[TABLE]
Hence, by Lemma 3.3, we obtain [N1,N2]=[N1,N3]=[N1,N4]=[N1,N0]=1.
Furthermore, H0−1H0n5=H0−1H0n6=H0−1H0n7=1 and
by Lemma 3.3 we obtain [N0,N2]=[N0,N3]=[N0,N4]=1. Hence,
K≃⟨N1⟩×⟨N0⟩×⟨N2,N3,N4⟩≃Z6×Z2×S4 is a complement for Tσn.
Torus 9. In this case w=w1w2w3w5 and CW(w)=⟨w0w2⟩×⟨w7⟩×⟨w5,w58w59⟩≃Z6×Z2×D8.
Let n=h4n1n2n3n5 and α∈Fp such that α2=−1. Denote N1=n0n2, N2=H1n7,
N3=H1h6n5, and N4=H1h1h4n58n59, where H1=(−1,α,1,−1,−α,1,α). We claim that
K=⟨N1,N2,N3,N4⟩ is a complement for Tσn in Nσn.
Using MAGMA, we see that [n,n7]=[n,h6n5]=[n,h1h4n58n59]=1, so n7, h6n5, and h1h4n58n59
belong to Nσn. Now we verify that H1∈Tσn. By Lemma 3.2,
[TABLE]
Applying this to H1, we find that H1n=(−1,−α−1,1,−1,−(−α)−1,1,α)=H1.
Then H1σ=(−1,αq,1,−1,(−α)q,1,αq)=(1,αq−1,1,1,αq−1,1,αq−1)H1=H1 and hence H1∈Tσn. So N2, N3, and N4 belong to Nσn.
Calculations in MAGMA show that N16=1. Now we verify that [N1,N2]=1, [N1,N3]=1, and [N1,N4]=1.
Using MAGMA, we see that [N1,n7]=[N1,h6n5]=[N1,h1h4n58n59]=1. By Lemma 3.2,
[TABLE]
Therefore, H1−1H1N1=h2h5h7=1 and hence N1 commutes with N2, N3 and N4 by Lemma 3.3.
So K≃⟨N1⟩×⟨N2,N3,N4⟩
Now we prove that N22=1 and [N2,N3]=[N2,N4]=1. Since n72=h7, we have N22=H1H1n7h7. By Lemma 3.2,
Hn7=(λ1,λ2,λ3,λ4,λ5,λ6,λ6λ7−1) and so H1n7=h7H1. Therefore, N22=h2h5h7h72=1.
Using MAGMA, we see that [n7,h6n5]=h7 and [n7,h1h4n58n59]=1. By Lemma 3.2,
[TABLE]
[TABLE]
Applying to H1, we find that H1n5=H1 and H1n58n59=h2h5H1=h7H1.
We see above that H1n7=H1h7, so h7H1n5=H1n7=H1n58n59.
It follows from Lemma 3.3 that [N2,N3]=[N2,N4]=1.
Finally, we show that ⟨N3,N4⟩≃D8. Namely, we verify that N32=N42=(N3N4)4=1.
Using MAGMA, we see that (h6n5)2=1 and (h1h4n58n59)2=h7.
By Lemma 3.2,
[TABLE]
[TABLE]
Applying these equations to H1, we find that N32=N42=1.
Since
[TABLE]
we infer that (N3N4)4=(h6n5h1h4n58n59)4=1.
Thus, K≃Z6×Z2×D8≃CW(w)
and hence K is a required complement.
Torus 13. In this case w=w3w2w5w4 and CW(w)≃Z6×Z2×S4. Using GAP, we see that
[TABLE]
and w, w7, w23w24, w20w21 are elements of CW(w) that satisfy this set of relations.
Let α,β∈Fp such that α2=−1 and βq−1=(−1)(q+1)/2. Put n=n3n2n5n4, N1=H1n, N2=H2n7, N3=h4h6n23n24, and N4=h3h5n20n21, where H1=(−1,α,1,−1,−α,1,α) and H2=(1,α,−1,1,α,1,β).
We claim that K=⟨N1,N2,N3,N4⟩ is a complement for Tσn in Nσn.
First, we verify that H1 and H2 belong to Tσn. By Lemma 3.2,
[TABLE]
Therefore, H1n=H1 and H2n=(1,−α,−1,1,−α,1,β).
Then
[TABLE]
and H2nσ=(1,α(−1)(q+1)/2,−1,1,α(−1)(q+1)/2,1,βq).
Since βq=β(−1)(q+1)/2, we infer that H2nσ=H2 and hence H1, H2
belong to Tσn.
Calculations in MAGMA show that [n,n7]=[n,N3]=[n,N4]=1, and hence N1, N2, N3, and N4 lie in Nσn. Since n6=1, Lemma 3.2 implies that
(Hn)6=(λ16,λ13λ63,λ16λ63,λ16λ66,λ13λ66,λ66,λ76). So N16=(1,−1,1,1,−1,1,−1)=1.
Now we prove that N1 commutes with N2, N3, and N4. By Lemma 3.3, it suffices to verify that H1−1H1N3=H1−1H1N4=1 and H1−1H1n7=H2−1H2n. By Lemma 3.2, we see that
[TABLE]
[TABLE]
[TABLE]
Therefore, H1−1H1N3=1, H1−1H1N4=1, and H1−1H1n7=h7.
We know that H2n=(1,−α,−1,1,−α,1,β)=h2h5H2=h7H2 and hence H2−1H2n=H1−1H1n7.
Thus, K≃⟨N1⟩×⟨N2,N3,N4⟩.
Calculations in MAGMA show that N32=N42=(N3N4)3=1, so it remains to verify that N22=(N2N3)4=(N2N4)2=1.
Since n72=h7, (n7N3)4=h2h3, and (n7N4)2=h7, Lemma 3.2 implies that
[TABLE]
[TABLE]
[TABLE]
We apply these equations to H2 and obtain N22=h2h5h7=1, (N2N3)4=1, and (N2N4)2=h2h5h7=1.
Thus, K≃CW(w) and hence K is a complement for Tσn.
Torus 19. In this case w=w2w5w3w4w6 and CW(w)=⟨w⟩×⟨w0⟩×⟨w63⟩≃Z8×Z2×Z2.
Put n=n2n5n3n4n6. Using MAGMA, we see that [n,n63]=1 and hence n63∈Nσn.
Let α be an element of Fp such that α2=−1.
Put H2=(−1,α,1,−1,−α,1,α).
By Lemma 3.2,
[TABLE]
Using this equality, we see that H2σn=H2σ=H2 and hence H2∈Tσn.
Let N2=H2n63. We claim that K=⟨n,n0,N1⟩ is a complement for Tσn in Nσn.
Calculations in MAGMA show that n8=1.
Since H2n=H2, we have H2−1H2n=1 and the equality nN2=N2n follows from Lemma 3.3.
So K≃⟨n⟩×⟨n0,N2⟩.
By Lemma 3.2, Hn63=(λ1−1,λ1−2λ2,λ1−3λ3,λ1−4λ4,λ1−3λ5,λ1−2λ6,λ1−1λ7).
Then N22=H2H2n63h2h3=1. Finally, we see that H2−1H2n0=H2−2=(1,−1,1,1,−1,1,−1)=1 and hence N2N0=N0N2 by Lemma 3.3.
Thus, K=⟨N1⟩×⟨N2⟩×⟨N0⟩≃Z8×Z2×Z2, as claimed.
Torus 20. In this case w=w23w5w4w3w2 and CW(w)=⟨w⟩×⟨w0⟩×⟨w63⟩≃Z12×Z2×Z2.
Let α be an element of Fp such that α2=−1 and n=n23n5n4n3n2.
Put N1=h1n and N2=H2n63, where H2=(−1,−α,1,1,α,−1,α).
By Lemma 3.2,
[TABLE]
Using this equality, we see that
H2n=h2h3H2 and hence H2nσh2h3=H2. Calculations in MAGMA show that [N1,n63]=h2h3.
Therefore, H2nσ[N1,n63]=H2 and N2∈Nσn by Lemma 3.1.
We claim that K=⟨N1,N2,n0⟩ is a complement for Tσn in Nσn.
Calculations in MAGMA show that N112=1. By Lemma 3.2,
[TABLE]
Then N22=H2H2n63h2h3=1.
Furthermore, we have H2−1H2N0=H2−2=(1,−1,1,1,−1,1,−1)=1 and hence N2n0=n0N2 by Lemma 3.3.
Thus, K≃Z12×Z2×Z2, as claimed.
Torus 26. In this case w=w1w4w6w3w7 and CW(w)=⟨w⟩×⟨w0⟩×⟨w59⟩≃Z12×Z2×Z2.
Let α be an element of Fp such that α2=−1.
Put n=n1n4n6n3n7, N1=h2n, and N2=H2h3n59, where H2=(1,α,1,1,−α,−1,α).
Calculations in MAGMA show that [N1,h3n59]=1 and hence h3n59∈Nσn.
By Lemma 3.2,
[TABLE]
Using this equality, we see that
H2σn=H2σ=H2. Thus, H2∈Nσn and N2∈Nσn.
We claim that K=⟨N1,n0,N2⟩ is a complement for Tσn in Nσn.
Calculations in MAGMA show that N112=1.
Since H2n=H2, we have H2−1H2n=1. Hence, the equality N1N2=N2N1 follows from Lemma 3.3.
By Lemma 3.2,
[TABLE]
Then N22=H2H2n59h1h4=1.
Furthermore, we see that H2−1H2n0=H2−2=(1,−1,1,1,−1,1,−1)=1 and by Lemma 3.3 we have N2n0=n0N2.
Thus, we conclude that K=⟨N1⟩×⟨N2⟩×⟨n0⟩≃Z12×Z2×Z2, as claimed.
Torus 30. In this case w=w39w3w5w1w4w6 and CW(w)=⟨w⟩×⟨w0⟩×⟨w53⟩≃Z8×Z2×Z2.
Put n=n39n3n5n1n4n6. Using MAGMA, we see that [n,n53]=1 and hence n53∈Nσn.
Let α be an element of Fp such that α2=−1.
Put H2=(1,α,−1,−1,−α,1,−α).
By Lemma 3.2,
[TABLE]
Using this equality, we see that H2σn=(1,−α,−1,−1,α,1,α)σ=H2 and hence H2∈Tσn.
Let N1=n and N2=H2n53. We claim that K=⟨N1,n0,N2⟩ is a complement for Tσn in Nσn.
Calculations in MAGMA show that N18=1.
Since H2n=H2, we have H2−1H2n=1 and the equality N1N2=N2N1 follows from Lemma 3.3. By Lemma 3.2,
[TABLE]
Then N22=H2H2n53h1h4h6=(1,α,−1,−1,−α,1,−α)(−1,α,−1,1,−α,−1,−α)h1h4h6=1.
Further, H2−1H2n0=H2−2=(1,−1,1,1,−1,1,−1)=1 and by Lemma 3.3 we have N2n0=n0N2.
Thus, K=⟨N1⟩×⟨N2⟩×⟨n0⟩≃Z8×Z2×Z2, as claimed.
∎
Lemma 4.4**.**
Suppose that w or ww0 is conjugate to one of the following:
w1w5w3w6, w1w4w3w2, w1w5w3w6w2, w1w4w5w3w53, w1w4w6w3w5,
w1w5w2w3w6w53, w1w4w6w3w5w53, w1w4w6w3w2w5,
w1w4w6w16w3w2w6, w1w4w16w3w2w40, w1w4w6w2w3w7, or
w1w4w6w3w5w7. Then T has a complement in its algebraic normalizer.
Proof.
Our strategy is the same in all cases.
For an element w, we choose n∈T such that π(n)=w and a set of relations that defines CW(w). Then we verify that
generators of a subgroup of Nσn satisfy this set of relations and generate a complement for Tσn. All data is listed in Table LABEL:t:split_E7.
As an example, we consider w=w1w5w3w6 that corresponds to Torus 10 in Table LABEL:t:main:E7.
In this case CW(w)=⟨ww0⟩×⟨w2,w53⟩×⟨w2w32w35w46,w2w28w42w43⟩≃Z6×S3×S3.
Using GAP, we see that CW(w) has the following presentation:
[TABLE]
Moreover, we verify that ww0, w2, w53, w2w32w35w46, w2w28w42w43 satisfy this set of relations.
Consider a=n0n, b=h53n2, c=h2n53, d=h1h6n2n32n35n46, and e=h1h3h6n2n28n42n43.
Then calculations in MAGMA show that a, b, c, d, e satisfy relations for CW(w) and hence K=⟨a,b,c,d,e⟩
is a homomorphic image of CW(w). On the other hand, we have
π(K)=CW(w) and hence K≃CW(w). Let n=n1n5n3n6. Finally, we verify that
[n,a]=[n,b]=[n,c]=[n,d]=[n,e]=1 and hence a,b,c,d,e∈Nσn by Lemma 3.1. Thus, K is a complement for Tσn in Nσn.
Other cases can be verified in the same way. The first column of Table LABEL:t:split_E7 contains numbers of tori in accordance with Table LABEL:t:main:E7. The second column for each w contains a set of relations S(w) that defines CW(w).
The third column contains examples of generators of a complement. All such generators lie in T.
Therefore, it is easy to verify in MAGMA that generators satisfy S(w). The natural preimage of w in T
is denote by n. In each case we choose an element x∈T and consider the groups Tσx and Nσx. Usually x=n but sometimes they differ. To verify that a generator y lies in Nσx, one can check that [x,y]=1 and apply Lemma 3.1. Elements from the third column
generate a complement for Tσx and hence N(G,T) splits over T
by Remark 2.4.
For convenience, we add all verified equations in [20].
∎
5 The proof for type E8
In this section, we prove Theorem 1.3. We assume that G is a finite group of Lie type E8 and q is odd.
The extended Dynkin diagram of E8 is the following.
r12r34r46r55r64r73r82r23-1-r0
In this case the Weyl group W is isomorphic to (2.O8+(2)):2 (in the notation of [14]). So W has the central involution w0. It is easy to verify that w0=w1w2w5w7w44w71w89w120.
There are in total 112 conjugacy classes in W.
In contrast to the type E7, it is possible that w∈W and ww0
are conjugate in W. There are in total 22 such conjugacy classes.
All the other classes are divided into pairs such that one class in a pair can be obtained from the other by multiplying on w0. This information can be verified in GAP.
We list 67 of representatives of conjugacy classes in Table LABEL:t:main:E8 and the remaining 45 representatives can be obtained by multiplying on w0.
Remark 5.1**.**
There are two comments concerning [6, Table I]. The representatives of the conjugacy classes of W in this table were obtained from [4].
- (i)
According to **[4, Table 11]**, a representative with number 50 has an admissible diagram D5×A2. The element u=w2w3w5w4w8w6w120 has such a diagram and we choose it for further computations. In **[6, Table I]** the representative u′=w2w3w5w4w8w6w35 does not have such a diagram, but u is conjugate to u′w0 in W. Hence, the representative u′ corresponds to a maximal torus with cyclic structure (q−1)(q3+1)(q4+1).
2. (ii)
According to **[4, Table 11]**, a representative with number 51 has an admissible diagram D5(a1)×A2. In **[6, Table I]** the representative v′=w26w5w4w3w2w7w8 contains a misprint, because v′ has the same admissible diagram as u′. The element v=w26w5w4w3w2w120w8 has a diagram D5(a1)×A2 and we choose it for further computations. Notice that the cyclic structure of the corresponding maximal torus in **[6, Table I]** is correct and equal to (q2+1)(q6−1).
Put n0=h2h5h7n1n2n5n7n44n71n89n120.
Calculations in MAGMA show that [n0,ni]=1 for i=1,…,8 and hence n0∈Z(T).
We divide our proof into two general cases. First, we consider maximal tori that do not have complements in their algebraic normalizers. If a maximal torus T does not have a complement and corresponds to the conjugacy class of element w∈W then we provide a lift of w of the same order. Secondly, we consider the remaining tori and present generators of their complements (up to conjugation).
5.1 Type E8: non-splitting cases
Throughout this subsection we suppose that T is a maximal torus corresponding to the conjugacy class of w∈W.
We write H=(λ1,λ2,λ3,λ4,λ5,λ6,λ7,λ8) for an arbitrary element of T. This notation means that H=i=1∏8hri(λi).
The main tool to show that T does not have a complement in N(G,T) is the following assertion similar to Lemma 4.1.
Lemma 5.2**.**
Let w∈W and CW(w)⩾⟨w2w5,w61,w97⟩≃Z2×Z2×Z2. Suppose that n∈N
such that π(n)=w and n2n5,n61,n97 lie in Nσn. Then Tσn does not have a complement in Nσn.
Proof.
Assume that Tσn has a complement K in Nσn.
Let N1,N2,N3 be preimages of w2w5, w61, and w97 in K, respectively. Then N1=H1n2n5, N2=H2n49, N3=H3n63, where
[TABLE]
are elements of Tσn.
Since w612=1, it is true that N22=1. Computations in MAGMA show
that n612=h2h3h7. By Lemma 3.2,
Hn612=(λ12,−λ1λ22λ6−1λ8,−λ1λ32λ6−1λ8,λ12λ42λ6−2λ82,λ12λ52λ6−2λ82,λ12λ82,−λ1λ6−1λ72λ8,λ82).
Therefore, α1α22α8=−α6, α12=α82=1, and α42=α52=α62.
Since [n2n5,n61]=1, it follows from Lemma 3.3 that H1−1H1n61=H2−1H2n2n5.
By Lemma 3.2,
[TABLE]
[TABLE]
Applying to H1 and H2, we see that
[TABLE]
Then
μ1μ6−1μ8=1, so α22=α4 and α52=α4α6.
Since α52=α42=α62, we infer that α4=α6.
Now the equality α1α22α8=−α6 implies that α1α8=−1.
Since [n97,n61]=1, it follows from Lemma 3.3 that H3−1H3n61=H2−1H2n97.
By Lemma 3.2,
[TABLE]
Therefore,
[TABLE]
Comparing the second and the third coordinates, we find that α1−2α82=α1−3α83 and hence α1−1α8=1.
Since α12=1, we arrive at a contradiction with α1α8=−1.
∎
Now we consider cases where Lemma 5.2 can be applied.
Lemma 5.3**.**
Let w or ww0 be an element from the second column of Table LABEL:t:nonsplit_E8.
If a maximal torus T corresponds to the conjugacy class of w then
T does not have a complement in N(G,T). Nevertheless, w has a lift to N(G,T) of order ∣w∣.
Proof.
By Lemma 2.5, we can assume that w is an element from the second column of Table LABEL:t:nonsplit_E8. Using GAP, we find that in each case there exists w′∈W which is conjugate to w
and such that w2w5, w61, w97∈CW(w′). Examples of such w′ for every w
are listed in Table LABEL:t:nonsplit_E8. The first column of this table contains the number of a torus according to Table LABEL:t:main:E8,
and the third column contains an example of w′ for w. The fourth column contains elements nw′∈T
such that [nw′,n2n5]=[nw′,n61]=[nw′,n97]=1 and π(nw′)=w′.
Then Lemma 3.1 yields n2n5, n61, and n97 belong to Nσnw′ in each case.
Now Lemma 5.2 implies that Nσnw′ does not split over Tσnw′. By Remark 2.4, T does not have a complement in N(G,T). Finally, the fifth column contains another preimage n′ of w′ in T such that ∣w′∣=∣n′∣. The latter equality can be verified in MAGMA. Thus, n′ is a required lift of w′ to Nσn′. Since n0∈Z(T), we have n′n0 is a required lift of w′w0 and the lemma is proved.
∎
The remaining cases are covered by the following lemma.
Lemma 5.4**.**
Let w or w0 be one of the following:
w1w4w3w7w6w8, w2w3w4w8w120w18, w2w3w4w7w120w8w18,
w2w3w4w7w120w18w8w74. Assume that T is a maximal torus
which corresponds to the conjugacy class of w. Then T does not have a complement
in N(G,T) and w has a lift to N(G,T) of order ∣w∣.
Proof.
We consider each case for w separately.
Torus 36. In this case w=w1w4w3w7w6w8 and
[TABLE]
where
[TABLE]
Put n=n1n4n3n7n6n8. Let N1=H1n1n4n3, N2=H2n7n6n8 and N3=H3n6n8n69n91 be preimages of w1w4w3, w7w6w8 and w6w8w69w91 in K,
where H1=(αi), H2=(βi), and H3=(μi).
Using MAGMA, we see that [n,n1n4n3]=[n,n7n6n8]=[n,n6n8n69n91]=1. By Lemma 3.1, we have n1n4n3,n7n6n8,n6n8n69n91∈Nσn and H1,H2,H3∈Tσn.
Since (w7w6w8)4=1, we have N24=1. Using MAGMA, we see that (n7n6n8)4=h6h8 and hence Lemma 3.2 implies that
[TABLE]
Therefore, β5=−1 and β14=1.
Since [n1n4n3,n7n6n8]=1, Lemma 3.3 yields
H1−1H1n7n6n8=H2−1H2n1n4n3. Using MAGMA, we see that
[TABLE]
[TABLE]
Therefore, we have β4=β1β3, β32=β1β4, and β1β2β5=β3β4.
Hence, β32=β1β4=β1(β1β3). So β3=β12 and β4=β1β3=β13. Since β5=−1 and β14=1, we have β1β2(−1)=β3β4=β15=β1. Therefore, β2=−1.
Since [n6n8n69n91,n7n6n8]=1, Lemma 3.3 yields H3−1H3n7n6n8=H2−1H2n6n8n69n91. Using MAGMA, we see that
[TABLE]
Therefore, β2−2β5=1; a contradiction with
β2=β5=−1.
Calculations in MAGMA show that (h5n)4=1. Thus, h5n is a required lift of w to Nσh5n.
Torus 41. In this case w=w2w3w4w8w120w18 and CW(w)=⟨w2w0⟩×⟨i,j,k⟩≃Z6×(SL2(3):Z4),
where i=w6w19w26, j=w4w13w40, and k=w1w2w4w6w48w51. Using GAP, one can verify that ⟨i,j,k⟩≃⟨a,b,c\leavevmode ∣\leavevmode a4=b4=c3=[a,b]=ca3b2cb3=1⟩
and i, j, k satisfy these relations.
Put n=n2n3n4n8n120n18 and a=h6n6n19n26.
Then, using MAGMA, we see that [n,a]=1 and hence a∈Nσn. Suppose that there exists
a complement K for Tσn. Let N1=H1n, N2=H2a, and N0=H0h2h3h5n0 be preimages of w, i and, w0 in K with H1=(μi), H0=(βi) and H2=(αi).
Since [n,a]=1, Lemma 3.2 implies that H1−1H1a=H2−1H2n. By Lemma 3.2, we find that
[TABLE]
[TABLE]
Therefore, we conclude that α7α83=α83=α82=1. So α7=α8=1.
Then
[TABLE]
Since the product of the second and the third coordinates on the left side equals the fourth coordinate,
we have (α2−1α3α4−1α5)(α1α4−1α6)=α32α4−2α6 and hence α1α5=α2α3. Moreover, the fourth and the fifth coordinates coincide on the left side, so α32α4−2α6=α2−1α3α5−1α6 and hence
α2α3=α42α5−1.
Using MAGMA, we see that a4=h2h3 and hence Lemma 3.2 implies that
[TABLE]
Therefore, we infer that −α1−1=α22α32α5−2α72, α12=1.
Since α7=1, we have −α1α52=α22α32. However, we know that
α1α5=α2α3 and hence α1=−1. Now from
α2α3=α42α5−1, we obtain α52=−α42.
Calculations in MAGMA show that [a,n0]=1. By Lemma 3.3, we infer that
H0−1H0a=H2−1H2n0. Then H2−1H2n0=(α1−2,α2−2,α3−2,α4−2,α5−2,α6−2,α7−2,α8−2). Applying the above equation for H−1Ha, we see that
H0−1H0a=H0−2=(∗,∗,∗,β1β2−1β3β5−1β6−1β72,β1β2−1β3β5−1β6−1β72,∗,∗,∗).
Since α42=−α52, we arrive at a contradiction.
Using MAGMA, we see that n12=1 and so n is a required lift of w to Nσn. Then nn0 is a required lift of ww0.
Torus 49. In this case w=w2w3w4w7w120w8w18 and
[TABLE]
Put n=h4n2n3n4n7n120n8n18. Observe that w15w119w8∈CW(w). Using MAGMA, we see that [h2n15n119n8,n]=1 and hence h2n15n119n8∈Nσn.
Suppose that exists a complement K for Tσn in Nσn. Then N0=H0n0, N1=H1n, N2=H2h2n15n119n8 belong to K for some H0, H1=(μi) and H2=(αi).
Since [n,h2n15n119n8]=1, Lemma 3.3 implies that
H2−1H2N1=H1−1H1N2. By Lemma 3.2,
[TABLE]
[TABLE]
Applying this to H2−1H2N1=H1−1H1N2, we see that the third and the sixth coordinates on the right side coincide, so α1α4−1α6α7−4α84=α7−4α84 and hence
α1α4−1α6=1. Moreover, we see that the square of the second coordinate equals the fourth coordinate, so (α2−1α3α4−1α5α7−3α83)2=α32α4−2α6α7−6α86
and hence α2−2α52=α6. Finally, since the fifth coordinate equals the product of the first and the second coordinates,
we infer that (α7−2α82)(α2−1α3α4−1α5α7−3α83)=α2−1α3α5−1α6α7−5α85. Whence, α4−1α52=α6.
Since [h2n15n119n8,n0]=1, we conclude that
[TABLE]
We use the same equalities of coordinates for the right side as in the previous
paragraph and get that α32=α62, α24=α42, and α12α22=α52.
We obtain above α52α2−2=α6, so α12=α6.
Calculations in MAGMA show that (h2n15n119n8)4=h2h5. By Lemma 3.2,
[TABLE]
Therefore, α24=−α63. On the other hand, we see above that α24=α42 and hence
α42=−α63. Squaring up the equation α1α4−1α6=1, we obtain α12(−α6−1)=1; a contradiction with the equation α12=α6.
Since (h6n)4=1 and hence h6n is a required lift of w in Nσh6n. Then h6nn0 is a required lift of ww0.
Torus 59. In this case w=w2w3w4w7w120w18w8w74 and
[TABLE]
Moreover, CW(w) is isomorphic to the following group:
[TABLE]
and the elements w1w99, w2w5, w4w17, w6w35, and w9w79 of CW(w) satisfy this set of relation.
Put n=h4n2n3n4n7n120n18n8n74. Using MAGMA, we see that [n,h3h5n1n99]=[n,h4h7n2n5]=[n,h3h8n9n79]=1. Therefore, h3h5n1n99,h4h7n2n5,h3h8n9n79∈Nσn. Let N1,N2,N3,N4 be preimages of w, a, b, and e in K, respectively. Then N1=H1n, N2=H2h3h5n1n99, N3=H3h4h7n2n5 and N4=H4h3h8n9n79, where
H1=(μi), H2=(αi), H3=(βi), and H4=(δi), are elements of Tσn.
Since b2=1, we have N32=1. Using MAGMA, we see that (h4h7n2n5)2=1. By Lemma 3.2,
[TABLE]
Therefore, β4=β6=1 and β12=β32=β72=β82=1.
Since [a,b]=1 and [h3h5n1n99,h4h7n2n5]=1, Lemma 3.3 implies that H3−1H3N2=H2−1H2N3.
By Lemma 3.2,
[TABLE]
[TABLE]
Applying to H2 and H3, we see that β1−2β32β4−1β6β7−1=β3β4−1β6β7−1=1. Since β4=β6=β12=β32=1,
we infer that β3=1 and hence β7=1 as well.
Since [w,b]=1, Lemma 3.2 implies that H3−1H3N1=H1−1H1N3.
By Lemma 3.2,
[TABLE]
Applying this to H3 and using the above equation for H−1HN3, we find that (∗,∗,∗,∗,∗,∗,∗,β1−1)=(∗,∗,∗,∗,∗,∗,∗,1) and hence β1=1.
Since [b,e]=1 and [h4h7n2n5,h3h8n9n79]=h2h8, Lemma 3.2 implies that
H3−1H3N4=H4−1H4N3h2h8.
By Lemma 3.2,
[TABLE]
Since β1=β3=β4=β7=1, we get H3−1H3N4=1. On the other hand, we apply the above equation for
H4−1H4N3 and find that h2h8=(1,δ2−2δ4,1,1,δ4δ5−2δ6,1,1,1);
a contradiction.
Using MAGMA, we see that n4=1, so n is a lift of w in Nσn of the same order. Then nn0 is a required lift of ww0.
∎
5.2 Type E8: splitting cases
We consider all cases in a similar manner. First, we need the following lemma for maximal tori of odd order.
Lemma 5.5**.**
Let w∈W and elements x1,x2,…,xm generate CW(w).
Suppose n∈N such that π(n)=w and there exist y1,y2,…,ym∈T
such that π(yi)=xi. If ∣Tσn∣ is odd then ⟨y1,y2,…,ym⟩ is a complement for Tσn in Nσn.
Proof.
Since x1,x2,…,xm generate CW(w),
there exists a set of defining relations S of CW(w) such that x1,x2,…,xm satisfy S.
Therefore, each relation of S holds for y1,y2,…,ym up to some elements of Tσn. Let h be any of such elements.
Since y1,y2,…,ym∈T, we infer that h∈T.
Since T∩Tσn⩽H, we have h∈Tσn∩H. However, H is an elementary abelian 2-group and hence
H∩Tσn=1. Thus, all relations of S hold in ⟨y1,y2,…,ym⟩,
and hence it is a complement for Tσn.
∎
Our strategy is similar in all cases.
For an element w, we find n∈T such that π(n)=w and a set of relations that defines CW(w). Then we present
elements in Nσn that satisfy this set of relations. All data is listed in Tables LABEL:t:split_E8.
As an example, we consider w=w1w5w3w6 that corresponds to Torus 10 in Table LABEL:t:main:E8.
In this case
[TABLE]
One can verify in GAP that CW(w) is isomorphic the group defined as follows:
[TABLE]
We replace the set of relations defining that x commutes with all other generators by R(x).
Now, if we take a=w, b=w0, c=w2, d=w69, e=w8, f=w120,
g=w2w37w40w57, i=w2w32w51w52, and j=w1w34w36w84 then
all these elements lie in CW(w) and satisfy the above relations for this group.
Finally, we put a=n1n5n3n6, b=n0, c=h69n2, d=h2n69, e=h120n8,
f=h8n120, g=h1h6n2n37n40n57, i=h1h3h6n2n32n51n52,
j=h2h8n1n34n36n84, and K=⟨a,b,c,d,e,f,g,i,j⟩. Then π(K)=CW(w).
Computations in MAGMA show that generators of K satisfy the set of relations above, so
K≃CW(w). On the other hand, a commutes with other generators of K, so
a,b,c,d,e,f,g lie in Nσa by Lemma 3.1.
Therefore, Nσa splits over Tσa.
Other cases can be verified in the same way. We divide information into two parts.
Table LABEL:t:split_E8 contains information for maximal tori of even order.
The first column of Table LABEL:t:split_E8 contains numbers of tori in accordance with Table LABEL:t:main:E8. The second column for each w contains a set of relations S(w) that defines CW(w). As above, we replace the set of relations defining that x commutes with all other generators by R(x).
The third column contains examples of generators of a complement. All such generators lie in T.
Therefore, it is easy to verify in MAGMA that the generators satisfy relations S(w). The natural preimage of w in T is denoted by n. In each case we choose an element x that defines the group Tσx. Usually x=n but sometimes they differ. To verify that a generator y lies in Nσx, one can check that [x,y]=1 and apply Lemma 3.1.
In Table LABEL:t:split_E8_odd, we list information for maximal tori of odd order.
The first column of this table contains numbers of maximal tori
in accordance with Table LABEL:t:main:E8. The second column contains examples of generators of a complement. All such generators lie in T. The natural preimage of w in T is denoted by n. Using MAGMA,
we see that for every generator y it is true that [n,y]=1 and hence in each case y∈Nσn by Lemma 3.1. Now
Lemma 5.5 implies that in all these cases listed elements generate corresponding complements.
For convenience, we add all verified equations in [21].
6 Results for all types
In this section we collect the results of Theorems 1.1-1.3 in one table. Information for types E7 and E8 is taken from Tables LABEL:t:main:E7 and LABEL:t:main:E8, respectively.
We use the symbol ’+’ if the corresponding torus has a complement in its algebraic normalizer, otherwise we put the symbol ’–’ into the cell.
To obtain results for 2E6(q) one can start with the maximal torus Tσ≃Zq+16 and then use the proofs from [11] replacing q by −q everywhere. So we take information for type E6ε
from [11, Table 1]. The sign ’±’ for the case 14 means that
the algebraic normalizers splits over the torus if and only if q≡ε1(mod4).
Acknowledgement
This research was supported by the Russian Science Foundation (project no. 14-21-00065).