On the equivariant cohomology of
isotropy actions
Als Dissertation
dem Fachbereich Mathematik und Informatik
der PhilippsโUniversitยจ
at Marburg vorgelegt
von M. Sc. Sam Hagh Shenas Noshari
aus Hamburg
Erstgutacher:
Prof. Dr. Oliver Goertsches
Zweitgutachter:
Prof. Dr. AugustinโLiviu Mare
Einreichungstermin:
24. Juli 2018
Pr ยจ
ufungstermin:
26. September 2018
Hochschulkennziffer:
1180
Zusammenfassung
Sei๐บ eine kompakte, zusammenhยจangende Lie Gruppe und ๐พ โ ๐บ eine abgeschlossene Untergruppe. Wir zeigen, dass die Isotropiewirkung von๐พ auf ๐บ/๐พ ยจaquivariant formal ist und der Raum ๐บ/๐พ formal im Sinne rationaler Homotopietheorie, falls es sich bei ๐พ um die Identitยจatskomponente des Schnitts der Fixpunktmengen zweier verschiedener Involutionen auf๐บ handelt, ๐บ/๐พ also ein โค2ร โค2โsymmetrischer Raum ist. Ist๐พ die Identitยจats-komponente der Fixpunktmenge einer einzelnen Involution und๐ป โ ๐บ eine abgeschlossene, zusammenhยจangen-de Untergruppe, die๐พ enthยจalt, so zeigen wir, dass auch die Wirkung von ๐พ auf ๐บ/๐ป durch Linksmultiplikation ยจ
aquivariant formal ist. Letztere Aussage ist ยจaquivalent zum Hauptresultat in [6], wird hier aber mit anderen Mit-teln bewiesen, nยจamlich durch Angabe eines algebraischen Modells f ยจur die ยจaquivariante Kohomologie gewisser Wirkungen.
Abstract
Let๐บ be a compact connected Lie group and ๐พ โ ๐บ a closed subgroup. We show that the isotropy action of ๐พ on ๐บ/๐พ is equivariantly formal and that the space ๐บ/๐พ is formal in the sense of rational homotopy theory whenever๐พ is the identity component of the intersection of the fixed point sets of two distinct involutions on ๐บ, so that ๐บ/๐พ is a โค2ร โค2โsymmetric space. If๐พ is the identity component of the fixed point set of a single involution and๐ป โ ๐บ is a closed connected subgroup containing ๐พ, then we show that the action of ๐พ on ๐บ/๐ป by leftโmultiplication is equivariantly formal. The latter statement is equivalent to the main result of [6], but is proved by different means, namely by providing an algebraic model for the equivariant cohomology of certain actions.
Contents
Zusammenfassung iii
Abstract v
I. Introduction 1
1. Introduction and background 1
2. Previous results 3
II. โค2ร โค2โsymmetric spaces 5
1. โค2ร โค2โsymmetric spaces 5
2. Preliminaries 6
3. Automorphisms 8
4. Normal forms for strongly orthogonal roots 11
5. Cohomology of associated subalgebras 22
6. Equivariant and ordinary cohomology of simpleโค2ร โค2โsymmetric spaces 32
III. An algebraic model for the eqivariant cohomology of isotropy actions 43
1. gโactions 43
2. Constructinggโactions 46
3. Compatibility with existing actions 49
4. An exact sequence 50
5. Applications to smooth manifolds 52
Chapter I.
Introduction
1. Introduction and background
This thesis is concerned with๐บโspaces, that is, topological spaces together with a continuous (left) action of a fixed (smooth) Lie group๐บ, and a certain invariant associated with such spaces, their equivariant cohomology. To motivate its definition, consider the problem of assigning to a๐บโspace ๐ an invariant that gives the same answer on any๐บโspace isomorphic to ๐ but yet discerns as many distinct isomorphism classes of ๐บโspaces as possible. Perhaps among the easiest such invariants that one might come up with (apart from the isomorphism class of๐) is the cohomologyH(๐/๐บ) of the orbit space ๐/๐บ; of course, one might consider arbitrary coefficient groups, but here and thereafter we confine ourselves to singular real cohomology or to de Rham cohomology if the space under consideration happens to be a smooth manifold. In any case, it appears to be common understanding thatH(๐/๐บ) is a reasonable invariant if the ๐บโaction is free, but less wellโbehaved for actions with nonโtrivial isotropy. A frequently given example of an action justifying this last statement is the action of the circle๐1on the unit sphere๐2by rotation about a fixed axis. This action has exactly two fixed points, namely the poles of the rotation axis, and its orbit space is homeomorphic to the closed unit interval, hence has trivial cohomology. To overcome this difficulty one replaces๐ by what is now called the Borel construction and usually denoted ๐๐บ. Originally introduced in [2], this is the space๐๐บโถ= (๐ธ๐บ ร๐)/๐บ obtained from a contractible space ๐ธ๐บ on which ๐บ acts freely (from the right), such as the total space in the universal ๐บโbundle ๐ธ๐บ โ ๐ต๐บ over the classifying space๐ต๐บ. The action of ๐บ on ๐ธ๐บ ร ๐ is the diagonal action, induced by the assignment ๐.(๐, ๐ฅ) = (๐๐โ1, ๐๐ฅ) for ๐ โ ๐บ and (๐, ๐ฅ) โ ๐ธ๐บร๐, and the equivariant cohomology then is defined as H๐บ(๐) โถ= H(๐๐บ). Note that the ๐บโ
action on๐ธ๐บ ร ๐ is free. Another indication that H๐บ(๐) is a useful invariant is that it can actually be computed in many situations: quite generally, if๐บ acts locally freely on a space ๐, then the map ๐๐บ โ ๐/๐บ induced by the quotient map๐ โ ๐/๐บ yields an isomorphism H(๐/๐บ) โ H๐บ(๐), cf. [12, Section C.2.1]. On the other hand,H๐บ(โ ) satisfies the axioms of a generalized cohomology theory with morphisms replaced by ๐บโequivariant morphisms, so that, for example, an equivariant MayerโVietoris sequence is available. In very much the same way as the MayerโVietoris sequence can be used to compute the ordinary cohomology of spheres, its equivariant counterpart can be utilized to compute the๐1โequivariant cohomology of the action on๐2considered above, e. g. by means of the open cover consisting of the two open sets that one obtains by removing one of the poles of the rotation axis at a time. The conclusion now is thatH
๐1(๐2) = H(๐ต๐1) โ H(๐ต๐1) in nonโzero degrees, because for any Lie group๐บ the equivariant cohomology of a single point is given by H
๐บ(โ) = H(๐ต๐บ) and ๐1
acts freely on๐2outside its fixed point set.
The previous eaxmple can be written more concisely asH
๐1(๐2) = H(๐ต๐1)โH(๐2) (recall that the classifying space of๐1 isโ๐โ, whose cohomology ring is a polynomial algebra in one variable of degree2), and if one considersH
๐1(๐2) as a H(๐ต๐1)โmodule via the morphism of rings H๐1(โ) โ H๐1(๐2) induced by the constant map๐2โ {โ}, then this equality is even valid as H(๐ต๐1)โmodules, showing that the ๐1action on๐2is in fact equivariantly formal. This name was coined in [10] for actions of compact connected Lie groups๐บ on topological spaces๐, although its defining property, the collapse of the Serre spectral sequence associated with the fibration ๐ โช ๐๐บ โ ๐ต๐บ on the second page, was already investigated in [2], mostly for actions of tori and finite cyclic
groups of prime order. It is also worth pointing out that for a general fibration๐น โช ๐ธ โ ๐ต with connected fiber๐น and pathโconnected base ๐ต of finite type the degeneration of the associated Serre spectral sequence at the ๐ธ2โterm is equivalent to surjectivity of the inclusion induced mapH(๐ธ) โ H(๐น). In this situation, ๐น is traditionally said to be(totally) nonโcohomologous to zero in๐ธ, see [21, p. 148]. This shows the equivalence of the first two items in the following list of wellโknown characterizations of equivariant formality.
Proposition 1.1. Let๐บ be a compact connected Lie group with maximal torus ๐ and ๐ a connected ๐บโspace. The following statements are equivalent.
(1) The๐บโaction on ๐ is equivariantly formal.
(2) Fiber inclusion of the fibration๐ โช ๐๐บโ ๐ต๐บ induces a surjection H๐บ(๐) โ H(๐). (3) The๐โaction on ๐ obtained by restriction of the ๐บโaction is equivariantly formal.
(4) TheH(๐ต๐บ)โmodule H๐บ(๐) is free.
(5) We have an equality of total Betti numbersdim H(๐) = dim H(๐๐), where ๐๐ is the fixed point set of the induced๐โaction.
Actions on spaces with vanishing odd degree cohomology are equivariantly formal, as are symplectic mani-folds with a Hamiltonian action [10, Theorem 14.1]. Further examples of equivariantly formal actions areisotropy actions on symmetric spaces [6] and, more generally, on homogeneous spaces๐บ/๐พ in which the subgroup ๐พ is the connected component of the fixed point set of an arbitrary Lie group automorphism on๐บ, see [8]. Here, the isotropy action associated with a homogeneous space๐บ/๐พ is the action of ๐พ on ๐บ/๐พ induced by left multiplica-tion, that is, by the assignment(๐, ๐๐พ) โฆ ๐๐๐พ for all ๐ โ ๐พ, ๐๐พ โ ๐บ/๐พ. Our main contribution with this thesis now is that we extend the list of actions which are known to be equivariantly formal by one more item.
In theorem II.1.2 below we will show that the isotropy action associated with๐บ/๐พ is equivariantly formal if ๐พ is the connected component of the common fixed point set of two distinct commuting involutions on ๐บ, in which case๐บ/๐พ is said to be a โค2รโค2โsymmetric space, provided that none of the automorphisms is the identity map. The proof borrows some ideas from the proof of the main result of [8], which we therefore summarize in section 2. The key step is to construct a subgroup๐ป of ๐บ which shares a maximal torus with ๐พ and for which the cohomology of๐บ/๐ป is more accessible than that of ๐บ/๐พ, as then the isotropy action associated with ๐บ/๐ป is equivariantly formal if and only if so is the isotropy action associated with๐บ/๐พ. Since eventually we want to be able to give a description of a maximal torus of๐พ in terms of a maximal torus of ๐บ, we thus study in section II.2 the problem of reconstructing a maximal torus of๐บ from a fixed maximal torus ๐ of ๐พ. There is a general solution to this problem. Namely, upon fixing a reference torus๐ which is maximal in ๐บ and contains ๐, one finds that the complexification of the Lie algebra of the centralizer of๐ in ๐บ, which abstractly is the union of all maximal tori of๐บ containing ๐, is the direct sum of the complexification tโoft and the weight spaces of all gโโroots that vanish ons. While it is known that no such root exists if๐บ/๐พ is a symmetric space, certain gโโ roots might (and in general will) restrict to zero ons if๐บ/๐พ is โค2ร โค2โsymmetric, even if the automorphisms defining๐พ are both inner. Fortunately, however, the set of all such roots is strongly orthogonal, meaning that the sum of two elements of that set is not a root (see [16, p. 396]), and already sets of orthogonal roots in irreducible root systems can be classified up to application of a Weyl group element. This we have done in section II.4.
What makes this classification particularly useful is that in the present situation the maximal torus๐ of ๐พ is the intersection of the kernels of all roots vanishing ons and the fixed point set on๐ of one of the automorphisms defining๐พ. All of this data can be formulated in terms of the root system of gโand the list of possible sets of roots vanishing ons is further constrained by the requirement that the automorphisms defining๐พ be involutive. At this point, one could thus go through the list of all possible candidates for๐ and verify that the subalgebra ๐ acts in an equivariantly formal fashion on ๐บ/๐. We proceed differently and show that we may sequentially modify the automorphisms defining๐พ so as to almost always assume that one of them is an inner automorphism and that the semisimple part of the fixed point set of this inner automorphism realizes a subdiagram of the Dynkin diagram ofgโ. Homogeneous spaces arising from such subgroups have tractable cohomology, which we determine in section II.5. Building on these results, in section II.6 we finally traverse the list of simple Lie groups, determine in each case the desired subgroup๐ป, and show that the isotropy action of ๐ป on ๐บ/๐ป is equivariantly formal.
Our second contribution, which actually is equivalent to the main theorem of [6], is theorem III.5.10. The statement here is that for every compact connected Lie group๐บ and the connected component ๐พ of the fixed
point set of any involution on๐บ the action of ๐พ on ๐บ/๐ป by leftโmultiplication is equivariantly formal whenever ๐ป is a closed connected subgroup of ๐บ that contains ๐พ. Of course, the novelty is not the statement itself, but rather its proof, as it relies on an algebraic model for the equivariant cohomology of the๐พโaction on ๐บ/๐ป which is solely built out of the Lie algebras of๐บ, ๐ป, and ๐พ, and the inclusions of the latter two into the former. We note that such a model has been realized only very recently in [4, Sect. 3.1] using methods from rational homotopy theory, while our model is established by quite elementary means using the Cartan model for equivariant coho-mology. The drawback of our method is that it only captures theA
kโmodule structure ofH๐พ(๐บ/๐ป), Akโ S(kโ)
the space ofkโinvariant polynomials on kโ, whereas the model given in [4] is isomorphic toH๐พ(๐บ/๐ป) via an isomorphism ofA
kโalgebras. To explain this deficiency, consider an action of a compact connected Lie group๐บ
on a smooth manifold๐. The basic observation we exploit to construct our model is that there is a sequence of vector subspacesฮฉ(๐)๐บ,igฮฉ(๐)๐บ,(ig)2ฮฉ(๐)๐บ, โฆ whose sum is stable under the differential on ฮฉ(๐); here, ฮฉ(๐)๐บ is the space of๐บโinvariant forms on ๐ and igdenotes the image of the operatori โถ g โ End(ฮฉ(๐)), ๐ โฆ i๐, contracting a form with the vector field induced by๐ โ g. This leads to an additive, quasiโisomorphic model ofฮฉ(๐) and hence to a model of H๐บ(๐) which is isomorphic as an A
gโmodule.
Despite the lack of a ring structure our proof of theorem III.5.10, in contrast to the original proof in [6], does not rely on any classification result. Again, it has to be noted that a classificationโfree proof of the main theorem of [6] and even of [8, Theorem 1.1] was already achieved in [4, Theorem 7.8]. However, the proof presented in [4] uses๐พโtheory and relies on a reduction to the case when ๐บ is simple, while our proof works equally well for simple and nonโ simple Lie groups and only uses the decomposition ofg into the eigenspaces of the involution defining๐พ.
2. Previous results
Starting with this section we will almost exclusively consider isotropy actions on homogeneous spaces and be concerned with the question when such an action is equivariantly formal. It thus seems appropriate to make the following definition: given a compact connected Lie group๐บ and a closed connected subgroup ๐พ, we say that the pair(๐บ, ๐พ) is equivariantly formal if the action of ๐พ on ๐บ/๐พ by leftโmultiplication is equivariantly formal; we also say that(๐บ, ๐พ) is formal or a Cartan pair if the homogeneous space ๐บ/๐พ is formal in the sense of rational homotopy theory, which means that there exist commutative differential gradedโโalgebras ๐ด1, โฆ , ๐ด๐ and a
chain of morphismsฮฉ(๐บ/๐พ) โ ๐ด1 โ ๐ด2 โ โฆ โ ๐ด๐ โ H(๐บ/๐พ), each of which induces an isomorphism
on the level of cohomology. While this definition is valid for arbitrary (connected) manifolds, not just๐บ/๐พ, we prefer to use the following equivalent characterization of formality which is available in this particular situation: we recall from [11] that the spaceฮฉ(g)gofgโinvariant forms on g is an exterior algebra over an oddly graded subspace๐g โ ฮฉ(g)gof dimensionrank g, called primitive space of g, and that the Samelson subspace ๐ of the pair(g, k) is the graded subspace of ๐gwhose elements, considered as elements ofH(g), are contained in the image of the inclusion induced mapฮฉ(g, k) โ ฮฉ(g). Then we have dim ๐ โค rank ๐ โ rank k, cf. [11, Theorem V, sect. 10.4], and the pair(๐บ, ๐พ) is formal if and only if the previous inequality is actually an equality; see [11, Theorem VIII, sect. 10.4] for this and various other reformulations of formality.
These preliminary notions being introduced, we briefly summarize the proof of the main result in [8] and show how [8] is related to [7].
Theorem 2.1 ([8, Theorem 1.1]). Let๐บ be a compact connected Lie group and ๐พ โ ๐บ the identity component of the fixed point set of an automorphism on๐บ. Then the pair (๐บ, ๐พ) is (equivariantly) formal.
Note that according to [4, Theorem A] an equivariantly formal pair(๐บ, ๐พ) with both ๐บ and ๐พ connected is necessarily formal as well. That formality of a pair(๐บ, ๐พ) does not necessarily enforce equivariant formality of (๐บ, ๐พ) is shown in [8, Example 3.7].
The proof of theorem 2.1 given in [8] can be divided into two major steps: the first step is to show that it suffices to consider pairs (๐บ, ๐พ) satisfying the assumptions of theorem 2.1 and for which ๐บ is simple. In the
second step one actually proves theorem 2.1 for simple groups๐บ. Both steps crucially rely on the following general principle.
Theorem 2.2 ([4, Theorem 2.2]). Let๐พ and ๐ป be equal rank closed connected subgroups of a compact connected Lie group๐บ and such that ๐ป โ ๐พ. Then (๐บ, ๐พ) is equivariantly formal if and only if so is (๐บ, ๐ป).
A proof of theorem 2.2 is also contained in [8, Proposition 3.5] under the additional hypothesis that the pairs(๐บ, ๐พ) and (๐บ, ๐ป) are formal. Since by [22, p. 212] the pair (๐บ, ๐พ) is formal if and only if so is (๐บ, ๐ป), it follows from [4, Theorem A] that this seemingly more restrictive setting is actually equivalent to the general situation considered in theorem 2.2; the proof of the first item of [8, Proposition 3.5], which essentially states that formality of(๐บ, ๐พ) is equivalent to that of (๐บ, ๐ป), is erroneous though1.
The most important consequence of theorem 2.2 is that whenever๐ป and ๐พ are closed connected subgroups of a compact connected Lie group๐บ and ๐ is a maximal torus of both ๐ป and ๐พ, then the pair (๐บ, ๐พ) is equivariantly formal if and only if(๐บ, ๐ป) is equivariantly formal, because this property is satisfied by either one of the pairs if and only if it is satisfied by the pair(๐บ, ๐). In this way one can reduce the question of equivariant formality of pairs(๐บ, ๐พ) as in theorem 2.1 and with ๐บ simple to pairs for which ๐พ is the identity component of the fixed point set of a finiteโorder automorphism. The homogeneous space๐บ/๐พ arising from such a pair (๐บ, ๐พ) is called a๐โsymmetric space (๐ โฅ 0 the order of the automorphism defining ๐พ) or generalized symmetric space, and the question whether or not(๐บ, ๐พ) is equivariantly formal was already answered affirmatively in [7]. In fact, by [7, Proposition 3.7]๐พ shares a maximal torus with a subgroup ๐ป dubbed โfolded subgroupโ in [7], because its Dynkin diagram is obtained from the Dynkin diagram of๐บ by a process commonly called folding, and it was observed in [7, Theorem 5.5] that๐ป is (totally) nonโcohomologous to zero in ๐บ, that is, the fiber inclusion in the fibration๐ป โช ๐บ โ ๐บ/๐ป induces a surjection in cohomology. That (๐บ, ๐ป) is formal then is a classical result (cf. [11, Corollary I, sect. 10.19]) and equivariant formality follows from
Proposition 2.3 ([7, Proposition 2.6]). Let๐บ be a compact connected Lie group, ๐พ a closed connected subgroup. If๐พ is totally nonโcohomologous to zero in ๐บ, then (๐บ, ๐พ) is equivariantly formal.
The question of (equivariant) formality being settled for pairs in which the ambient group is simple, we return to the general situation considered in theorem 2.1. One now observes that whenever(๐บ, ๐พ) and (๐บโฒ, ๐พโฒ) are two pairs of compact and connected Lie groups such that there is an isomorphism of Lie algebra pairs (g, k) โ (gโฒ, kโฒ), then (๐บ, ๐พ) is (equivariantly) formal if and only if so is (๐บโฒ, ๐พโฒ), cf. [7, Corollary 2.4]. Thus, we
call a Lie algebra pair(u0, h0) equivariantly formal if there exists a compact connected Lie group ๐ and a closed connected subgroup๐ป such that (๐ , ๐ป) is equivariantly formal and (u, h) is isomorphic to (u0, h0), for then any other compact connected Lie group pair with matching Lie algebras is equivariantly formal as well. Passing to the level of Lie algebras, we denote by๐ the automorphism on g whose fixed point set is k. Then g decomposes as a direct sum of๐โinvariant subalgebras g1, โฆ , g๐which are minimal in the sense that none of them contains a nonโtrivial proper๐โinvariant subalgebra, k decomposes accordingly as the direct sum of the subalgebras g1โฉ k, โฆ , g๐โฉ k, and it only remains to check that each of the pairs (g๐, g๐โฉ k) is (equivariantly) formal. This is indeed the case: the pair(g๐, g๐โฉ k) is isomorphic to a Lie algebra pair (u โ โฆ โ u, ฮ(f)), where u is a compact simple Lie algebra andฮ(f) is the diagonal embedding of the fixed point set f of an automorphism on u, and ฮ(u) is totally nonโcohomologous to zero in g; these two facts together imply that (g๐, g๐โฉ k) is (equivariantly)
formal, see [8, Section 5] for more details.
1
Namely, instead of the displayed equation in the proof of the first part of [8, Proposition 3.5] one has to consider an equation of the form ๐(๐)|t= โ๐๐๐|tโ ๐๐with๐๐polynomials in the image of the transgression and๐๐nonโconstant polynomials invariant under the Weyl group of๐ป. Averaging both sides over the Weyl group of ๐พ gives the desired conclusion.
Chapter II.
โค
2ร โค
2โsymmetric spaces
1. โค2ร โค2โsymmetric spaces
There is yet another generalization of symmetric spaces that also incorporates the notion of๐โsymmetric spaces, the soโcalledฮโsymmetric spaces introduced in [19].
Definition 1.1. Letฮ be a finite Abelian group, ๐บ a connected Lie group, and ๐พ โ ๐บ a closed subgroup. The homogeneous space๐บ/๐พ is called ฮโsymmetric if there exists an injective group homomorphism ฮ โช Aut(๐บ) such that(๐บฮ)0โ ๐พ โ ๐บฮ, where๐บฮis the common fixed point set of the automorphismsฮ โ Aut(๐บ).
Since every finite Abelian group is a product of cyclic groups, the above definition can be rephrased by saying that a homogeneous space๐บ/๐พ with ๐บ connected and ๐พ โ ๐บ closed is ฮ = โค๐
1ร โฆ ร โค๐๐โsymmetric if there exist๐ distinct commuting automorphisms ๐1, โฆ , ๐๐ of๐บ, with ๐๐of order๐๐, such that
(๐บ๐1โฉ โฆ โฉ ๐บ๐๐)0โ ๐พ โ (๐บ๐1โฉ โฆ โฉ ๐บ๐๐).
Theorem 1.2. Let๐บ be a compact connected Lie group, ๐1and๐2two involutions on๐บ, and suppose that ๐บ/๐พ is aโค2ร โค2โsymmetric space, where๐พ = (๐บ๐1โฉ ๐บ๐2)0. Then the pair(๐บ, ๐พ) is (equivariantly) formal.
We note that the classification ofโค2รโค2โsymmetric spaces๐บ/๐พ with ๐บ a simple Lie group was achieved in [1] and [17], but while we do make use of the classification of simple Lie algebras and finiteโorder automorphisms thereon, our proof of theorem 1.2 does not rely on the classification ofโค2ร โค2โsymmetric spaces.
Recall (cf. [14, p. 130]) that a Lie algebrag is compact, if so is the connected subgroup ofAut(g) with Lie algebra{ad๐ | ๐ โ g}. According to [14, Corollary 6.7, chap. II] this is the case if and only if there is a compact Lie group with Lie algebra (isomorphic to)g. If g is compact and semisimple, then every connected Lie group with Lie algebrag is compact (see [14, Theorem 6.9, chap. II]), and we call a subalgebra hโ g compact, if the connected subgroup๐ป โ ๐บ with Lie algebra h is compact, where ๐บ is the simplyโconnected Lie group with Lie algebrag. For the sequel and for the proof of theorem 1.2 it will be convienent to introduce the following relation on the set of all compact subalgebras of a compact semisimple Lie algebrag: two such subalgebras h, k โ g are related, if there exists a sequence of compact subalgebrasm0, โฆ , m๐+1ofg such that m0 = h, m๐+1 = k and if for all๐ = 0, โฆ , ๐ the subalgebras m๐ andm๐+1share a common maximal torus, that is, if there exists a maximal toruss โ m๐ which also is maximal torus of m๐+1. This defines an equivalence relation and we denote the equivalence class of a subalgebrak by[k]f. Note that ifk โ g is a compact subalgebra, then the pair (g, k) is (equivariantly) formal if and only if there exists a subalgebrahโ [k]fsuch that(g, h) is so. Now theorem 1.2 will be a consequence of
Theorem 1.3. In addition to the hypotheses of theorem 1.2 assume that๐บ is simple. Then there exists a compact subalgebrahโ [k]fwhich is totally nonโcohomologous to zero ing.
Proof of theorem 1.2 using theorem 1.3. Let[g, g] = g1โ โฆ โ g๐ be the decomposition of the semisimple part ofg into its simple ideals and consider the subgroupฮ = {idg, ๐1, ๐2, ๐1๐2} inside the group of Lie algebra automorphisms ofg. It is isomorphic toโค2รโค2and acts naturally on๎ต โถ= {g1, โฆ , g๐}. Moreover, as was already observed in [8, Section 5], it will suffice to check that for each๐ the pair (m, m โฉ k), where m = โ๐พโฮ๐พ(g๐), is (equivariantly) formal.
Sethโถ= g๐and choose representatives๐พ1ฮh, โฆ , ๐พ๐ฮhfor each class inฮ/ฮh, whereฮhis the isotropy subgroup ath of the action ofฮ on ๎ต, ๐ = |ฮ/ฮh|, and ๐พ1= idg. Then an isomorphism of Lie algebras is given by the map
ฮฆโถ h โ โฆ โ h โ m, (๐1, โฆ , ๐๐) โฆ ๐พ1(๐1) + โฆ + ๐พ๐(๐๐),
because๐พ๐ (h) and ๐พ๐ก(h) are distinct ideals of [g, g] for ๐ โ ๐ก and m = โจ๐๐ =1๐พ๐ (h). Moreover, if f โ h is the common fixed point set of all elements inฮh, thenฮฆ maps ฮ(f), the diagonal embedding of f, isomorphically onto m โฉ k: in fact, any element๐พ โ ฮ permutes ฮ/ฮh, so there exist a permutation๐ on {1, โฆ , ๐} and elements ๐พ๐ โฒโ ฮhfor each๐ such that ๐พ๐พ๐ก = ๐พ๐(๐ก)๐พ๐กโฒfor all๐ก. Then we have, for all ๐ โ f:
๐พ(ฮฆ(๐, โฆ , ๐)) = โ๐
๐ =1๐พ๐พ๐ (๐) = ๐
โ
๐ =1๐พ๐(๐ )(๐) = ฮฆ(๐, โฆ , ๐).
To prove the converse inclusion, note that ifฮฆ(๐1, โฆ , ๐๐) is fixed by some ๐พ๐, then๐๐ = ๐1, because we chose ๐พ1 = id and because ๐พ๐๐พ๐(๐๐) โ h only holds if ๐ = ๐. Hence, if ฮฆ(๐1, โฆ , ๐๐) is fixed by all elements of ฮ, then
๐1= ๐2= โฆ = ๐๐and also๐1โ f, because every ๐พ โ ฮhleavesh invariant.
Thus, it will suffice to check that(โจ๐๐ =1h, ฮ(f)) is (equivariantly) formal. But an orbit of ฮ is either of length 1, 2, or 4, and if ๐ = 1, then f is just the common fixed point set of ๐1and๐2, whence the pair in question is (equivariantly) formal by theorem 1.3. If๐ = 2, then ฮhcontains one nonโtrivial element๐, so f = h๐is the fixed point set of an involution, and it was observed in [8, Section 5] that(h โ h, ฮ(f)) is (equivariantly) formal in this case as well: indeed, if we choosenโ [f]fto be totally nonโcohomologous to zero inh, which is possible by [8, Section 4] or [7, Theorem 5.5], thenฮ(n) is totally nonโcohomologous to zero in h โ h as well and ฮ(n) โ [ฮ(f)]f. Finally, if๐ = 4, then ฮh is trivial, whencef= h. As is wellโknown, ฮ(h) is totally nonโcohomologous to zero inhโ h โ h โ h.
2. Preliminaries
Let๐บ be a compact connected Lie group and ๐ a finiteโorder automorphism on ๐บ. It follows from [14, Lemma 5.3, chap. X], that the centralizerZg(s) in g of any maximal torus s of g๐ is a maximal torus ofg, and hence the unique maximal torus ofg containing s. Thus, if๐1, โฆ , ๐๐ are commuting automorphisms of๐บ, then there is a maximal torus ofg which is invariant for all๐๐,๐ = 1, โฆ , ๐. In fact, put ๐๐+1= id๐บ and suppose that for some๐, 1 โค ๐ โค ๐, t๐is a maximal torus ofk๐, where
k๐ โถ= g๐๐โฉ โฆ โฉ g๐๐+1,
and thatt๐ is invariant under๐1, โฆ , ๐๐+1; such a torus exists for๐ = 1, because k1is the common fixed point set of๐1, โฆ , ๐๐, whence any maximal torus ofk1is fixed by each๐๐. Since all๐๐ commute,๐๐ then restricts to a finiteโorder automorphism๐๐โถ k๐+1 โ k๐+1with fixed point setk๐. Ask๐+1is the common fixed point set of ๐๐+1, โฆ , ๐๐ and thus the Lie algebra of a compact Lie group, we conclude thatt๐+1= Zk
๐+1(t๐) is a maximal torus ofk๐+1. By definition,t๐+1is fixed by๐๐+1, โฆ , ๐๐, and if๐ โค ๐, then ๐๐(t๐+1) is a maximal torus of k๐+1containing t๐, hence must be equal tot๐+1. Continuing in this way, we eventually obtain a maximal torust๐+1ofk๐+1 = g with๐๐(t๐+1) = t๐+1for all๐ = 1, โฆ , ๐.
Proposition 2.1. Let๐บ be a compact connected Lie group, a โ g an Abelian subalgebra, and t a maximal torus ofg containing a. Denote byฮ โ (tโ)โthe set of roots with respect to the Cartan subalgebratโofgโand by ฮ โ ฮ the set of roots vanishing on a. Then, as a vector space,
Ngโ(a) = Zgโ(a) = tโโ โจ ๐ผโฮg
โ ๐ผ.
Proof. Thattโis contained inZ
definition, for every๐ โ a:
[๐, ๐] = ยฑ๐ผ(๐)๐ = 0, hencegโยฑ๐ผis contained inZ
gโ(a). Conversely, let ๐ โ Ngโ(a), and write
๐ = ๐0+ โ ๐ผโฮ๐๐ผ,
where๐0โ tโand๐๐ผ โ gโ๐ผ. For๐ โ a we have
tโโ a โ [๐, ๐ ] = โ
๐ผโฮ๐ผ(๐)๐๐ผ โ โจ๐ผโฮg โ ๐ผ,
which is only possible if๐ผ(๐)๐๐ผ = 0 for all ๐ผ โ ฮ. Hence, if ๐๐ผ โ 0, then a โ ker ๐ผ and ๐ผ โ ฮ. We have shown: tโโ โจ ๐ผโฮg โ ๐ผ โ Zg(a) โ Ng(a) โ tโโ โจ ๐ผโฮ0g โ ๐ผ.
For the remainder of this section we fix a compact connected Lie group๐บ, two commuting involutions ๐1and ๐2on๐บ (not necessarily different), and an Adโinvariant negative definite inner product โจโ , โ โฉ on g for which ๐1 and๐2are isometries. Note that any negative definiteAdโinvariant inner product (โ , โ ) on g gives rise to such an inner product: just take
(โ , โ ) + ๐1โ(โ , โ ) + ๐2โ(โ , โ ) + (๐1๐2)โ(โ , โ ).
Moreover, we put๐พ1โถ= (๐บ๐1)0,๐พ2โถ= (๐บ๐2)0, and choose a maximal torus๐ โ (๐บ๐1โฉ ๐บ๐2)0. According to our previous observations,๐1= Z๐พ
1(๐) then is a maximal torus in ๐พ1and๐ โถ= Z๐บ(๐1) is a maximal torus in ๐บ. Let ฮ be the gโโroots with respect totโ,ฮ+a choice of positive roots,ฮ โ ฮ the set of roots vanishing on s, and ฮ+โถ= ฮ โฉ ฮ+. We also set๐๐ผ โถ= ๐ผโฆ๐ whenever ๐ผ is a root and ๐ is an automorphism on g leaving t invariant. Proposition 2.2. Let g= k1โ p1be the decomposition ofg into the1โ and (โ1)โeigenspaces of ๐1. Then
(1) the root spacegโ๐ผ is contained inpโ1 for all๐ผ โ ฮ; (2) if๐ผ โ ฮ, then ๐1๐ผ = ๐ผ and
(3) ๐2๐ผ = โ๐ผ;
(4) any two roots๐ผ, ๐ฝ โ ฮ are strongly orthogonal, that is, neither ๐ผ + ๐ฝ nor ๐ผ โ ๐ฝ is a root;
(5) denoting for a root๐ผ by ๐ป๐ผ โ it the element with โจ๐ป๐ผ, โ โฉ = ๐ผ, we have tโ= โ
๐ผโฮ+ker ๐ผ โ โจ๐ผโฮ+โ๐ป๐ผ,
and any two summands in this decomposition are mutually orthogonal with respect toโจโ , โ โฉ.
Proof.
(1) Pick๐ผ โ ฮ and note that ๐1๐ผ still vanishes on s. Thus, the space ๐ โถ= gโ๐ผ + gโ๐
1๐ผ is๐1โinvariant, and so decomposes as the direct sum๐ = (๐ โฉ kโ1) โ (๐ โฉ pโ1). Now proposition 2.1 implies that
๐ โฉ kโ1 โ Zgโ(s) โฉ kโ1 = Zkโ1(s) = tโ1 โ tโ,
(2) We have just seen that given๐ผ โ ฮ the root space gโ๐ผ is contained inp1โ. So, if we pick๐ธ๐ผ โ gโ๐ผ and ๐ธโ๐ผ โ gโโ๐ผ, then[๐ธ๐ผ, ๐ธโ๐ผ] โ kโ1. We may assume thatโจ๐ธ๐ผ, ๐ธโ๐ผโฉ = 1, and then ๐ป๐ผ = [๐ธ๐ผ, ๐ธโ๐ผ] for the element๐ป๐ผ โ it with โจ๐ป๐ผ, โ โฉ = ๐ผ. Therefore,
๐ผ = โจ๐ป๐ผ, โ โฉ = โจ๐1(๐ป๐ผ), โ โฉโฆ๐1= โจ๐ป๐ผ, โ โฉโฆ๐1= ๐1๐ผ.
(3) According to the previous item,๐ป๐ผ โ kโ1, and since๐1and๐2commute,๐2(๐ป๐ผ) must be contained in kโ1 as well. Therefore,
๐ป๐ผ+ ๐2(๐ป๐ผ) โ tโโฉ kโ1 โฉ kโ2 = sโ.
Now for๐ โ sโwe compute
โจ๐, ๐ป๐ผ + ๐2(๐ป๐ผ)โฉ = 2โจ๐, ๐ป๐ผโฉ = 2๐ผ(๐) = 0.
Butโจ โ , โ โฉ is nonโdegenerate on sโ, hence we must have๐ป๐ผ + ๐2(๐ป๐ผ) = 0, which is equivalent to saying that๐2๐ผ = โ๐ผ, because ๐2is an isometry ofโจ โ , โ โฉ.
(4) Let๐ผ, ๐ฝ โ ฮ and suppose that ๐ผ + ๐ฝ was a root for a contradiction. We could choose nonโzero root vectors ๐๐ผ โ gโ๐ผ and๐๐ฝ โ gโ
๐ฝ, and then[๐๐ผ, ๐๐ฝ] โ gโ
๐ผ+๐ฝ would be a nonโzero root vector as well. ButZ
gโ(s) is a
Lie algebra and๐๐ผ,๐๐ฝare elements ofZ
gโ(s), so according to the first item
[๐๐ผ, ๐๐ฝ] โ kโ1 โฉ Zgโ(s) = tโ1 โ tโ,
which is impossible. Therefore,๐ผ + ๐ฝ is not a root.
(5) Let๐ผ, ๐ฝ โ ฮ+be two distinct roots. It is well known (cf. [16, Proposition 2.48, sect. II.5]) that the๐ผโstring containing๐ฝ, that is, the subset of ฮ โช {0} consisting of elements ๐ฝ + ๐๐ผ with ๐ โ โค, has no gaps and that the integers๐, ๐ โฅ 0 such that (๐ฝ + ๐๐ผ โ ฮ โช {0}) โบ (โ๐ โค ๐ โค ๐) satisfy ๐ โ ๐ = 2โจ๐ผ, ๐ฝโฉ/โจ๐ผ, ๐ผโฉ. Since neither๐ผ + ๐ฝ nor ๐ผ โ ๐ฝ is a root, we hence must have
0 = โจ๐ผ, ๐ฝโฉ = โจ๐ป๐ผ, ๐ป๐ฝโฉ.
In particular, the elements๐ป๐ผ,๐ผ โ ฮ+, are linearly independent. Now let ๐ = โจ
๐ผโฮ+โ๐ป๐ผ and๐
โฒ= โ ๐ผโฮ+ker ๐ผ.
Then the equation๐ผ(๐) = โจ๐ป๐ผ, ๐โฉ for ๐ โ tโshows that๐โฒ = tโโฉ ๐โ, and sotโ= ๐ โ ๐โฒ. 3. Automorphisms
We continue to use the notation of the previous section. Given๐ผ โ ฮ, denote by ๐ ๐ป
๐ผโถ it โ it the reflection along the hyperplane orthogonal to๐ป๐ผ, i.e. the map
๐ ๐ป๐ผ(๐) = ๐ โ 2โจ๐ปโจ๐ป๐ผ๐ผ, ๐ป, ๐โฉ๐ผโฉ. Since the elements ofฮ are mutually orthogonal, we immediately have
Proposition 3.1. The members of{๐ ๐ป๐ผ| ๐ผ โ ฮ+} commute pairwise.
Note that proposition 2.2 suggests that๐2acts as a product of hyperplane reflections on a certain subspace oft. This subspace will be a proper subspace in general, but if๐2is an inner autormophism, then it actually is
all oft. We shall show that under some mild assumptions on๐1the maximal toruss of k1โฉ k2can in fact be recovered fromฮ.
Proposition 3.2. Suppose that๐2 = ๐๐โฆ๐ holds for some element ๐ โ ๐บ and some automorphism ๐ on ๐บ that fixest1pointwise. Then
(1) the element๐ is contained in Z๐บ(๐) โฉ N๐บ(๐1), (2) the maximal torust is๐โinvariant,
(3) ๐2|it= โ๐ผโฮ+(๐ ๐ป
๐ผ)โฆ(๐|it). Proof.
(1) By assumption,๐1 is contained in the1โeigenspace of ๐ and ๐2 = ๐๐โฆ๐. Since ๐ is contained in the 1โeigenspace of ๐2and๐1is๐2โinvariant, the same statement is true with๐๐in place of๐2.
(2) Just note that๐(๐) is a maximal torus of ๐บ containing ๐1, so๐(๐) = ๐.
(3) We already observed that๐ centralizes ๐ and it is a wellโknown fact (see [16, Corollary 4.51, sect. IV.5]) that centralizers of tori are connected, so, according to proposition 2.1, we may express๐ as ๐ = exp(๐), where๐ = ๐0+๐ฮfor certain elements๐0โ t and ๐ฮโ โจ๐ผโฮgโ๐ผ. In particular, if๐ โ ๐ฟ, ๐ฟ โถ= โ๐ผโฮ+ker ๐ผ, then[๐, ๐] = 0. Thus, Ad๐fixes๐ฟ โฉ it pointwise, as do the elements ๐ ๐ป
๐ผ with๐ผ โ ฮ+. On the other hand, if๐ฝ โ ฮ+is arbitrary, then๐ป๐ฝ โ it1by proposition 2.2, so
Ad๐(๐ป๐ฝ) = ๐2(๐ป๐ฝ) = โ๐ป๐ฝ = ( โ
๐ผโฮ+๐ ๐ป๐ผ) (๐ป๐ฝ). Therefore,Ad๐restricts toโ๐ผโฮ+๐ ๐ป
๐ผ onit, whence the ๐โinvariance of t implies the claim. Corollary 3.3. Suppose that๐2= ๐๐โฆ๐ and t1โ g๐, and put๐ฟ โถ= โ๐ผโฮ+ker ๐ผ. Then
it โฉ ๐ฟ = (is) โ i(t โฉ p1), it1= (is) โ โจ
๐ผโฮ+โ๐ป๐ผ, and rank(k1โฉ k2) = rank(k1) โ |ฮ
+|.
Proof. We know from proposition 2.2 thatit = (it โฉ ๐ฟ) โ โจ๐ผโฮ+โ๐ป๐ผ is a decomposition into two๐1โinvariant subspaces and thatโจ๐ผโฮ+โ๐ป๐ผ is entirely contained init1. Thus, we must haveit โฉ ๐ฟ = (it1โฉ ๐ฟ) โ i(t โฉ p1) and it1= (it1โฉ ๐ฟ) โ โจ๐ผโฮ+โ๐ป๐ผ. Now recall that(it1)๐2= is, while (it1)๐2= it1โฉ ๐ฟ holds by proposition 3.2.
Ifg is simple the condition that๐2is a composition of an inner automorphism and an automorphism fixingt1 is not too restrictive: in fact, we will see later that if๐1is an outer automorphisms, then, except for Lie algebras of typeD4, we may assume that๐1 = ๐๐กโฆ๐ and ๐2 = ๐๐โฆ๐ or that ๐1 = ๐๐กโฆ๐ and ๐2 = ๐๐ for some involution ๐ โถ ๐บ โ ๐บ and elements ๐ก โ ๐1,๐ โ N๐ป(๐1) โ ๐, where ๐ป = (๐บ๐)0.
The following propositions state that in this case we may trade๐ก โ ๐1for some element๐กโฒ โ ๐ to first assume that๐ โ ๐ป and that ๐2= ๐๐โฆ๐; afterwards we may replace ๐1by an inner automorphism.
Proposition 3.4. Suppose that๐1= ๐๐กโฆ๐ and ๐2= ๐๐โฆ๐, where ๐ is an involution, ๐ = ๐ or ๐ = id๐บ,๐ก โ ๐1, and ๐ โ N๐ป(๐1) โ ๐, with ๐ป = (๐บ๐)0. Then there exist elements๐กโฒ โ ๐ and โ โ N๐ป(๐1) such that ๐๐กโฒโฆ๐ and ๐โโฆ๐ are commuting involutions whose common fixed point set hass as a maximal torus.
Proof. First suppose that๐ = ๐. Then we choose ๐ โ exp(t โฉ p1), โ โ N๐ป(๐1) with ๐ = โ๐ and set ๐ฟ โถ= โ๐ผโฮ+ker ๐ผ. Note that t โฉ ๐ฟ decomposes, by corollary 3.3, as t โฉ ๐ฟ = s โ (t โฉ p1) and that the elements of t โฉ ๐ฟ are
fixed byAdโ, because๐1|t= ๐|tand hence proposition 3.2 applies. So if we pick๐ โ t โฉ p1with๐ = exp(๐) and put๐ = exp(๐/2), then ๐๐โ1โฆ๐2โฆ๐๐ is an involution,๐ = ๐2, and๐(๐) = ๐โ1. Therefore, we have
similarly,๐๐โ1โฆ๐1โฆ๐๐ = ๐๐โ1๐กโฆ๐. Thus, ๐๐โ1๐กโฆ๐ and ๐โโฆ๐ are two commuting involutions. Since their common fixed point subalgebra is conjugate tok1โฉ k2viaAd๐โ1 andAd๐โ1fixess, the claim follows.
Now assume that๐ = id๐บ. Choose a decomposition๐ = โ๐ as before and use corollary 3.3 to additionally find๐ โ exp(s), ๐ โ exp(โจ๐ผโฮ+โ(i๐ป๐ผ)) with ๐ก = ๐ ๐. We will show that ๐ โถ= ๐๐๐ โฆ๐ is an involution, that ๐ commutes with๐1, and thats is a maximal torus of g๐1โฉ g๐. The previous case then implies the claim, because ๐๐ โ N๐ป(๐1) โ ๐. To begin with, we assert that (๐๐ )2 = (๐๐)2; indeed,๐โ and๐๐ coincide ont, whence we have ๐โ(๐) = ๐โ1and๐โ(๐ ) = ๐ (cf. proposition 3.2), so this follows from
๐๐= ๐1โฆ๐๐โฆ๐1= ๐๐กโฆ๐๐(๐)โฆ๐๐ก= ๐๐ โฆ๐๐โฆ๐โโฆ๐๐โ1โฆ๐๐ก= ๐โโฆ๐๐ โฆ๐๐โ1โฆ๐๐โ1โฆ๐๐ก
together with the commutativity of๐, ๐ , and ๐. Also note that โ, ๐, and ๐ commute with each other and that ๐ป contains๐ . These observations imply that ๐ is an involution commuting with ๐1, since
๐2= ๐๐๐ โฆ๐โฆ๐๐๐ โฆ๐ = ๐๐๐ โฆ๐โ๐โ1๐ = (๐โ)2โฆ(๐๐ )2= (๐๐)2= id
and since๐๐ ,๐, and ๐๐commute with๐1. Finally, note that any maximal torus ofg๐1โฉg๐containings is a subset ofZg(s) and that by propositions 2.1 and 2.2 ๐1only fixest1onZg(s). Then s must be a maximal torus of g๐1โฉg๐, astโandโจ๐ผโฮgโ๐ผ are๐โinvariant subspaces and ๐|t
1= ๐2|t1only fixess.
Proposition 3.5. Suppose that๐1= ๐๐กโฆ๐ and that ๐2= ๐โโฆ๐, where ๐ is an involution, ๐ก is contained in ๐, and โ is an element of N๐ป(๐1), with ๐ป = (๐บ๐)0. Letฮ odd โ ฮ be the set of all roots ๐ฝ โ ฮ for which the integer โ๐ผโฮ+2โจ๐ผ, ๐ฝโฉ/โจ๐ผ, ๐ผโฉ is odd. Then ๐(๐ผ) โ ๐ผ for all ๐ผ โ ฮ odd.
Lemma 3.6. Under the assumptions of proposition 3.5 we have gโ๐ผ โ hโfor each root๐ผ โ ฮ.
Proof. Observe that the requirements of proposition 3.2 are met, soโ is an element of Z๐บ(๐) โฉ ๐ป = Z๐ป(๐). Since Z๐ป(๐) is connected, we may express โ as โ = exp(๐) for some element ๐ โ Zh(s) = Zg(s) โฉ h, say ๐ = ๐0+ โ๐ผโฮ+๐๐ผ, with๐0โ t and ๐๐ผ โ gโ๐ผ โ gโโ๐ผfor each root๐ผ โ ฮ+. Recall that๐1coincides with๐ on t, because๐๐กis the identity ont, so as๐1fixes each root๐ผ โ ฮ+,๐ fixes each element of ฮ+too. Therefore,gโ๐ผ and gโโ๐ผare eigenspaces of๐, whence ๐๐ผnecessarily vanishes ifg๐ผโโ hโ. However, if๐ฝ โ ฮ+was a root with๐๐ฝ = 0, then, as the elements ofฮ are strongly orthogonal, we also would have [๐, ๐ป๐ฝ] = 0, and hence Adโ(๐ป๐ฝ) = ๐ป๐ฝ. But this is impossible, because we know from proposition 3.2 thatAdโ(๐ป๐ฝ) = โ๐ป๐ฝ. Consequently,๐๐ฝ โ 0 and gโ๐ฝ โ hโ.
Proof of proposition 3.5. The decompositiont= sโsโฒ, withsโฒ= โจ๐ผโฮ+โ(i๐ป๐ผ)โ(tโฉp1) yields a decomposition ๐ = ๐+๐โfor every element๐ โ ๐, where ๐+โ exp(s) and ๐โโ exp(sโฒ). Moreover, ๐2restricts toid on s and to (โid) on sโฒ, so the condition that๐๐โฆ๐ commutes with ๐2can be rephrased as
๐๐โฆ๐ = ๐2โฆ๐๐โฆ๐โฆ(๐2)โ1โบ ๐๐โฆ๐ = ๐๐2(๐)โฆ(๐โ)2โฆ๐ โบ (๐๐โ)2= (๐โ)2;
but๐โis an involution, because๐โcommutes with๐ and ๐2is an involution, so๐๐โฆ๐ commutes with ๐2if and only if๐๐
โis an involution. In particular, if we let๐ก = ๐ก+๐กโ, then๐๐ก
โ is an involution. With this characterization at hand we can show that no root inฮ
oddis fixed by๐: let us further decompose
๐กโas๐กโ = ๐๐, where ๐ โ exp(t โฉ p1), ๐ = exp(๐), and ๐ = โ๐ผโฮ+๐ก๐ผi๐/โจ๐ผ, ๐ผโฉ๐ป๐ผ for certain real numbers๐ก๐ผ. Recalling that each element๐ฝ โ ฮ is contained in the (โ1)โeigenspace of ๐1, but in the fixed point set of๐, and thatsโ (t โฉ p1) is the common kernel of the elements of ฮ on t, we find that
โidgโ
๐ฝ = ๐1|gโ๐ฝ = Ad๐|gโ๐ฝ = ๐
i๐๐ก๐ฝid;
so,(๐ก๐ฝโ 1) โ 2โค. On the other hand, if ๐ฝ โ ฮ with ๐(๐ฝ) = ๐ฝ is arbitrary, then Ad๐restricts toยฑ id on gโ
tโฉ p1is the(โ1)โeigenspace of ๐ on t. Combined with the fact that ๐๐ก
โ is an involution this gives idgโ
๐ฝ = (Ad๐๐||gโ๐ฝ)
2
= (Ad๐|gโ๐ฝ)
2= (โ1)โ๐ผโฮ+2โจ๐ผ,๐ฝโฉโจ๐ผ,๐ผโฉ โ id, because2โจ๐ผ, ๐ฝโฉ/โจ๐ผ, ๐ผโฉ is an integer and (๐ก๐ผโ 1) is an even number. Therefore, ๐ฝ โ ฮ
odd.
Corollary 3.7. In addition to the hypotheses of proposition 3.5 assume that g is semisimple. Letฮ even= ฮ โงตฮ odd
and choose, for each๐ผ โ ฮ
odd,๐๐ผ โ {ยฑ1} with ๐๐ผ = โ๐๐(๐ผ). There exists๐ โ exp(t โฉ p1) such that
(1) Ad๐is equal to(๐๐ผi) โ id on gโ๐ผ and to the identity ongโ
๐ฝ for all๐ผ โ ฮ
odd,๐ฝ โ ฮ even,
(2) the automorphism๐ = ๐๐ exp(๐), where๐ = โ๐ผโฮ+i๐/โจ๐ผ, ๐ผโฉ๐ป๐ผ, is an involution, and (3) ๐2commutes with๐ and s is a maximal torus of g๐โฉ g๐2.
Proof. Choose๐ โ t such that ๐ผ(๐) = 0 for all ๐ผ โ ฮ evenand such that๐ผ(๐) = ๐๐ผi๐/2 for all roots ๐ผ โ ฮ odd; this
is possible, because the restrictions of the elements ofฮ constitute a basis of (it)โ. Then๐ is necessarily contained intโฉ p1, because๐ผ(๐ + ๐(๐)) vanishes for all ๐ผ by choice of the integers ๐๐ฝ,๐ฝ โ ฮ
odd. We set๐ โถ= exp(๐) and
observe thatAd๐indeed is equal to(๐๐ผi) โ id on gโ๐ผ, if๐ผ โ ฮ
odd, and toid else. Thus, for each simple root ๐ผ โ ฮ
the maps(Ad๐)2and(Adexp(๐))2coincide ongโ๐ผ and are equal toid or (โid), so ๐ = Ad๐ exp(๐)is an involution. Moreover,๐ commutes with ๐2, because๐โฆ๐2= ๐โ1โฆ๐2.
Hence, it remains to show that s is a maximal torus of g๐ โฉ g๐2, and to this end it suffices to verify the maximality ofs. However, we already know that the complexification ofZg(s) is the sum of the ๐1โ and๐2โ invariant subspacestโandโจ๐ผโฮgโ๐ผ. By construction,Ad๐ equalsid on the latter space, because ๐(๐ผ) = ๐ผ for ๐ผ โ ฮ, while Adexp(๐)is just(โid) by proposition 2.2; hence ๐ only fixes t in Zg(s), and the fixed point set of ๐2 ont is precisely s, because t1= t๐. Thus, onlys is fixed by both๐ and ๐2inZg(s).
4. Normal forms for strongly orthogonal roots 4.1. Abstract normal forms
In the previous sections we learned that for a suitable choice of Cartan subalgebra the set of roots vanishing on a maximal torus of the joint fixed point subalgebra of two commuting inner involutions is strongly orthogonal and satisfies a certain involutivity condition. The purpose of this section is to establish a normal form for all sets of roots satisfying these properties.
Recall (cf. [16, p. 149]) that an(abstract) root system(๐ , โจโ , โ โฉ, ฮ) consists of a finiteโdimensional Euclidean vector space(๐ , โจโ , โ โฉ) together with a nonโempty set ฮ โ ๐ of nonโzero vectors such that
(1) ๐ = spanโฮ,
(2) for each๐ผ โ ฮ the reflection
๐ ๐ผโถ ๐ โ ๐ , ๐ฃ โฆ ๐ฃ โ 2โจ๐ผ, ๐ฃโฉโจ๐ผ, ๐ผโฉ ๐ผ,
mapsฮ into itself, and
(3) the number2โจ๐ผ, ๐ฝโฉ/โจ๐ผ, ๐ผโฉ is an integer whenever ๐ผ and ๐ฝ are elements of ฮ.
A root systemฮ is reduced if ๐ผ โ ฮ implies that 2๐ผ โ ฮ. It is called reducible if there exists a nonโtrivial disjoint decompositionฮ = ฮโฒโ ฮโฒโฒsuch thatโจ๐ผโฒ, ๐ผโฒโฒโฉ = 0 for all ๐ผโฒโ ฮโฒand๐ผโฒโฒโ ฮโฒโฒ. If no such decomposition exists, thenฮ is irreducible.
(1) A pair(๐ , ฮฉ) is a root subsystem of ฮ if
a) ฮฉ โ ฮ is nonโempty,
b) ๐ = spanโฮฉ, and c) ๐ ๐ผ(ฮฉ) โ ฮฉ for all ๐ผ โ ฮฉ.
(2) Theroot subsystem ofฮ spanned by ๐, ๐ โ ฮ a nonโempty set, is the pair (spanโ๐, ฮ โฉ spanโค๐). Remark 4.2. Letฮ be a root system.
(1) If(๐ , ฮฉ) is a root subsystem of ฮ, then (๐ , โจโ , โ โฉ|๐ ร๐, ฮฉ) is a root system. If ๐ โ ฮ is a nonโempty subset, then the root subsystem spanned by๐ is a root subsytem in the sense of definition 4.1.
(2) Let(๐ , ฮฉ) be a root subsystem of ฮ. We can identify the Weyl group ๐ (ฮฉ) of ฮฉ, which by definition is a subgroup ofO(๐ , โจโ , โ โฉ|๐ ร๐), with a subgroup ๐ (ฮฉ, ฮ) of the Weyl group ๐ (ฮ) of ฮ, where
๐ (ฮฉ, ฮ) โถ={๐ค โ ๐ (ฮ) | ๐ค = ๐ ๐ผ1โฆ โฆ โฆ๐ ๐ผ๐,๐ผ๐โ ฮฉ}โ O(๐ , โจโ , โ โฉ).
In fact, the map๐ โถ ๐ (ฮฉ, ฮ) โ ๐ (ฮฉ) restricting an element ๐ค โ ๐ (ฮฉ, ฮ) to ๐ is a homomorphism of groups. Moreover, if๐ค โ ๐ (ฮฉ), say with ๐ค = ๐ก๐ผ
1โฆ โฆ โฆ๐ก๐ผ๐, where๐ผ๐ โ ฮฉ and ๐ก๐ผ
๐โถ ๐ โ ๐ denotes
reflection along the hyperplane in๐ perpendicular to ๐ผ๐, then ๐(๐ ๐ผ1โฆ โฆ โฆ๐ ๐ผ๐) = ๐ค;
and if๐ค โ ker ๐, then ๐ค = id๐, because๐(๐ค) = id๐ and๐ค(๐ฃโฒ) = ๐ฃโฒfor all๐ฃโฒโ ๐โby definition.
Recall that any choice of positive rootsฮ+in a root systemฮ determines a set of simple roots ฮ โ ฮ+, and that any root๐ผ can be uniquely written as ๐ผ = โ๐ฝโฮ ๐๐ฝ๐ฝ for integers ๐๐ฝof the same sign. The numberโ๐ฝโฮ ๐๐ฝis commonly referred to as thelevel of the root๐ผ.
Proposition 4.3. Letฮ be a reduced irreducible root system, ฮ+ โ ฮ a choice of positive roots, and ๐ผ0 โ ฮ. There exists a unique root๐ฟ of maximal level in the orbit ๐ โ ๐ผ0of the Weyl group๐ = ๐ (ฮ), and this root satisfiesโจ๐ฟ, ๐ผโฉ โฅ 0 for all ๐ผ โ ฮ+.
Proof. Choose any root๐ฟ of maximal level in ๐ โ ๐ผ0= {๐ค(๐ผ0) | ๐ค โ ๐ }. If ๐ผ โ ฮ+is a root withโจ๐ฟ, ๐ผโฉ < 0, then๐ ๐ผ(๐ฟ) is a root having higher level than ๐ฟ and still is contained in ๐ โ ๐ผ0, which is impossible. Therefore, we haveโจ๐ฟ, ๐ผโฉ โฅ 0 for any positive root ๐ผ. In order to prove the uniqueness statement, let ฮ โ ฮ+be the simple roots associated with the given choice of positivity and note that๐ฟ is positive, so we may write
๐ฟ = โ
๐ผโฮ ๐๐ผ๐ผ,
with๐๐ผ โ โคโฅ0. We claim that each of the integers๐๐ผis nonโzero. For if this was not the case, thenฮ = ฮ โฒโชฮ โฒโฒ withฮ โฒ= {๐ผ | ๐๐ผ = 0} and ฮ โฒโฒ= {๐ผ | ๐๐ผ > 0} would be a nonโtrivial disjoint union. Moreover, for any ๐ฝ โ ฮ โฒ we would have
โจ๐ฟ, ๐ฝโฉ = โ
๐ผโฮ โฒโฒ๐๐ผโจ๐ผ, ๐ฝโฉ,
and the right hand side is nonโpositive, because the inner product of two distinct simple roots already is nonโ positive. By what we have just shown, โจ๐ฟ, ๐ฝโฉ โฅ 0, and so โจ๐ฟ, ๐ฝโฉ = 0 and hence โจ๐ผ, ๐ฝโฉ = 0 would have to hold for all๐ผ โ ฮ โฒโฒand๐ฝ โ ฮ โฒ. But this is impossible, because we are assumingฮ to be irreducible. Now let ๐พ โ ๐ โ ๐ผ0be another root of maximal level. The same line of reasoning as before also applies to๐พ and shows
that๐พ = โ๐ผโฮ ๐๐ผ๐ผ for integers ๐๐ผ > 0. In particular, since there is some simple root ๐ฝ โ ฮ with โจ๐ฟ, ๐ฝโฉ > 0, we also must haveโจ๐ฟ, ๐พโฉ > 0. Therefore, ๐ฟ โ ๐พ is either a positive or a negative root (or 0), and since
๐ฟ โ ๐พ = โ
๐ผโฮ (๐๐ผ โ ๐๐ผ)๐ผ,
it follows that(๐๐ผ โ ๐๐ผ)๐ผโฮ is either a sequence of nonโnegative or nonโpositive integers. But๐ฟ and ๐พ have the same level, that is,โ๐ผโฮ ๐๐ผ = โ๐ผโฮ ๐๐ผ, and therefore๐๐ผ = ๐๐ผ for all๐ผ โ ฮ .
Letฮ be a reduced irreducible root system and ฮ+a choice of positive roots. A wellโknown consequence of the classification of such root systems is that any two simple roots of the same length are contained in the same Weyl group orbit. On the other hand, every root is contained in the Weyl group orbit of a simple root (see [16, Proposition 2.62, sect. II.6]), so if๐ฟ is the length of a root in ฮ+, then by proposition 4.3 we may unambiguously speak of the highest root (with respect to the level) of length๐ฟ.
Now letฮ โ ฮ+be the simple roots andฮ โ ฮ a nonโempty set of (not necessarily strongly) orthogonal roots such thatฮ = (โฮ). We further suppose that all elements of ฮ are of the same length ๐ฟ > 0 and put ฮ+= ฮ โฉ ฮ+. We claim that there is a way to describe the possible elements thatฮ may contain, up to application of a Weyl group element. To this end, let us introduce some notation for nonโempty subsets๐ด โ ฮ that we will make use of in the sequel. Given such a set๐ด we write ฮ๐ดto denote the root subsystem ofฮ spanned by ๐ด and we putฮ+
๐ด = ฮ๐ดโฉ ฮ+, which is a notion of positivity with simple roots๐ด. Moreover, we call ๐ด irreducible if ฮ๐ด is irreducible, and refer to a nonโempty subset๐ดโฒ โ ๐ด as an irreducible component of ๐ด if ๐ดโฒis maximal (with respect to inclusion) among all irreducible subsets of๐ด. Note that ๐ด decomposes as ๐ด = ๐ด1โช โฆ โช ๐ด๐, where each๐ด๐is an irreducible component of๐ด and the members of ๐ด๐are orthogonal to๐ด๐for all๐ โ ๐. Finally, if ๐ด is irreducible and admits roots of length๐ฟ, then we write ๐ฟ(๐ด) to denote the highest root of length ๐ฟ in ฮ๐ด(with respect toฮ+๐ด).
Next, we recursively define a family(๎ญ๐)๐=0,โฆ,๐of nonโempty subsets of๎ผ(ฮ ) (the power set of ฮ ) as follows. We put๎ญ0โถ= {ฮ } and suppose that for some ๐ โฅ 0 the sets ๎ญ0, โฆ , ๎ญ๐are already defined. Then a nonโempty subset๐ด โ ฮ is contained in ๎ญ๐+1if and only if
(1) ฮ๐ดis irreducible and admits roots of length๐ฟ,
(2) there exists a (possibly empty) set๐ต โ ฮ whose members are orthogonal to each member of ๐ด and a set ๐ฃ(๐ด) โ ๎ญ๐such that
๐ฟ(๐ฃ(๐ด))โโฉ ๐ฃ(๐ด) = ๐ต โช ๐ด;
in other words,๐ด is an irreducible component of ๐ฟ(๐ฃ(๐ด))โโฉ ๐ฃ(๐ด) that admits roots of length ๐ฟ. We put ๐ โถ= ๐ if no such๐ด exists and call ๎ญ0, โฆ , ๎ญ๐thenormal form tree for(ฮ, ฮ+) and ๐ฟ.
Remark 4.4. Closely related to the normal form tree construced above is the soโcalled cascade of strongly orthogonal roots defined in [18, Section 1]: indeed, if๐ด โ ๎ญ๐for some๐ > 1, then in the notation of [18] ๐ฟ(๐ด) is an offspring of๐ฟ(๐ฃ(๐ด)). If ๐ด0, โฆ , ๐ด๐are such that๐ด๐ โ ๎ญ๐, then{๐ฟ(๐ด0), โฆ , ๐ฟ(๐ด๐)} is called a chain cascade in [18].
Proposition 4.5. Any two distinct sets๎ญ๐,๎ญ๐ are disjoint andฮ๐ด,ฮ๐ดโฒ are perpendicular for all๐ด, ๐ดโฒ โ ๎ญ๐ with๐ด โ ๐ดโฒ. Moreover, for๐ด โ ๎ญ๐+1the element๐ฃ(๐ด) is the only set in ๎ญ๐with๐ด โฉ ๐ฃ(๐ด) โ โ .
Remark 4.6. Thus, we may define a graph with vertices the elements of๎ญ0โชโฆโช๎ญ๐, where๐ด, ๐ดโฒare connected by an edge if and only if๐ด = ๐ฃ(๐ดโฒ). The resulting graph is a tree, hence the name.
Proof. We first show by induction on๐ = 0, โฆ , ๐ that ๐ฃ(๐ด) is the only set in ๎ญ๐intersecting๐ด โ ๎ญ๐+1nonโ trivially and that๐ด, ๐ดโฒ โ ๎ญ๐ have nonโtrivial intersection only if๐ดโฒ = ๐ด. This is immediate if ๐ = 0, because ๎ญ0 = {ฮ }, so suppose that the induction hypothesis has been established for some natural number ๐ โฅ 0.
Choose๐ด, ๐ดโฒโ ๎ญ๐+1arbitrarily and note that by the induction assumption๐ฃ(๐ด) and ๐ฃ(๐ดโฒ) are the unique sets in๎ญ๐with๐ด โฉ ๐ฃ(๐ด) โ โ and ๐ดโฒโฉ ๐ฃ(๐ดโฒ) โ โ . Hence, if ๐ด โฉ ๐ดโฒis nonโempty, then, since๐ด โ ๐ฃ(๐ด) and ๐ดโฒโ ๐ฃ(๐ดโฒ) holds by definition, also๐ฃ(๐ด)โฉ๐ฃ(๐ดโฒ) is nonโempty, so by the induction assumption we must have ๐ฃ(๐ด) = ๐ฃ(๐ดโฒ). The defining property of๐ฃ(๐ด) is that ๐ฟ(๐ฃ(๐ด))โโฉ ๐ฃ(๐ด) = ๐ต โช ๐ด holds for some subset ๐ต โ ฮ whose members are orthogonal to each member of๐ด. Therefore,
ฮ๐ดโฒ= (ฮ๐ดโฒโฉ spanโค(๐ดโฒโฉ ๐ต)) โช (ฮ๐ดโฒโฉ spanโค(๐ดโฒโฉ ๐ด))
is a decomposition into two sets whose members are mutually orthogonal, whence by irreducibility ofฮ๐ดโฒwe must haveฮ๐ดโฒโ spanโค(๐ดโฉ๐ดโฒ). Thus, ๐ดโฒโฉ๐ต is empty and ๐ดโฒ โ ๐ด. Exchanging the roles of ๐ด and ๐ดโฒwe conclude that๐ด = ๐ดโฒ, so two sets in๎ญ๐+1intersect nonโtrivially only if they are equal. To finish the induction step, just note that if๐ด โ ๎ญ๐+2is arbitrary and๐ต โ ๎ญ๐+1intersects๐ด nonโtrivially, then also ๐ฃ(๐ด) โฉ ๐ต โ โ , because ๐ด โ ๐ฃ(๐ด), so by what we have just shown ๐ต = ๐ฃ(๐ด).
Now suppose that๐ด, ๐ดโฒ โ ๎ญ๐ are two distinct sets and let๐ โฅ 0 be the smallest integer such that ๐ฃ๐+1(๐ด) = ๐ฃ๐+1(๐ดโฒ). By definition we have ๐ฟ(๐ฃ๐+1(๐ด))โโฉ ๐ฃ๐+1(๐ด) = ๐ต โช ๐ฃ๐(๐ด) for some set ๐ต which is perpendicular to
๐ฃ๐(๐ด) and hence intersects ๐ฃ๐(๐ด) trivially. Since we just showed that ๐ฃ๐(๐ด) intersects ๐ฃ๐(๐ดโฒ) trivially as well, we conclude that๐ฃ๐(๐ดโฒ) must be contained in ๐ต. Thus, ๐ฃ๐(๐ดโฒ) is perpendicular to ๐ฃ๐(๐ด), whence ๐ด and ๐ดโฒ are perpendicular too, because๐ด โ ๐ฃ๐(๐ด) and ๐ดโฒ โ ๐ฃ๐(๐ดโฒ). Finally, suppose that ๐ด is contained in ๎ญ๐โฉ ๎ญ๐+๐ for integers๐ โฅ 0 and ๐ โฅ 1. Then ๐ฃ๐(๐ด) โ ๎ญ0โฉ ๎ญ๐, whence๐ฃ๐(๐ด) = ฮ . This is impossible, however, because each element of๎ญ๐is a proper subset ofฮ .
Corollary 4.7. For๐ต โ ๎ญ๐, and all๐ด โ ๎ญ0โช โฆ โช ๎ญ๐such that๐ด โ ๐ต we have ๐ต โ ๐ฟ(๐ด)โ.
Proof. If๐ด โ ๎ญ๐, the statement follows readily from proposition 4.5, so we suppose that๐ด โ ๎ญ๐โ๐ for some ๐ โฅ 1. If ๐ด is different from ๐ฃ๐(๐ต), then even ๐ฃ๐(๐ต) and ๐ด are perpendicular. If ๐ด is equal to ๐ฃ๐(๐ต), then ๐ฃ๐โ1(๐ต) โ
๐ฟ(๐ด)โholds by definition, so๐ต โ ๐ฃ๐โ1(๐ต) is perpendicular to ๐ฟ(๐ด).
Corollary 4.8. Let๐ต โ ๎ญ๐. Any๐ค โ ๐ (๐ต) permutes the members of {ฮ๐ด| ๐ด โ ๎ญ๐}, if ๐ โค ๐.
Proof. Fix some๐ โฅ 0 and put ๐ โถ= ๐ โ ๐. If ๐ด โ ๎ญ๐ is different from๐ฃ๐(๐ต), then ๐ด and ๐ฃ๐(๐ต) are perpendicular, whence so are๐ด and ๐ต. Since ๐ค is a product of reflections ๐ ๐ผ with๐ผ โ ๐ต, ๐ค hence fixes ๐ด and ฮ๐ดin this case. On the other hand, if๐ด = ๐ฃ๐(๐ต), but ๐ > 0, let ๐ถ1, โฆ , ๐ถ๐ โ ฮ be the irreducible components of ๐ฟ(๐ด)โโฉ ๐ด. Note that๐ถ๐is contained in๎ญ๐+1if and only ifฮ๐ถ
๐admits roots of length๐ฟ, so we may further assume that for some ๐ โฅ 1 the sets ๐ถ1, โฆ , ๐ถ๐ contain roots of length๐ฟ, while ๐ถ๐ +1, โฆ , ๐ถ๐do not, and that๐ฃ๐โ1(๐ต) = ๐ถ1. Now observe that the root subsystem spanned by๐ฟ(๐ด)โโฉ ๐ด is precisely ๐ฟ(๐ด)โโฉ ฮ๐ด. Indeed, any root๐ผ โ ฮ๐ดis aโคโฅ0โ or โคโค0โlinear combination of elements in๐ด, so if โจ๐ฟ(๐ด), ๐ผโฉ = 0, then ๐ผ must be a linear combination of elements in๐ฟ(๐ด)โโฉ ๐ด, because โจ๐ฟ(๐ด), ๐ฝโฉ โฅ 0 holds for all ๐ฝ โ ๐ด by proposition 4.3. Hence, we have
๐ฟ(๐ด)โโฉ ฮ๐ด= ฮ๐ถ1โช โฆ โช ฮ๐ถ๐.
Also note that๐ต is perpendicular to ๐ฟ(๐ด), but contained in ฮ๐ด, so๐ค leaves ๐ฟ(๐ด)โโฉ ฮ๐ดinvariant. Hence, since๐ค is an isometry andฮ๐ถ
๐ is irreducible, we must have๐ค(ฮ๐ถ
๐) โ {ฮ๐ถ1, โฆ , ฮ๐ถ๐} for each ๐. Moreover, if ฮ๐ถ๐ admits roots of length๐ฟ, then so does ๐ค(ฮ๐ถ
๐), whence ๐ค even permutes the set {ฮ๐ถ1, โฆ , ฮ๐ถ๐ }. Theorem 4.9. There exists a Weyl group element๐ค โ ๐ (ฮ) such that
(1) ๐ค(ฮ) โฉ ฮ+โ {๐ฟ(๐ด) | ๐ด โ ๎ญ0โช โฆ โช ๎ญ๐} and
(2) if๐ฟ(๐ด) is contained in ๐ค(ฮ) โฉ ฮ+, then either๐ด = ฮ or ๐ฟ(๐ฃ(๐ด)) is contained in ๐ค(ฮ).
Lemma 4.10. If๐ผ โ ฮ๐ด,๐ด โ ๎ญ๐, is perpendicular to๐ฟ(๐ด), then ๐ผ โ ฮ๐ดโฒfor some irreducible component๐ดโฒof ๐ฟ(๐ด)โโฉ ๐ด. If in addition ๐ผ is of length ๐ฟ, then ๐ < ๐ and ๐ดโฒis contained in๎ญ๐+1.
Proof. Express๐ผ as ๐ผ = โ๐ฝโ๐ด๐๐ฝ๐ฝ for integers (๐๐ฝ)๐ฝโ๐ด of the same sign. Since โจ๐ผ, ๐ฟ(๐ด)โฉ = 0 holds by assumption, we conclude that only those coefficients๐๐ฝwithโจ๐ฟ(๐ด), ๐ฝโฉ = 0 can be nonโzero, and since ๐ผ is a root, some๐๐ฝmust be nonโzero. Hence,๐ฟ(๐ด)โโฉ ๐ด is nonโempty and ๐ฟ(๐ด)โโฉ ๐ด = ๐ถ1โช โฆ โช ๐ถ๐, where๐ถ1, โฆ , ๐ถ๐ are the irreducible components. Thus, if๐ฝ โ ๐ถ๐for some๐ and some ๐ฝ with ๐๐ฝ โ 0, then also ๐ผ โ ฮ๐ถ
๐. Moreover, if๐ผ is of length ๐ฟ, then ฮ๐ถ
๐admits roots of length๐ฟ, so ๐ถ๐ โ ๎ญ๐+1and๐ < ๐.
Proof of theorem 4.9. Put๎ญ๐+1โถ= โ and denote for each ๐ = โ1, โฆ , ๐ by ๐ฟ(๎ญโค๐) the set {๐ฟ(๐ด) | ๐ด โ ๎ญ0โช โฆ โช ๎ญ๐}. We inductively prove that for ๐ = โ1, โฆ , ๐ there exists an element ๐ค โ ๐ (ฮ) such that
(1) every element in(๐ค(ฮ) โฉ ฮ+) โงต ๐ฟ(๎ญโค๐) is contained in ฮ๐ดfor some๐ด โ ๎ญ๐+1and (2) ๐ฟ(๐ฃ(๐ด)) โ ๐ค(ฮ) whenever ๐ผ โ ๐ค(ฮ) โฉ ฮ๐ดfor some๐ด โ ๎ญ1โช โฆ โช ๎ญ๐+1.
For ๐ = โ1 the set ๐ฟ(๎ญโค๐) is empty and ๎ญ0 = {ฮ }, so we may take ๐ค = id in this case. Now suppose that the induction hypothesis holds for some number๐ โค ๐, so there exists ๐ค โ ๐ (ฮ) verifying the two properties above. In particular, there exist elements๐ด1, โฆ , ๐ด๐โ ๎ญ๐+1such that each element ofฮโฒโถ= (๐ค(ฮ) โฉ ฮ+) โงต ๐ฟ(๎ญโค๐) is contained in someฮ๐ด
1, โฆ , ฮ๐ด๐, and we may assume๐ to be the minimal number of elements required to satisfy this property. Thus, we may choose an element๐พ๐ โ ฮโฒโฉ ฮ๐ด
๐for each๐ = 1, โฆ , ๐. Since ฮ๐ด
๐ is reduced and irreducible, all roots of the same length are contained in one Weyl group orbit, so there exists an element ๐ค๐โ ๐ (ฮ๐ด๐) such that ๐ค๐(๐พ๐) is the highest root of ฮ๐ด๐having length๐ฟ, that is, ๐ค๐(๐พ๐) = ๐ฟ(๐ด๐). Now consider the element๐คโฒโถ= ๐ค1โฆ โฆ โฆ๐ค๐. We know from proposition 4.5 that๐คโฒleaves each of the root systemsฮ๐ด
๐invariant, because each๐ค๐is a product of root reflections๐ ๐ผ with๐ผ โ ๐ด๐. The same reasoning combined with corollary 4.7 shows that๐ค๐fixes๐ฟ(๐ด๐) for all ๐ โ ๐ and also all roots in ๐ฟ(๎ญโค๐). Hence, ๐คโฒfixes the elements in๐ฟ(๎ญโค๐), so if we put ฬ๐ค โถ= ๐คโฒโฆ๐ค, then the set ฬ๐ค(ฮ)โฉฮ+fully contains๐ค(ฮ)โฉ๐ฟ(๎ญโค๐) and all of the roots ๐ฟ(๐ด1), โฆ , ๐ฟ(๐ด๐). Moreover, each root๐ผ in ( ฬ๐ค(ฮ) โฉ ฮ+) โงต ๐ฟ(๎ญโค๐+1) is contained in some ฮ๐ด
๐, because the same is true for(๐คโฒ)โ1(๐ผ) โ ฮโฒ. Since the roots inฮ are pairwise orthogonal, such an ๐ผ hence is orthogonal to ๐ฟ(๐ด๐), because ๐คโฒ(๐พ๐) = ๐ฟ(๐ด๐), and therefore already contained inฮ๐ดfor some๐ด โ ๎ญ๐+2by lemma 4.10; in particular, no such๐ผ exists if ๐ = ๐ โ 1. It remains to verify the second property, so suppose that we are given a positive root๐ผ โ ฬ๐ค(ฮ) โฉ ฮ๐ตfor some ๐ต โ ๎ญ1โช โฆ โช ๎ญ๐+2. We already know from the induction assumption that either๐ผ โ ๐ฟ(๎ญโค๐) or ๐ผ โ ฮ๐ด
๐ must
hold, and if๐ผ โ ๐ฟ(๎ญโค๐), then ๐ต must be contained in ๎ญ1โช โฆ โช ๎ญ๐ by corollary 4.7. Since ๐คโฒ fixes๐ฟ(๎ญโค๐) pointwise, the induction statement for๐ shows that ๐ฟ(๐ฃ(๐ต)) must be contained in ฬ๐ค(ฮ) if ๐ผ โ ๐ฟ(๎ญโค๐). If ๐ผ โ ฮ๐ด
๐ for some๐ and ๐ต โ ๎ญ๐+1โ๐for some๐ โฅ 0, then ฮ๐ตandฮ๐ฃ๐(๐ด
๐)intersect nonโtrivially, hence๐ต and ๐ฃ๐(๐ด๐) must be equal by proposition 4.5. Moreover,(๐คโฒ)โ1(๐ผ) and ๐ผ both are contained in ฮ๐ฃ๐(๐ด
๐), because๐คโฒleaves invariant ฮ๐ด๐, so by corollary 4.8(๐คโฒ)โ1must leaveฮ๐ฃ๐(๐ด
๐)andฮ๐ตinvariant as well. Therefore,(๐คโฒ)โ1(๐ผ) is contained in ๐ค(ฮ) โฉ ฮ๐ต, whence by induction assumption๐ฟ(๐ฃ(๐ต)) โ ๐ฟ(๎ญโค๐) is contained in ๐ค(ฮ) and also ฬ๐ค(ฮ). The final case to consider is that๐ผ is an element of some ฮ๐ด
๐, but that๐ต โ ๎ญ๐+2. Then๐ด๐ = ๐ฃ(๐ต), and ๐ฟ(๐ด๐) is contained in ฬ๐ค(ฮ) by construction.
4.2. Normal forms for simply laced root systems
Withis this section, we fix a reduced irreducible root systemฮ whose roots are all of the same length, a set of positive rootsฮ+with corresponding simple rootsฮ , and a nonโempty set of strongly orthogonal roots ฮ โ ฮ. As before, we also setฮ+= ฮ โฉ ฮ+and we additionally suppose that the integer
๐(๐ผ) โถ= ๐(ฮ, ฮ, ๐ผ) โถ= โ
๐ฝโฮ+ 2โจ๐ผ, ๐ฝโฉ
โจ๐ฝ, ๐ฝโฉ
is even for all roots๐ผ. Note that if ๐ค โ ๐ (ฮ) is arbitrary, then ๐(ฮ, ๐ค(ฮ), ๐ผ) still is even, because this number is equal to๐(๐คโ1(๐ผ)). Hence, we may use theorem 4.9 to assume that ฮ+is contained in{๐ฟ๐ด| ๐ด โ ๎ญ0โช โฆ โช ๎ญ๐} and that each๐ฟ(๐ฃ(๐ด)) is contained in ฮ whenever ๐ฟ(๐ด) is an element of ฮ and ๐ด โ ฮ .
Example 4.11 (Normal form forA๐). It will be convenient to associate with any reduced irreducible root system ฮฉ with positive roots ฮฉ+and simple rootsฮฆ a modified Dynkin diagram. By this we shall mean the graph with verticesฮฆ โช {๐ฟ๐}๐, where๐ฟ๐denotes the highest root of length๐ and ๐ ranges over all root lengths in ฮฉ, and whose edge set is built according to the rules of an ordinary Dynkin diagram. The resulting diagram for root systems of typeA๐,๐ โฅ 1, is given in figure 1. If ฮ is of type A๐ and we label the simple rootsฮ = {๐ผ1, โฆ , ๐ผ๐}
๐ผ1 ๐ผ2
๐ฟ
๐ผ๐โ1 ๐ผ๐
Figure 1. Modified Dynkin diagram for root systems of typeA๐,๐ โฅ 1. The highest root is ๐ฟ = ๐ผ1+ โฆ + ๐ผ๐. as in figure 1, we can immediately read off the sets๎ญ0, โฆ , ๎ญ๐. In fact,
๎ญ0= {ฮ }, ๎ญ1={{๐ผ2, โฆ , ๐ผ๐โ1}}, โฆ , ๎ญ๐ ={{๐ผ๐+1, โฆ , ๐ผ๐โ๐}}, โฆ
soฮ+ = {๐ฟ(๐ด0), โฆ , ๐ฟ(๐ด๐)} for some ๐ < โ๐/2โ, where ๐ด๐ = {๐ผ๐+1, โฆ , ๐ผ๐โ๐}. However, the constraint ๐(๐ผ) โ 2โค can only be satisfied if๐ is odd and ๐ = (๐ โ 1)/2, for otherwise ๐ผ๐+1โ ๐ฟ(๐ด๐) is a root and ๐(๐ผ๐+1) = 1. Therefore, ๐ = 2๐ + 1 and ฮ+is equal to{๐ฟ1, โฆ , ๐ฟ๐+1}, where ๐ฟ๐= ๐ผ๐+ โฆ + ๐ผ๐โ๐+1.
Example 4.12 (Normal form forD๐). Suppose thatฮ is of type D๐, ๐ โฅ 4, and enumerate the simple roots ฮ = {๐ผ1, โฆ , ๐ผ๐} as in figure 2. We first assume that ๐ = 2๐ + 1 is odd. Then we have, for ๐ โฅ 1:
๐ผ1 ๐ผ2
๐ฟ
๐ผ๐โ2
๐ผ๐โ1
๐ผ๐
Figure 2. Modified Dynkin diagram for root systems of typeD๐,๐ โฅ 4. The highest root is ๐ฟ = ๐ผ1+ 2๐ผ2+ โฆ + 2๐ผ๐โ2+ ๐ผ๐โ1+ ๐ผ๐.
โฆ , ๎ญ๐ ={{๐ผ2๐โ1}, {๐ผ2๐+1, โฆ , ๐ผ๐}}, โฆ , ๎ญ๐โ1={{๐ผ2๐โ3}, {๐ผ2๐โ1, ๐ผ2๐, ๐ผ2๐+1}},๎ญ๐={{๐ผ2๐โ1}}. Thus, if we let๐ด๐ = {๐ผ๐, โฆ , ๐ผ๐}, then there exists a maximal integer 1 โค ๐ โค ๐ such that ฮ+contains the element ๐ฟ(๐ด2๐โ1), and then ฮ+will also contain๐ฟ(๐ด1), ๐ฟ(๐ด3), โฆ , ๐ฟ(๐ด2๐โ3), because ๐ฃ(๐ด2๐+1) = ๐ด2๐โ1. No element๐ผ2๐โ1 with๐ < ๐ โค ๐ can be contained in ฮ+, for otherwise we could choose๐ maximal with ๐ผ2๐โ1 โ ฮ+, and then๐ผ2๐โ1is the only element ofฮ+not perpendicular to๐ผ2๐, whence๐(๐ผ2๐) = โ1. Similarly, if ๐ผ2๐โ1is contained inฮ+for some 1 < ๐, then ๐ผ2๐โ3is contained inฮ+as well, for otherwise๐(๐ผ2๐โ2) = โ1 would hold. On the other hand, ๐ผ2๐โ1 must be contained inฮ+to ensure๐(๐ผ2๐) โ 2โค, hence ฮ+is equal to{๐ผ1, ๐ฟ(๐ด1), ๐ผ3, ๐ฟ(๐ด3), โฆ , ๐ผ2๐โ1, ๐ฟ(๐ด2๐โ1)}, for some1 โค ๐ โค ๐. Now suppose that ๐ = 2๐. This time we have
๎ญ๐={{๐ผ2๐โ1}, {๐ผ2๐+1, โฆ , ๐ผ๐}}for๐ < ๐ โ 1 and ๎ญ๐โ1={{๐ผ2๐โ3}, {๐ผ2๐โ1}, {๐ผ2๐}}.
We again let๐ด๐= {๐ผ๐, โฆ , ๐ผ๐} and define 1 โค ๐ โค ๐ โ2 to be the maximal integer such that ฮ+contains๐ฟ(๐ด2๐โ1). If๐ < ๐ โ2, then the same argument as in the case of odd rank shows that ฮ+is equal to{๐ผ2๐โ1, ๐ฟ(๐ด2๐โ1) | ๐ โค ๐}. If๐ = ๐ โ2, then an odd (in particular nonโzero) number of elements of {๐ผ2๐โ3, ๐ผ2๐โ1, ๐ผ2๐} must be contained in ฮ+, for otherwise๐(๐ผ2๐โ2) is not even, and if ๐ผ2๐โ3is contained inฮ+, then the same reasoning as in the previous case shows that๐ผ1, โฆ , ๐ผ2๐โ3actually are contained inฮ+. For later reference, let us summarize all the cases we