Non-group gradings on simple Lie algebras (Q2668070)

This enjoyable paper answers in the negative an old conjecture, providing an example of a (set) grading on the orthogonal Lie algebra \(\mathfrak{so}(26,\mathbb C)\) which cannot be realized as a group-grading. This is later generalized to provide an infinite family of examples of set gradings which are not group gradings: for any natural number \(n\) congruent to \(1\) mod 12 and for any simple split Lie algebra of type \(\mathfrak{so}(2n)\), where the field is arbitrary of characteristic not \(2\). A \textit{set grading} on a Lie algebra \(L\) over a field is simply a decomposition of \(L\) into a direct sum of vector subspaces \(\Gamma:L=\oplus_{s\in S}L_s\), where \(L_s\ne 0\) for any \(s \) in the grading set \(S\), such that the bracket of two homogeneous components \([L_{s_1},L_{s_2}]\) is contained in another homogeneous component. In this case \(\Gamma\) can be \textit{realized as a group grading} if there is a group \(G\) and a one-to-one map \(\iota\colon S\to G\) such that \([L_{g_1},L_{g_2}]\subset L_{g_1g_2}\), where \(L_{\iota{(s)}}:=L_s\) and \(L_g:=0\) if \(g\notin \iota (S)\). Note that if \(L\) is a complex simple finite-dimensional Lie algebra, then the famous root decomposition is a group grading over the group \(G=\mathbb Z\Phi\), for \(\Phi\) the root decomposition relative to a Cartan subalgebra. This tells us that group gradings have been a key piece in Lie theory. Although gradings were present from the beginning of the theory, \textit{J. Patera} and \textit{H. Zassenhaus} initiated in [Linear Algebra Appl. 112, 87--159 (1989; Zbl 0675.17001)] a systematic study of gradings on Lie algebras. One of their assertions was that any set grading on a finite-dimensional Lie algebra can be realized as a semigroup grading. That assertion was a mistake, since some easy counterexamples were found by A. Elduque, the first one on a nilpotent Lie algebra of dimension \(16\), and then simpler examples on a solvable Lie algebra of dimension \(4\) and on a semisimple Lie algebra of dimension \(6\). The next question is up in the air since those works in 2006: Can any set grading on a finite-dimensional simple Lie algebra \(L\) be realized as a group grading? Although no one had provided a counterexample, in general the mathematical community thought that the answer should be affirmative, mainly due to some hints and results as the following: if a grading on \(L\) can be realized as a semigroup grading, then it can also be realized as a group grading (the abelian group generated by the support). Hence this paper has been received as a nice surprise. Besides answering an open question, it contributes to understand some properties of the gradings on simple Lie algebras. The main idea used to construct this example has been to decompose the set of positive roots as a disjoint union of pairs of orthogonal roots. If two roots are orthogonal, then the bracket of their corresponding root spaces annihilates and certain control on these products is got. Every simple Lie algebra of rank \(\ell \) admits a fine group grading (the pieces could not be made smaller) over the group \(\mathbb Z_2^{\ell+1}\), with one of the homogeneous components equal to a Cartan subalgebra: simply take the grading produced by the order two automorphisms in a maximal torus and refine it with an automorphism which is minus the identity in the Cartan subalgebra. The group gradings where a Cartan subalgebra is contained in a homogeneous component \(L_g\) with \(g\neq 1\) were studied by Hesselink, who called them \textit{pure}. In the case of \(D_4\), a coarsening of the above mentioned \(\mathbb Z_2^5\)-grading is a \(\mathbb Z_2^4\)-grading which serves as an inspiration and as a basic piece for the new (non group!) grading on \(D_{13}\): the \(12\) positive roots in \(D_4\) are decomposed in \(6\) pairs of orthogonal roots. Note that this is still a group grading. But now the author decomposes the set of \(156\) positive roots of \(D_{13}\) as a disjoint union of \(78\) pairs of orthogonal roots: starting with the projective plane over the field of \(3\) elements (which has just \(13\) points and \(13\) lines, each one with \(4\) points), he gets \(6\) pairs of positive roots related to each set of 4 points (line) exactly as in the example of \(D_4\). The partition is artful, but it is not difficult at all to prove that it is really a set grading: it is based on the fact that any two different lines meet in a unique point. Instead, the theory of pure gradings is necessary to prove that it can not be realized as a group grading. With little extra effort, the author generalizes this grading to the orthogonal Lie algebra \(\mathfrak{so}(V,b)\) for \(V\) a vector space of dimension \(2n\) over an arbitrary field of characteristic not 2 and \(b\) a nondegenerate symmetric bilinear form of maximal Witt index, for any \(n\) congruent with \(1\) mod \(12\). Then, a combinatorial object called Steiner system gives a collection of subsets of \(I=\{1,\dots,n\}\), or blocks with \(4\) elements, such that any \(2\)-element subset of \(I\) is contained in exactly one block.