The answers to this question indicate that there is a fairly vast literature on the index of Lie algebras. Unfortunately I was not able to find in this literature (or maybe to extract from it) an answer to the following simple-looking question:

Is it true (and under which conditions) that for a Lie algebra $\mathfrak{g}$ with a Levi decomposition $\mathfrak{g}=\mathfrak{h}\triangleright\mathfrak{s}$, where $\mathfrak{s}$ is solvable and $\mathfrak{h}$ is semisimple, we have $$\mathrm{ind}\ \mathfrak{g}\leq \mathrm{ind}\ \mathfrak{h}+\dim\mathfrak{s},$$ and, if yes, when the above inequality is strict (i.e. has $<$ instead of $\leq$) or, on the other hand, when does it turn into an equality (except for the trivial case when $\mathfrak{g}$ is abelian)?

More broadly, are there any other inequalities estimating the index of a Lie algebra from above (or perhaps from below) using its Levi decomposition?

Many thanks in advance.

NOTE: The original question, to which Francois Ziegler provided the counterexample in his answer, concerned the opposite inequality $\mathrm{ind}\ \mathfrak{g}\geq \mathrm{ind}\ \mathfrak{h}+\dim\mathfrak{s}$, but this very answer has motivated me to modify the question.

  • $\begingroup$ If someone gives a good answer, you should not modify the question to the opposite. It would be better to accept the answer, and then post a better question. $\endgroup$ Dec 9 '14 at 11:10

Your initially proposed inequality fails if $G$ is Euclid's group $\mathrm{SO}(3)\ltimes\mathbf R^3$. Indeed we have $\operatorname{ind}\mathfrak g = 2$ (generic coadjoint orbits of $G$ have dimension 4), $\operatorname{ind}\mathfrak{so}(3) = 1$ (generic coadjoint orbits of $\mathrm{SO}(3)$ have dimension 2), hence $ \operatorname{ind}\mathfrak g < \operatorname{ind}\mathfrak{so}(3) + \dim\mathbf R^3. $

Edit. The reverse inequality, on the other hand, is true (under the sole assumption that $\mathfrak s$ is an ideal: it needs not be the solvable radical, nor a semidirect factor). Indeed, dualizing $0\to\mathfrak s\to\mathfrak g\to\mathfrak h\to0$ gives an exact sequence $$ 0\to\mathfrak h^*\overset\iota\hookrightarrow\mathfrak g^*\to\mathfrak s^*\to0 $$ by means of which each coadjoint orbit $\mathcal O$ of $H=G/S$ becomes a coadjoint orbit $\iota(\mathcal O)$ of $G$ (on which $S$ acts trivially). Now take $\mathcal O$ to have maximal dimension among coadjoint orbits of $H$. Then we have \begin{align} \operatorname{ind}\mathfrak g &\leqslant \operatorname{codim}_{\mathfrak g^*}\iota(\mathcal O)\\ %&=\dim\mathfrak g^*-\dim\iota(\mathcal O)\\ &=\dim\mathfrak g^*-\dim\mathcal O\\ &=\dim\mathfrak g^*- (\dim\mathfrak h^*-\operatorname{codim}_{\mathfrak h^*}\mathcal O)\\ &=\dim\mathfrak s + \operatorname{ind}\mathfrak h. \end{align} References. You may be interested in §40.4 "Index and semi-direct products" in this book (2005). Also, see A. Panasyuk, Reduction by stages and the Raïs-type formula for the index of a Lie algebra with an ideal (2008).

  • $\begingroup$ Many thanks! Your answer suggests that the opposite inequality is far more likely to hold in some generality, and is hence more interesting, so I changed the question accordingly (and restored the reference-request tag). $\endgroup$ Dec 7 '14 at 19:48
  • $\begingroup$ Extra thanks for the added reference, will have a look. $\endgroup$ Dec 7 '14 at 20:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.