6
$\begingroup$

A colleague of mine asked me the question below, and since I could not answer it, I thought I might have more luck on MO.

In Encyclopedia of Mathematics, a finite dimensional Lie algebra $L$ over a field$\newcommand{\bK}{\mathbb{K}}$ $\bK$ is defined to be supersolvable if all eigenvalues of all operators $\DeclareMathOperator{\ad}{ad}$ $\ad(X)$ belong to $\bK$. It is then stated that any supersolvable Lie algebra over a field of characteristic $0$ can be isomorphically imbedded in the Lie algebra of upper-triangular matrices with coefficients from $\bK$. Does anyone know a proof or a reference for the last statement?

Edit 1. Here is the precise reference: Lie algebra, supersolvable. V.V. Gorbatsevich (originator), Encyclopedia of Mathematics.

Edit 2. Let me refine the original question following YCor's suggestions. Let $\frak{g}$ be a Lie algebra over $\mathbb{R}$. Let’s call $\frak{g}$ supersolvable if it admits a complete flag made up of ideals. (By the way, Knapp calls such algebras split-solvable.)

Gorbatsevich-Onishchik-Vinberg state without a proof (see Lie groups and Lie algebras III, p.50) that if $\frak{g}$ is supersolvable, then it is triangulable. Any idea how to prove it?

Clearly not every faithful representation of a supersolvable $\frak{g}$ is triangulable, even for abelian $\frak{g}=(\mathbb{R},+)$, so Ado’s theorem is not enough.

$\endgroup$
6
  • $\begingroup$ I'm not sure where this definition is stated, but your version doesn't make sense. Can you be more exact about sources (and try to avoid anonymous sources)? $\endgroup$ May 20, 2015 at 20:44
  • $\begingroup$ @JimHumphreys: I agree that precise references would be useful, but the question sounds precise to me. (It is easy to see that a finite-dimensional Lie algebra in characteristic zero is supersolvable iff it has a complete flag consisting of ideals, which is possibly a more common definition for "supersolvable".) $\endgroup$
    – YCor
    May 20, 2015 at 21:30
  • $\begingroup$ @JimHumphreys I have added the precise reference. $\endgroup$ May 21, 2015 at 0:58
  • $\begingroup$ As spotted by Dave, the assumption "solvable" is obviously forgotten in the wiki you link at, and should be part of your definition of "supersolvable". $\endgroup$
    – YCor
    May 21, 2015 at 8:22
  • 1
    $\begingroup$ Why have you restricted the question to the real case? it sounds natural for a general field of characteristic zero. $\endgroup$
    – YCor
    May 21, 2015 at 14:38

2 Answers 2

6
$\begingroup$

I've checked the published book ((on Springer's site) Lie groups and Lie algebras III by Gorbatsevich-Onishchik-Vinberg, from which the site linked by Liviu most likely refers.

The book contains a wealth of results, but with few proofs, and indeed makes the mistake spotted by Dave. So let me clarify the picture.

Let $K$ be a field. Let's say that a finite-dimensional Lie algebra $\mathfrak{g}$ over $K$ is

  • pointwise ad-triangulable if for every $x\in\mathfrak{g}$, all eigenvalues of $\mathrm{ad}(x)$ are in $K$;
  • supersolvable if it admits a complete flag made up of ideals
  • triangulable if it is isomorphic to a Lie subalgebra of the algebra of upper triangular matrices in large enough dimension.

Immediate implications:

  • supersolvable implies solvable and pointwise ad-triangulable
  • triangulable implies supersolvable;

If $K$ has characteristic zero, the first implication is an equivalence:

  • (char$(K)=0$) solvable and pointwise ad-triangulable implies supersolvable.

Indeed if $\mathfrak{g}$ is abelian there is nothing to do, otherwise the center $\mathfrak{z}$ of $[\mathfrak{g},\mathfrak{g}]$ (which is nilpotent; I possibly use char. zero here) is nontrivial and the adjoint action of $\mathfrak{g}$ on $\mathfrak{z}$ factors through an abelian group and has an common eigenvector as it consists of a commuting family of trigonalizable operators. [This argument works with no characteristic assumption if $\mathfrak{g}$ is assumed nilpotent-by-abelian from the beginning, i.e. has nilpotent derived subalgebra.]

  • So I guess the post should be corrected as whether (in char. zero) supersolvable implies triangulable. Actually in the above-quoted book, they say that it can be proved by following the proof of Ado's theorem (which is quite complicated). [But as I mentioned as a common, it is not enough to just evoke Ado's theorem and then use the Lie-Kolchin theorem, because the representation produced by Ado's theorem could fail to be conjugate to a triangular representation, even when we start from an abelian Lie algebra.]

  • Finally, we can wonder about the implication whether pointwise ad-triangulable implies supersolvable, which, in characteristic zero (as I henceforth assume), is, by the above, equivalent to wonder whether pointwise ad-triangulable implies solvable. As I said, this implication is claimed to always hold but Dave mentioned that it clearly fails when the field is algebraically closed. The authors were misled by the fact they had the real case in mind, in which case the implication does hold. To clarify, let me state the general case:

Proposition. Let $K$ be a field of characteristic zero. Equivalences: (i) Every finite-dimensional pointwise ad-triangulable Lie algebra over $K$ is solvable (and hence supersolvable). (ii) $K$ admits a field extension of degree 2.

Examples of fields with no extension of degree 2 (so that pointwise ad-triangulable fails to imply solvable) are not only algebraically closed field, but also, for instance, the field generated by iterated square roots of rationals. On the other hand the reals, number fields, and finite extensions of $p$-adic fields, admit extensions of degree 2.

Proof of the proposition: assume that $K$ admits an extension of degree 2 (i.e. admits a non-square $t$). It's enough to show that there is no simple Lie algebra $\mathfrak{g}$ over $K$ that is pointwise ad-triangulable. Indeed, since $\mathfrak{g}$ admits a Cartan subalgebra defined over $K$, we immediately see that $\mathfrak{g}$ is $K$-split so admits $\mathfrak{sl}_2(K)$ as a subalgebra. Notice that $\mathfrak{sl}_2(K)$ contains an element $x$ such that ad$(x)$ is not trigonalizable: just pick $x=\begin{pmatrix}0 &t\\ 1 & 0\end{pmatrix}$; then the characteristic polynomial of ad$(x)$ (in $\mathfrak{sl}_2(K)$) is $X(X^2-4t)$, which is not split since $t$ is not a square, and in $\mathfrak{g}$, the characteristic polynomial of ad$(x)$ is divisible by $X(X^2-4t)$, so is not split. Thus $\mathfrak{g}$ is not pointwise ad-triangulable.

Conversely if every element in $K$ is a square, then for every $x\in\mathfrak{sl}_2(K)$, the characteristic polynomial $\chi(X)$ of ad$(x)$ has degree 3 and is divisible by $X$, and hence is split (since the degree 2 quotient $\chi(X)/X$ is then split using that the discriminant is a square); thus in this case $\mathfrak{sl}_2(K)$ is pointwise ad-trigonalizable but not solvable.

$\endgroup$
1
  • $\begingroup$ Thanks for your clarifications. I am adding a refined question following your suggestions. $\endgroup$ May 21, 2015 at 14:08
4
$\begingroup$

This is false. If $\mathbb{K}$ is algebraically closed, then it is obvious that all eigenvalues are in $\mathbb{K}$, but many Lie algebras (such as $\mathfrak{sl}_2(\mathbb{K})$) are not solvable and therefore cannot be embedded in the upper-triangular matrices.

However, by a standard argument (the proof of the Lie-Kolchin Theorem), it is not difficult to see that every solvable Lie algebra with this property can be embedded in the upper-triangular matrices. Since every semisimple Lie algebra over $\mathbb{R}$ contains $\mathfrak{sl}_2(\mathbb{R})$, and therefore has elements with non-real eigenvalues, this implies that the result is true for the special case $\mathbb{K} = \mathbb{R}$. By looking at the web page cited by the OP's reference, it seems that the mistake is an erroneous generalization from $\mathbb{R}$ to all fields of characteristic zero.

The correct definition of supersolvable is given in YCor's comment. It does imply that all eigenvalues of every $\mathrm{ad}(X)$ are in $\mathbb{K}$, but it is not always equivalent to this condition.

$\endgroup$
2
  • 4
    $\begingroup$ Dave, a supersolvable Lie subalgebra $\mathfrak{h}$ of the algebra $\mathfrak{gl}_n(K)$ of matrices (even in the real case) need not be conjugate (over the ground field $K$) to upper triangular matrices. This is even false if $\mathfrak{h}$ is abelian! So I'm skeptical about just saying that it's enough to apply Lie-Kolchin; one issue is to find a good linear representation. $\endgroup$
    – YCor
    May 21, 2015 at 7:56
  • $\begingroup$ YCor is correct. Applying the argument from the proof of Lie-Kolchin to the adjoint representation can only show that the image of the adjoint representation is conjugate to an upper-triangular algebra. It cannot not say anything at all about the center of the Lie algebra. $\endgroup$ May 22, 2015 at 3:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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