Suppose $\mathfrak g$ is a finite-dimensional Lie algebra over a field $k$, which we can assume of characteristic zero. In Milne's LAG, Proposition 6.4 claims that $\mathfrak g$ is a reductive Lie algebra if and only if there exists a faithful and completely reducible finite-dimensional representation of $\mathfrak g$.

I understood the proof in the book saying that if $\mathfrak g$ is reductive, we can take the direct sum of the adjoint representation for $Z(\mathfrak g) \oplus \mathcal D \mathfrak g = \mathfrak g$ given by the adjoint representation for $\mathcal D\mathfrak g$ and a direct sum of $1$-dimensional faithful representations for $Z(\mathfrak g)$ given by adding morphisms of the form $k \simeq \mathfrak{gl}(k)$. What I don't understand is the converse, for which nothing is mentioned.

The existence of this faithful completely reducible representation of $\mathfrak g$ implies that we have an inclusion $\mathfrak g \subseteq \mathfrak{gl}(V)$ for some finite-dimensional $k$-vector space $V$. Since $\mathcal D(\mathfrak{gl}(V)) = \mathfrak{sl}(V)$ is semisimple, $\mathrm{rad}(\mathfrak{sl}(V)) = 0$, thus $\mathcal D(\mathfrak g) \subseteq \mathfrak{sl}(V)$ is also semisimple. If $x \in Z(\mathfrak g) \cap \mathcal D \mathfrak g$, this also means $x \in Z(\mathcal D \mathfrak g) = 0$, so we know that $Z(\mathfrak g) \oplus \mathcal D \mathfrak g$ is an ideal of $\mathfrak g$. That's what I managed to do so far.

Two questions (which are kind of linked) :

  • Why doesn't Milne mention anything about this direction? Did he forget or am I missing something obvious?
  • Does anyone have a proof?

Let ${\cal G}$ be a Lie algebra defined over a field of characteristic $0$. Suppose that $D{\cal G}$ is semi-simple, write the Levi Decomposition ${\cal G}=S\oplus rad({\cal G})$, where $S$ is semi-simple and $rad({\cal G})$ solvable. We have $[S,S]=S$, this implies that $S\subset [{\cal G},{\cal G}]$ and $[{\cal G},{\cal G}]=S\oplus U$ where $U\subset rad({\cal G})$ thus $U$ is solvable. We deduce that $[{\cal G},{\cal G}]$ is semi-simple, if and only if $[{\cal G},{\cal G}]=S$. This implies that $[rad({\cal G},rad({\cal G})]=0$ and $rad({\cal G})$ is commutative.

But I don't understand your incomplete argument, you seem to assume that a subalgebra of a semi-simple Lie algebra is also semi-simple and this is not true.

Let $\phi:{\cal G}\rightarrow sl(V)$ be a complete reducible representation of ${\cal G}$, you can write $V=\oplus_iV_i$ as a sum of irreducible module.

Suppose that $dim V_i>1$, let $\phi_i:{\cal G}\rightarrow gl(V_i)$ be the representation induced by $\phi$ on $V_i$. Remark that $[rad({\cal G}),rad({\cal G})]$ is a nilpotent ideal. There exists $x\in V_i$ such that $[rad({\cal G}),rad({\cal G})]x=0$, since $[rad({\cal G}),rad({\cal G})]$ is an ideal, you deduce that $V=\{ x\in V_i, [rad({\cal G}),rad({\cal G})]x=0\}$ is a submodule thus it is $V_i$. This implies that $\phi_i(rad({\cal G})$ is commutative for $dim(V_i)>1$. Since $gl(k)$ is commutative, you deduce that $\phi(rad({\cal G})$ is commutative and $rad({\cal G})$ is commutative since $\phi$ is faithful.

  • $\begingroup$ I like the answer, but I am trying to follow Milne and the Levi decomposition comes later. At least you're telling me that the statement is true, which is still comforting. Thanks $\endgroup$ – Patrick Da Silva Aug 14 '16 at 18:16
  • $\begingroup$ @TsemoAristide The last conclusion should be that $rad(\mathfrak{g})$ is commutative. How does this imply that $\mathfrak{g}$ is reductive? (I know it suffices to prove that $rad(\mathfrak{g}) \subset Z(\mathfrak{g})$.) $\endgroup$ – Dominique R.F. Aug 14 '16 at 19:25
  • $\begingroup$ This is a definition of a reductive Lie algebra: A Lie algebra is reductive if its radical is commutative. $\endgroup$ – Tsemo Aristide Aug 14 '16 at 19:27
  • $\begingroup$ @TsemoAristide I know the three following equivalent definitions of an algebra being reductive : 1) $Z(\mathfrak{g}) = rad( \mathfrak{g})$, 2) $\mathfrak{g} = Z(\mathfrak{g}) \oplus [\mathfrak{g}, \mathfrak{g} ]$ with $[\mathfrak{g}, \mathfrak{g}]$ semisimple and 3) The adjoint representation is completely reducible. Could you please indicate how your definition relates to any of those? A reference would suffice. (All the books that I know use one of the three definitions I have mentioned) $\endgroup$ – Dominique R.F. Aug 14 '16 at 19:32
  • $\begingroup$ In characteristic zero, take the Levi decomposition ${\cal G}=S\oplus rad({\cal G})$, if $rad({\cal G})$ is commutative, then you have 3) since $[S,S]=S$. $\endgroup$ – Tsemo Aristide Aug 14 '16 at 19:34

You can actually find a proof of this claim in Milne, though you have to read it carefully. Right after proposition 6.4, he defines the nilpotent radical $\mathfrak{s}$ of $\mathfrak{g}$, which is the intersection of the kernels of all simple representations of $\mathfrak{g}$. Since $\mathfrak{s}$ is contained in the kernel of any semisimple representation of $\mathfrak{g}$, the hypothesis that $\mathfrak{g}$ has a faithful semisimple representation implies that $\mathfrak{s}=0$.

Milne then proves the following theorem (theorem 6.9):

Let $\mathfrak{g}$ be a Lie algebra, $\mathfrak{r}$ its radical and $\mathfrak{s}$ its nilpotent radical. Then $\mathfrak{s} = [\mathfrak{g}, \mathfrak{r}]$.

Looking at the proof, we see that the full theorem depends on proposition 6.4. However, the inclusion $[\mathfrak{g}, \mathfrak{r}] \subset \mathfrak{s}$ does not. It only depends on lemmas 6.11 and 6.12, which are elementary.

Since $\mathfrak{s}=0$, we conclude that $[\mathfrak{g}, \mathfrak{r}]=0$, so the radical $\mathfrak{r}$ is contained in the center $Z(\mathfrak{g})$. Note that the inclusion $Z(\mathfrak{g}) \subset \mathfrak{r}$ always holds, so we conclude $Z(\mathfrak{g})= \mathfrak{r}$. This is Milne's definition of an algebra being reductive.

In a sense, the $[\mathfrak{g}, \mathfrak{r}] \subset \mathfrak{s}$ part of the theorem should have been stated and proved before proposition 6.4. Proposition 6.4 then follows quite easily.

  • 1
    $\begingroup$ I managed to do it your way (I'm not into Levi decompositions yet, so I can't do it Tsemo's way). So thanks for that. Do you know anything about Corollary 6.10? He seems to use that if $\pi : \mathfrak g \to \mathfrak g/\mathrm{rad}(\mathfrak g)$ is the projection then $\pi(\mathrm{rad}(\mathfrak g)) = \mathrm{rad}(\pi(\mathfrak g))$, which is very wrong (just consider the two-dimensional non-abelian Lie algebra $\mathfrak g$ with basis $\{x,y\}$ with $[x,y] = x$ and send $y$ to zero). I'm disturbed by this book... this might be a different question, but then again maybe not. $\endgroup$ – Patrick Da Silva Aug 22 '16 at 1:46

Here is a possibly different way of doing it:

Claim: Let $\mathfrak g$ be Lie algebra over $\mathbb{R}$, and $I,J$ be two ideals of it such that $I \cap J=0$ and such that $\mathfrak g/I$,$\mathfrak g/J$ are reductive then $\mathfrak g$ is reductive as well. Proof: Let $p:\mathfrak g\rightarrow \mathfrak g /I,q:\mathfrak g\rightarrow \mathfrak g/J$ be the natural quotient maps. Let $A$ be an abelian ideal of $\mathfrak g$, then $p[A]$ is abelian ideal $\mathfrak g/I$ and by reductivity of $\mathfrak g/I$ we get that $p[A]\subseteq Z(\mathfrak g/I)$. Hence: $$p([A,\mathfrak g])=[p(A),p(\mathfrak g)]=[p(A),\mathfrak g/I]=0 \text{ (As $p[A]\subseteq Z(\mathfrak g$/I) ) }$$ So $[A,\mathfrak g]\subseteq I$, similarly we get $[A,\mathfrak g]\subseteq J $. Hence, $[A,\mathfrak g]\subseteq I\cap J=0$, so $A\subseteq Z(\mathfrak g)$. So every abelian ideal of $\mathfrak g$ lives in its center. We also have $$p(Z(\mathfrak g) \cap {\mathfrak g}^2)\subseteq Z(\mathfrak g/I) \cap (\mathfrak g/I)^2=0 \text{ , (By reductivity of $\mathfrak g/I$) } $$ Thus, $Z(\mathfrak g)\cap \mathfrak g^2\subseteq I$, a similar argument gives $Z(\mathfrak g)\cap \mathfrak g^2\subseteq J$, so $Z(\mathfrak g)\cap \mathfrak g^2\subseteq I \cap J=0$, and this completes the proof of reductivity of $\mathfrak g$ $\square$

Now, we prove the statement: "Let $\mathfrak g$ be a finite dimensional $\mathbb{R}$- Lie algebra that admits a faithful completely reducible $\mathfrak g$-module, then $\mathfrak g$ is reductive" by induction on $\dim(\mathfrak g)$.

Proof: Let $V$ be the faithful completely reducible $\mathfrak g$-module. If this representation is irreducible, then we are done because irreducible linear lie algebras are reductive. So assume that the representation is not irreducible. Let $S$ be a proper nontrivial $\mathfrak g$-submodule of $V$. By complete reducibility, we have a $\mathfrak g$-submodule $T$ of $V$ such that $V=S\oplus T$. $V$ is completely reducible , thus so are the $\mathfrak g$-submodules $S,T$. It follows that $S$ is a faithful completely reducible $\mathfrak g/Ann(S)$-module and $T$ is a faithful completely reducible $\mathfrak g/Ann(T)$-module. Hence, by induction we have that $\mathfrak g/Ann(S)$, $\mathfrak g/Ann(T)$ are reductive. By faithfulness, we have $Ann(S)\cap Ann(T)=Ann(V)=0$ . So by previous claim, we are done. $\square$


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.