Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Let $T^*$ denote upper triangular matrices (of the appropriate size) with positive diagonal entries and $\mathrm{UT}$ upper triangular matrices with all diagonal entries equal to 1.

Does every (abstract group) embedding $\varphi:\mathrm{UT}(n,\mathbb{R})\to\mathrm{UT}(m,\mathbb{R})$ extend to $\bar{\varphi}:T^*(n,\mathbb{R})\to T^*(m,\mathbb{R})$?

(In the other direction, any $\bar{\varphi}:T^*(n,\mathbb{R})\to T^*(m,\mathbb{R})$ restricts to a homomorphism $\mathrm{UT}(n,\mathbb{R})\to\mathrm{UT}(m,\mathbb{R})$, since $\mathrm{UT}$ is the derived subgroup of $T^*$.)

An affirmative answer to this question implies a relatively easy affirmative answer to this question. I've recently answered the latter independently, but wonder if there's a general result that could be used, and which I should know about. I asked the current question on Math.stackexchange but without success.

EDIT: Florian Eisele has shown that as stated above the answer to the original question is no. This makes me wonder if there's a reasonably natural reformulation for which the answer is yes. For the sake of asking a concrete question, let me hazard the following.

Does every embedding $\varphi:\mathrm{UT}(n,\mathbb{Q})\to\mathrm{UT}(m,\mathbb{Q})$ extend to $\bar{\varphi}:T^*(n,\mathbb{Q})\to T^*(m,\mathbb{Q})$?

share|improve this question
add comment

2 Answers 2

up vote 3 down vote accepted

Here's a counterexample ($m=n=3$) which is a continuous homomorphism and actually probably also works for your second question with $\mathbf{Q}$.

In short: most automorphisms of $UT(3)$ do not extend to $T^*(3)$.

Since I deal with continuous automorphisms, it boils down to a Lie algebra problem. An automorphism of $UT(3)$ induces an automorphism of its abelianized subgroup, giving a $2\times 2$ matrix and every invertible matrix occurs this way. But my claim is that if the automorphism extends to $T^*(3)$ then (some power of) this matrix has to be diagonal.

Sketch of argument: I work upside down by considering an automorphism of $T^1(3)$, the group of determinant 1 matrices in $T^*(3)$ (it's enough and more convenient to deal with it) and describe its restriction to $UT(3)$.

First, because $UT(3)$ is the derived subgroup, it is stable. Write $T^1(3)=D.UT(3)$ where $D$ is the group of diagonal matrices. Since $D$ is a Cartan subgroup in $T^1(3)$ (that is, its Lie algebra is nilpotent and self-normalized, see Bourbaki, Groups and Lie algebras) and is unique up to conjugacy, it is mapped to a conjugate subgroup. So after composing by a conjugation, we can suppose $D$ mapped into itself. Now any automorphism of $T^1(3)$ induces a finite order automorphism of $D$: indeed, any one-parameter group of automorphisms of $T^1(3)$ induces the identity on $D$, as we see using the fact that the derived subgroup of the semidirect product $\mathbf{R}\ltimes T^1(3)$ is nilpotent (hence contained in the nilpotent radical $UT(3)$ of $T^1(3)$). So after conjugation, some power of the automorphism acts on $D$ as the identity and therefore preserves the weight decomposition of the Lie algebra of $UT(3)$ induced by the action of $D$, and thus induces a diagonal automorphism of the derived subgroup of $UT(3)$.

If you don't follow the argument you can also compute the group of continuous automorphisms of $T^*(3)$ and check that no one maps a matrix $\pmatrix{1 & x & *\cr 0 & 1 & y \cr 0 & 0 & 1\cr}$ to a matrix of the form $\pmatrix{1 & x+y & *\cr 0 & 1 & y \cr 0 & 0 & 1\cr}$.

NB: it's unclear from your question if you mean "extends as an automorphism" or "extends as an endomorphism". However, it is not hard to show that an endomorphism of $T^1(3)$ that is injective on $UT(3)$ remains injective on $T^1(3)$, and it easily follows that an automorphism of $UT(3)$ extends as an endomorphism of $T^*(3)$ iff it extends as an automorphism.

share|improve this answer
    
Great answer, thank you! It seems to me that the existence of the sort of extension I had in mind is far from typical in the absence of further assumptions. –  shane.orourke Jul 27 '12 at 18:37
add comment

No (at least if you really mean abstract group embeddings). Choose $m=n=2$. Note that $$UT(2,\mathbb R) \cong (\mathbb R,+)\quad \textrm{ and }\quad T^*(2,\mathbb R)\cong UT(2,\mathbb R)\rtimes (\mathbb R_+,\cdot)^2 $$

Note secondly that the image of the conjugation action of $T^*(2,\mathbb R)$ on $UT(2,\mathbb R)$ is equal to $(\mathbb R_+,\cdot)\leq Aut((\mathbb R,+))$. In particular, $UT(2,\mathbb R)$ splits up into only three orbits under this action. I claim there is an injective homomorphism from $UT(2,\mathbb R)$ into itself such that its image $G$ splits up into many more orbits under the action of the normalizer of $G$ in $T^*(2,\mathbb R)$. Such an embedding clearly cannot be extended to all of $T^*(2,\mathbb R)$.

As a $\mathbb Q$-vector space (and hence also as an abelian group), $\mathbb R\cong \bigoplus_I \mathbb Q$ for some index set $I$ of the same cardinality as $\mathbb R$. Let $S$ be a transcendence basis of $\mathbb R$ over $\mathbb Q$ (so the cardinality of $S$ will be the same as that of $I$). Let $G$ be the $\mathbb Q$-span of $S$. So there is an isomorphism from $(\mathbb R, +)$ to $(G,+)\subsetneq (\mathbb R,+)$. Now if $s_1\neq s_2\in S$, then the unique element in $\mathbb R-\{0\}$ sending $s_1$ to $s_2$ is $r=s_2/s_1$. But if $r\cdot s_2 = s_2^2/s_1$ could be written as a $\mathbb Q$-linear combination $\sum_i c_i s_i$ of elements in $S$, then we would get an algebraic relation $$ s_1(\sum_i c_i s_i)-s_2^2 $$ between the elements in the transcendence basis $S$. Therefore $r\cdot s_2 \notin G$, i.e. $r\cdot G \nsubseteq G$, thus proving that all elements of $S$ lie in different orbits under the action of the normalizer of $G$ in $T^*(2,\mathbb R)$.

share|improve this answer
    
Nice answer, thank you very much. And you're probably right to doubt that abstract group embeddings are the appropriate level of generality to consider: I'll have to think about this and possibly edit the question later. BTW: is the notation $m=n=1$ (rather than 2) above standard? –  shane.orourke Jul 24 '12 at 12:36
    
No, I suppose "m=n=1" was not standard notation. Changed it to "m=n=2". –  Florian Eisele Jul 24 '12 at 12:44
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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