Let $K$ be any field. Consider $GL_n(\mathbb{K})$ as an algebraic group. I know that it has a Gauss decomposition, i.e $GL_n(K)=I^- D I^+$, where $I^-$ and $I^+$ are the lower unipotent matrix and the upper unipotent matrices respectively and $D$ is the set of diagonal matrices in $GL_n(K)$. My question is the following:

Is the following statement true?

For any algebraic subgroup $G$ of $GL_n(K)$ we have the decomposition

$G=(I^-\cap G) (D \cap G) ( I^+ \cap G)$

I know that this is true when for example $G=SL_n(K)$.

Thanks in advance for help and reference.

Added: I think for $SL_n$ or $GL_n$ the above decomposition are incorrect. In order to get the decomposition one should work with their subgroups consisting of matrices with non-trivial principal minors.

After the answer by David my question is:

For what conditions on $G \subset GL_n(K)$ the gauss decomposition is possible?

  • $\begingroup$ The set of $n \times n$ matrices with non-vanishing principal minors is not a group. E.g. if $U = \begin{pmatrix} 1 & 1 \\ 0 & 1\end{pmatrix}$ and $L = \begin{pmatrix} 1 & 0 \\ 1 & 1\end{pmatrix}$ then $L U L^{-1} =\left(\begin{array}{rr} 0 & 1 \\ -1 & 2 \end{array}\right)$. $\endgroup$ Dec 16, 2015 at 16:20

1 Answer 1


This is false. Let $K = \mathbf{R}$, $n = 2$ and let $G$ be the matrices of the form $\begin{pmatrix} x & y \\ -y & x \end{pmatrix}$ with $x^2 + y^2 = 1$. Then $I^- \cap G = I^+ \cap G = \{1\}$ and $D \cap G = \{ \pm 1\}$, so the product $(I^- \cap G) (D \cap G) (I^+ \cap G) = \{\pm 1\}$ is much smaller than $G$.

  • $\begingroup$ @Loeffler: I understand from your argument that the statement in general is false. But what should be the assumptions on $G$ to make such a statement true? $\endgroup$ Dec 16, 2015 at 14:57
  • $\begingroup$ Google "Bruhat decomposition". $\endgroup$ Dec 16, 2015 at 15:09
  • $\begingroup$ In Bruhat decomposition the middle term is an element of the weyl group which is (in general) a permutation matrix. It is not the same as Gauss decomposition in which I am interested. $\endgroup$ Dec 16, 2015 at 15:18
  • $\begingroup$ If you take that viewpoint, then the "Gauss decomposition" does not exist even for $\mathrm{GL}_n$ (except when $n = 1$). You'll have some real trouble decomposing $\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}$. $\endgroup$ Dec 16, 2015 at 16:19
  • $\begingroup$ See math.stackexchange.com/questions/637921 $\endgroup$ Dec 16, 2015 at 16:19

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.