Let $G=GL_n$ and $U_-$ the set of all lower unipotent triangular matrices. Then by Gauss Decomposition, we have $G = U_-B$, where $B$ is the set of all upper triangular matrices. The group $U_-$ acts on $G$ by left multiplication. Since $G = U_-B$, we have an embedding $B \hookrightarrow U_-B/U_- = G/U_-$. Therefore $\mathbb{C}[G/U_-] \hookrightarrow \mathbb{C}[B]$. We have $\mathbb{C}[G/U_-] = \oplus_{\lambda} V_{\lambda}$, where $\lambda$ is a dominant weight and $V_{\lambda}$ is the highest weight module of $U(\mathfrak{g})$ with highest weight $\lambda$, where $\mathfrak{g}$ is the Lie algebra of $G$.

My question is: how to write $\mathbb{C}[G/U_-]=\oplus_{\lambda} V_{\lambda}$ explicitly?

For example, let $G = GL_3$. Then \begin{align} \mathbb{C}[G/U_-] = \mathbb{C}[g_{11}, g_{12}, g_{13}], \end{align} where $g_{11}, g_{12}, g_{13}$ are coordinate functions which sends a matrix $x$ to matrix coefficients $x_{11}, x_{12}, x_{13}$ respectively. How to write the $V_{\lambda}$ in \begin{align} \mathbb{C}[G/U_-] = \mathbb{C}[g_{11}, g_{12}, g_{13}]=\oplus_{\lambda} V_{\lambda} \end{align} explicitly? Thank you very much.

  • 8
    $\begingroup$ It is not true that $G = U_- B$ (unless $n = 1$, I guess). The set $U_- B$ is dense and open, but it is only the "big cell" of the Bruhat decomposition, and there are other (lower-dimensional) cells that are needed to get all of $G$. However, you are right that there is an embedding $\mathbb{C}[G/U_-] \hookrightarrow \mathbb{C}[B]$. (In your example, don't you also need the functions $g_{22}$, $g_{23}$, and $g_{33}$?) $\endgroup$ May 3, 2016 at 3:40

2 Answers 2


The subspace $V_\lambda$ is very easy to see. Since $U^-$ is normalized by the maximal torus $T$ there is an action of $T$ on $G/U^-$ on the right. This means that $\mathbb C[G/U^-]$ carries a multigrading and $V_\lambda$ is just one of the multihomogeneous pieces.

The problem is now that $\mathbb C[G/U^-]$ is very complicated as a ring. The case $G=SL(2)$ is deceiving. Here $\mathbb C[G/U^-]$ is indeed a polynomial ring in two variables and the $V_\lambda$ are the binary forms of degree $\lambda$.

In general, $\mathbb C[G/U^-]$ is generated by the coordinate functions on all fundamental representations. The relations can be shown to be all quadratic, a generalization of the Plücker relations. For example for $G=GL(n)$ the ring is generated by all "left justified" $m\times m$-minors for $m=1,\ldots,n$.

So your description of $\mathbb C[G/U^-]$ for $G=GL(3)$ is incorrect. In fact $G/U^-$ is $6$-dimensional. The right hand side is the affine cone over a partial flag variety namely the projective plane, hence affine $3$-space.

The description of $\mathbb C[G/U^-]$ as a ring is precisely the purpose of standard monomial theory. There, one constructs a set of monomials which forms a $\mathbb C$-basis and describes multiplication by straightening laws.

  • $\begingroup$ thank you very much. Is the action of $T$ on $G/U^-$ given by the following formula $G/U^- \times T \to G/U^-$, $(gU^-, t) \mapsto gt'U^-$ ($t'$ satisfies $t'U^- = U^- t$)? Since we have an action $G/U^- \times T \to G/U^-$, there is a coaction $\mathbb{C}[G/U^-] \to \mathbb{C}[G/U^-] \otimes \mathbb{C}[T]$. What is the corresponding multigrading on $\mathbb{C}[G/U^-]$? $\endgroup$ May 3, 2016 at 8:40
  • $\begingroup$ First: $t'=t$. Second: $f$ is homogeneous of degree $\lambda$ if the coaction sends $f$ to $f\otimes\lambda$. $\endgroup$ May 3, 2016 at 10:59
  • $\begingroup$ I have another question. How to write the map $\mathbb{C}[G/U_-] \hookrightarrow \mathbb{C}[B]$ explicitly? For example, let $G = GL_2$. Then we have $\mathbb{C}[G/U_-] = \mathbb{C}[g_{11}, g_{21}, g_{11}g_{22} - g_{12} g_{21}]$ and $\mathbb{C}[B] = \mathbb{C}[b_{11}, b_{12}, b_{22}]$. What are the images of $g_{11}, g_{21}, g_{11}g_{22} - g_{12} g_{21}$? $\endgroup$ May 3, 2016 at 12:33

It's slightly nicer to look at $M_n // U_-$ instead of $GL(n) // U_-$, since then we're looking at a subring of invariants inside a polynomial ring. Namely, the subring generated by all determinants that use any $k$ rows and the left $k$ columns (for all $k=1,\ldots,n$); there are $2^n-1$ choices of row set.

These Plücker coordinates form a SAGBI basis for this ring, making it easy to write down a basis for each $T\times T$-weight space, in bijection with the relevant Gel$'$fand-Cetlin patterns. (Specifically, there are $2^n-1$ patterns consisting of only $0,1$, which are in obvious bijection with the Plücker coordinates. Then any pattern is canonically (though not uniquely) a sum of these basic patterns, suggesting the corresponding monomial.) This is in Miller and Sturmfels' "Combinatorial Commutative Algebra", chapter 14.


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.