The Unapologetic Mathematician

Mathematics for the interested outsider

Elementary Matrices

Today we’ll write down three different collections of matrices that together provide us all the tools we need to modify bases.

First, and least important, are the swaps. A swap is a matrix that looks like the identity, but has two of its nonzero entries in reversed columns.

\displaystyle W_{i,j}\begin{pmatrix}1&&&&&&0\\&\ddots&&&&&\\&&0&\cdots&1&&\\&&\vdots&\ddots&\vdots&&\\&&1&\cdots&0&&\\&&&&&\ddots&\\{0}&&&&&&1\end{pmatrix}

where the two swapped columns (or, equivalently, rows) are i and j. The swaps generate a subgroup of \mathrm{GL}(n,\mathbb{F}) isomorphic to the symmetric group S_n. In fact, these are the image of the usual generators of S_n under the permutation representation. They just rearrange the order of the basis elements.

Next are the scalings. A scaling is a matrix that looks like the identity, but one of its nonzero entries isn’t the identity.

\displaystyle C_{i,c}=\begin{pmatrix}1&&&&&&0\\&\ddots&&&&&\\&&1&&&&\\&&&c&&&\\&&&&1&&\\&&&&&\ddots&\\{0}&&&&&&1\end{pmatrix}

where the entry c is in the ith row and column. The scalings generate the subgroup of diagonal matrices, which is isomorphic to \left(\mathbb{F}^\times\right)^n — n independent copies of the group of nonzero elements of \mathbb{F} under multiplication. They stretch, squeeze, or reverse individual basis elements.

Finally come the shears. A shear is a matrix that looks like the identity, but one of its off-diagonal entries is nonzero.

\displaystyle H_{i,j,c}=\begin{pmatrix}1&&&&&&0\\&\ddots&&&&&\\&&1&&c&&\\&&&\ddots&&&\\&&&&1&&\\&&&&&\ddots&\\{0}&&&&&&1\end{pmatrix}

where the entry c is in the ith row and jth column. If i<j, then the extra nonzero entry falls above the diagonal and we call it an “upper shear”. On the other hand, if i>j then the extra nonzero entry falls below the diagonal, and we call it a “lower shear”. The shears also generate useful subgroups, but the proof of this fact is more complicated, and I’ll save it for its own post.

Now I said that the swaps are the least important of the three elementary transformations, and I should explain myself. It turns out that swaps aren’t really elementary. Indeed, consider the following calculation

\displaystyle\begin{aligned}\begin{pmatrix}1&1\\{0}&1\end{pmatrix}\begin{pmatrix}1&0\\-1&1\end{pmatrix}\begin{pmatrix}1&1\\{0}&1\end{pmatrix}\begin{pmatrix}-1&0\\{0}&1\end{pmatrix}&\\=\begin{pmatrix}0&1\\-1&1\end{pmatrix}\begin{pmatrix}1&1\\{0}&1\end{pmatrix}\begin{pmatrix}-1&0\\{0}&1\end{pmatrix}&\\=\begin{pmatrix}0&1\\-1&0\end{pmatrix}\begin{pmatrix}-1&0\\{0}&1\end{pmatrix}&\\=\begin{pmatrix}0&1\\1&0\end{pmatrix}\end{aligned}

So we can build a swap from three shears and a scaling. It should be clear how to generalize this to build any swap from three shears and a scaling. But it’s often simpler to just thing of swapping two basis elements as a single basic operation rather than as a composition of shears and scalings.

On the other hand, we can tell that we can’t build any shears from scalings, since the product of scalings is always diagonal. We also can’t build any scalings from shears, since the determinant of any shear is always {1}, and so the product of a bunch of shears also has determinant {1}. Meanwhile, the determinant of a scaling C_{i,c} is always the scaling factor c\neq1.

August 26, 2009 Posted by | Algebra, Linear Algebra | 5 Comments