What is the connection between row reduction and matrix inversion?

[Wicketer]

I'll start off with my question:

Why do we use Gaussian Elimination when inverting a matrix? (this is only one of the methods...which is the only one that doesn't make sense to me).

I know how to do it, but I'm not sure why it works. When solving a system of linear equations, I understand why Gaussian Elimination works: to me, it's just the adding and subtracting of equations until a desirable form is reached. But Gaussian Elimination doesn't seem to be as obvious a tool in matrix inversion.

Thanks!

 

[Petr Mugver]

The Gauss method for finding the inverse B of a matrix A, AB = I, corresponds to solving n linear systems, one for each column of B.

 

[HallsofIvy]

If by "Gaussian elimination" you mean "row reduction", one way to think about it is this:
to every row operation, there exist an "elementary" matrix that you get by applying that row operation to the identity matrix.

That is, if we are working with 3 by 3 matrices and we apply the row operation "add 3 times the second row to the third row" to the identity matrix we get
[tex]\begin{bmatrix}1 & 0 & 0\\ 0 & 1 & 0 \\ 0 & 3 & 1\end{bmatrix}[/tex]

and, further, multiplying that "elementary matrix" by any matrix does that row operation:
[tex]\begin{bmatrix}1 & 0 & 0\\ 0 & 1 & 0 \\ 0 & 3 & 1\end{bmatrix}\begin{bmatrix}a & b & c \\ d & e & f\\ g & h & j\end{bmatrix}= \begin{bmatrix}a & b & c \\ d & e & f \\ g+ 3d & h+ 3e & j+ 3f\end{bmatrix}[/tex].

So, suppose some set of row operations, R1,then R2, ..., Rn reduce the matrix A to the identity matrix I. Call the corresponding elementary matrices M1, M2, ..., Mn so that Mn... M2M1A= I (notice the order- applying R1 first means we have to multiply M1 first). That is the same as (Mn...M2M1)A= I which is exactly saying that Mn...M2M1 is [itex]A^{-1}[/itex]. But now [itex]A^{-1}= Mn...M2M1= Mn...M2M1I[/itex] and multiplying those matrices, in that order, by the indentity matrix is the same as applying the row operations R1, R2, ..., Rn, in that order, to the identity matrix.

That is why "Gaussian Elimination" (row reduction) works. You determine row operations that will reduce the matrix A to the identity matrix while applying the same row operations to the identity matrix. That is the same as determining the matrices M1, M2, etc. while performing the matrix multiplications [itex]Mn...M2M1I= A^{-1}[/itex].