Skip to main content

Section 7.4 Proofs - Inversion, Kernels, Images

Subsection 7.4.1 Interactions between Definitions

In Section 7.3, I defined a number of terms: rowspace, columnspace, kernel, image and preimage. In this section, I want to prove some properties of these new definitions. All these terms, along with the matrix rank, fit togther very tightly.

Proof.

I know the dimension of the image is the rank, which is the number of leading ones in the reduced row echelon form. I also know that the dimension of the kernel is the dimension of the system of equations represented by the extended matrix \((M|0)\text{.}\) The dimension of a solution space to a system of equations is equal to the number of free parameters. Each column without a leading one in the extended matrix (left of the dividing line) is a free variable. Therefore, the dimension of the kernel is the number of columns without leading ones.
Each column in the row reduced form of \(M\) either has or does not have a leading one. The columns with leading ones contribute to the rank. The columns without leading ones are free parameters for the system. Therefore, the dimensions of the image (rank) plus the dimension of the kernel (free parameters) add up to the number of columns, which is \(n\text{.}\)

Proof.

For this if and only if statement, I need to prove in both directions. I’ll start with the forward implication. For this direction of the proof, I must assume the matrix is invertible.
Since the matrix is invertible, it row reduces to the identity matrix. (I know this from the inversion algorithm, where I must row reduce the left side of the extended matrix to the identity.) That means it has rank \(n\text{,}\) since each row has a leading one. This means that the image has dimension \(n\) since the dimension of the image is equal to the rank. Using the previous proposition, this fact implies that the kernel has dimension \(0\text{.}\) The only thing with dimension zero is a point, so the kernel is a point. The kernel always includes zero, since zero is also always sent to zero, so the kernel must just be the zero vector.
I have to prove the reverse implication as well. Luckily in this case, I can essentially reverse the same logic. A zero kernel has dimension zero. Therefore the image has to have dimension \(n\text{.}\) Therefore, there must be \(n\) leading ones. Therefore, the matrix must reduce to the identity matrix (the only \(n \times n\) matrix with \(n\) leading ones). Thus, the matrix is invertible.
It turns out that there will be many ways to describe an invertible matrix. I’ll start a list of equivalent properties of invertible matrices, which will be extended throughout the course. Assume \(A\) is an \(n \times n\) matrix. The following properties are equivalent (each implies all the rest).
  • \(A\) is invertible.
  • \(A\) row reduces to the identity matrix.
  • \(A\) has rank \(n\text{.}\)
  • The image of \(A\) is \(\RR^n\text{.}\)
  • The columnspace of \(A\) is \(\RR^n\text{.}\)
  • The columns of \(A\) are linearly independent.
  • The rowspace of \(A\) is \(\RR^n\text{.}\)
  • The rows of \(A\) are linearly independent.
  • The kernel of \(A\) is zero.
  • \(A\) preserves the dimension of any (linear or affine) object it acts upon
  • \(Au = v\) has a unique solution \(u\) for any choice of vector \(v \in \RR^n\text{.}\)