Linear Algebra I MT22, Column rank and row rank


Flashcards

If $A$ is a matrix, how can you write $\text{Im }A$ in terms of the column space or row space?

\[\text{Im }A = \text{col}(A)\]

If $A \in \mathbb{F}^{m\times n}$, then how could you write the column space $\text{col}(A)$, or equivalently the image $\text{Im} A$, as a set?

\[\\{Ax : x \in \mathbb{F}^{n\times 1}\\}\]

If $A \in \mathbb{F}^{m \times n}$, then how could you write the row space $\text{row}(A)$, as a set?

\[\\{x^\intercal A:x\in\mathbb{F}^{m\times1}\\}\]

Let $A \in \mathbb{F}^{m \times n}$, and its RREF $R \in \mathbb{F}^{m \times n}$. When proving that the $\text{rowrank}(A) = \text{colrank(A)}$, what long line of equalities do you make?

\[\begin{aligned} \text{rowrank}(A) &= \text{rowrank}(R) \\\\ &= \text{colrank}(R) \\\\ &= \text{colrank}(A) \end{aligned}\]

Let $A \in \mathbb{F}^{m \times n}$. When talking about the kernel or image of a matrix, what linear transformation $T$ are we referring to?

\[T : \mathbb{F}^{n \times 1} \to \mathbb{F}^{m\times 1} \]

Let $A \in \mathbb{F}^{m \times n}$. When talking about $\ker A$ or $\text{Im } A$, we are really referring to the kernel and image of a linear transformation $T$:

\[T : \mathbb{F}^{n \times 1} \to \mathbb{F}^{m\times 1} \]

Therefore, when we apply the rank-nullity theorem to $\dim(\ker A) + \dim(\text{Im } A)$, what is $\dim V$?

\[n\]

Let $A \in \mathbb{F}^{m \times n}$, and its RREF $R \in \mathbb{F}^{m \times n}$. Can you give a quick proof that $\ker A = \ker R$?

\[\begin{aligned} Ax = 0 &\iff EAx = 0 \\\\ &\iff Rx = 0 \end{aligned}\]

as $E$ is invertible (representing the EROs).

Let $A \in \mathbb{F}^{m \times n}$, and its RREF $R \in \mathbb{F}^{m \times n}$. When proving that the $\text{rowrank}(A) = \text{colrank(A)}$, you get to a stage where you need to show $\text{colrank}(A) = \text{colrank}(R)$. What implies this, by the rank-nullity theorem?

\[\ker A = \ker R\]

Bite-sized

Row rank equals column rank: for any $A \in \mathbb F^{m \times n}$, $\mathrm{rowrank}(A) = $ $\mathrm{colrank}(A)$, so the unqualified “rank” is well-defined. In NLA this is just $\mathrm{rank}(A) = $ the number of nonzero singular values of $A$ — the SVD makes the row/column duality manifest.

Source: Basicblock cards above in this entry; NLA MT25 §2.1 of the lecture notes.

@bite~

What does $\mathrm{rank}(A) = r$ tell you about the SVD of $A$?

$A$ has exactly $r$ nonzero singular values, so $A = U _ r \Sigma _ r V _ r^\top$ in “compact” SVD form where $U _ r \in \mathbb R^{m \times r}$, $\Sigma _ r \in \mathbb R^{r \times r}$ with positive diagonal, and $V _ r \in \mathbb R^{n \times r}$. The first $r$ columns of $U$ span the column space ($= \mathrm{Im}\,A$), the first $r$ columns of $V$ span the row space, and the last $n - r$ columns of $V$ span $\ker A$.

Source: NLA MT25 §2.1 of the lecture notes (rank, column/row space, etc).

@bite~