3

Suppose I have an linear transformation described by a square matrix $A$. Then, $A$ takes a vector $x\in\mathbb{R}^n$ to a vector $y\in\mathbb{R}^n$ related to $x$ by the equation $y=Ax$.

Grant Sanderson defined the determinant of $A$ on his youtube series on linear algebra in the following way:

Two vectors $x$ and $y$ determine a unique paralelogram. On the same way $Ax$ and $Ay$ determine another paralelogram. The ratio between its areas is defined to be the determinant of $A$. That is $|\det(A)|=\text{Area}(Ax,Ay)/\text{Area}(x,y)$. (The sign of the determinant is given by the orientation of the vectors. If $A$ preserves orientation, its determinant is positive.)

Can't we define a determinant to non-square matrices by the same formula?

Link to Grant's video: https://www.youtube.com/watch?v=v8VSDg_WQlA

EDIT: (Example) Take $$A=\begin{bmatrix} 1 & 2 \\ 3 & 4 \end{bmatrix}$$ and $$x=\begin{bmatrix} 1 \\ 0\end{bmatrix}, \quad y=\begin{bmatrix} 0 \\ 1\end{bmatrix}.$$ Using the notation defined above, we clearly have $\text{Area}(x,y)=1$. We have that $$Ax=\begin{bmatrix} 1 \\ 3\end{bmatrix}, \quad Ay=\begin{bmatrix} 2 \\ 4\end{bmatrix}$$ too. So, $\text{Area}(Ax,Ay)=2$. Then the ratio between the areas is $2$, which coincides with the modulus of the determinant of $A$. (Of course $A$ changes the orientation of the vectors, so it's determinant is negative.)

EDIT 2: Here's an explicit definition of the determinant. $$|\det(A)| = \begin{cases} \text{Area}(Ax,Ay)/\text{Area}(x,y) & \quad \text{if } \text{kernel}(A) \text{ is zero}\\ 0 & \quad \text{if } \text{kernel}(A) \text{ is non-zero}\\ \end{cases},$$ for any linearly independent vectors $x$ and $y$.

If $A$ preserves orientation, then $\det(A)>0$. $\det(A)\leq 0$ otherwise.

Gabriel
  • 4,513
  • 2
    One can certainly define a quantity as above, but it is not the determinant. – copper.hat Aug 17 '16 at 01:34
  • @copper.hat But it coincides with the determinant when $A$ is a square matrix right? – Gabriel Aug 17 '16 at 01:53
  • @GabrielRibeiro What about an $m\times n$ matrix with $n>m$? The determinant you defined is always zero? – velut luna Aug 17 '16 at 01:58
  • 1
    @AlphaGo The quantity $\text{Area}(Ax,Ay)/\text{Area}(x,y)$ makes sense even if $n>m$. I don't see why that is a problem. – Gabriel Aug 17 '16 at 02:04
  • @GabrielRibeiro Will it depend on the vectors you choose in the definition? – velut luna Aug 17 '16 at 02:04
  • I can prove that it does not in the $n=m=2$ case, but I don't know if it is always true. – Gabriel Aug 17 '16 at 02:05
  • It's straightforward to construct a $4 \times 3$ matrix with linearly independent columns such that $\text{Area}(Ax,Ay)/‌​\text{Area}(x,y)$ depends on the choice of vectors $x$ and $y$, so this is not even well-defined for a given matrix $A$. In any case, the generalization of the geometric interpretation of the determinant to higher dimensions is via parallelepipeds, not parallelograms; see this answer: http://math.stackexchange.com/a/427576/32785 – Jonas Aug 17 '16 at 02:54
  • The formula $|\det(A)|=\mathrm{Area}(Ax,Ay)/\mathrm{Area}(x,y)$ only holds for $2\times2$ matrices. It fails even for square matrices of size $>2$. – stewbasic Aug 17 '16 at 02:55
  • Can you show an example? – Gabriel Aug 17 '16 at 02:59

0 Answers0