2
$\begingroup$

I was trying to find the determinant for $A^TA$ where $$ A = \left( \begin{array}{ccc} a & b & c \\ d & e & f \\ \end{array} \right) $$ I tried out with some numbers in place of $a, b, c, \dots$ and saw that the determinant is always $0$. Is there any way to prove this?

One thing I thought of, was writing a matrix $B$

$$ B = \left( \begin{array}{ccc} a & b & c \\ d & e & f \\ 0 & 0 & 0 \\ \end{array} \right) $$ $B^TB$ is always the same as $A^TA$ because the $0$ elements don't contribute to anything, whereas both $B$ and $B^T$ are clearly singular, but I don't have a rigorous way to show that $B^TB = A^TA$ other than multiplying out the matrices which is the very thing I'm trying to avoid.

So, how do I show that the product of a matrix and it's transpose, when they are not square, is singular?

$\endgroup$
  • $\begingroup$ OP switched to $A^T A$. $\endgroup$ – mvw Jan 25 '16 at 18:47
  • $\begingroup$ Sorry for not pointing it out @mvw $\endgroup$ – Aritra Das Jan 25 '16 at 18:49
4
$\begingroup$

Note that $A$ (dimension $2\times 3$) has rank at most 2 and so cannot have full column rank. That is, there is $v$ ($3\times 1$) not identically $0$ such that $$ Av=0. $$ But then $(A'A)v=A'(Av)=0$, implying that $A'A$ cannot be invertible. Alternatively, note that $(A'A)v=0$ and $v\neq 0$ expose the fact that $A'A$ (a $3\times 3$ matrix) does not have full (column) rank.

$\endgroup$
3
$\begingroup$

In general $\text{rank}(AB) \le \max(\text{rank}(A), \text{rank}(B))$. In particular, if $A$ and $B$ are $m \times n$ and $n \times m$ with $n < m$, $AB$ must be singular. However, it's incorrect to say "the product of a matrix and its transpose, when they are not square, is singular": if $A$ is $m \times n$ with $n < m$ , $A A^T$ (which is $m \times m$) is singular, but $A^T A$ (which is $n \times n$) may well be nonsingular.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.