Say I have some nonsquare matrix $A$ and a vector $y$, how can I solve $Ax=y$? If $A$ was square, it would be simply $A^{-1}y=x$, but you can't invert a nonsquare matrix.

  • $\begingroup$ What about gaussian elimination? $\endgroup$ – Javi Apr 16 '18 at 19:06
  • 1
    $\begingroup$ Even if $A$ is square it need not be invertible. As Javi says Gaussian elimination is a standard way to solve any matrix equation $Ax=y$. $\endgroup$ – Dave Apr 16 '18 at 19:08
  • 1
    $\begingroup$ If $A\mathbf{x}=\mathbf{b}$ admits one and only one solution, then you can prove that $A^{\top}A$ is invertible. Hence $\mathbf{x}=\left(A^{\top}A\right)^{-1}A^{\top}\mathbf{b}$. If $A\mathbf{x}=\mathbf{b}$ has infinitely many solution, I am afraid there is no explicit form for the general solution... $\endgroup$ – hypernova Apr 16 '18 at 19:10
  • $\begingroup$ You have two completely different cases, according to the fact that $A$ is portrait or landscape (do you know these terms : more lines than columns for portrait, the contrary for lanscape ?). If $A$ is portrait, you have more constraints (equations) than you have unknowns : thus, in general, there are no solutions, unless you are in particular cases. On the contrary, for a landscape matrix, mith $n$ lines and $m$ columns $m>n$, you will have in general a solution depending for exemple on the $m-n$ ($m$ minus $n$) remaining unknowns. $\endgroup$ – Jean Marie Apr 16 '18 at 19:12
  • 1
    $\begingroup$ @AlexLostado: You are right. That is why I mentioned "if $A\mathbb{x}=\mathbb{b}$ admits one and only one solution". If otherwise, this applies to the least square problem only. In addition, if $A\mathbb{x}=\mathbb{b}$ admits no solution, $A^{\top}A$ could also be singular, so it is hard to say that this formulation applies in general. $\endgroup$ – hypernova Apr 16 '18 at 19:24

In general, you can't invert (square) matrices. Look up "Properties of Invertible Matrices" to see what conditions must be satisfied for a matrix to be invertible.

A matrix has to be square to b e invertible, but this does not necessarily hold the other way around.

Gaussian elimination is, I would say, the most used process to determine solutions to the equation

$$A \mathbf{x} = \mathbf{b}$$


$A$ is an $m \times n$ matrix, also written as $A_{m \times n}$ or $A \in \mathbb{M}^{m \times n}$

$\mathbf x$ is an $n \times 1$ vector, also written as $\mathbf{x} \in \mathbb{R}^n$

$\mathbf{b}$ is a $m \times 1$ vector, also written as $\mathbf{b} \in \mathbb{R}^m$


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.