0
$\begingroup$

I'm learning about LU decomposition with a math book, but this question is not about LU decomposition, just wanted to explain why am I wondering about this.

We assume that $A$ is $m×n$ matrix, and $s = min(m,n)$ which compares two values and returns smaller value.

We can express $LU$ decomposition about $A$ with column/row block matrix like this:

$L = \begin{bmatrix}l_1 & \cdots & l_s\end{bmatrix}, U = \begin{bmatrix}u_1\\\vdots\\u_s\\\end{bmatrix}$

But the book expressed $LU$ decomposition to:

$L = \begin{bmatrix}l_1 & \cdots & l_s\end{bmatrix}, U = \begin{bmatrix}u_1^T\\\vdots\\u_s^T\\\end{bmatrix}$

Looks like each element of $U$ has been transposed. How could it be? Is it okay to do this?

$\endgroup$
0
$\begingroup$

I think entry $l_1,\ldots ,l_s$ and $u_1,\ldots ,u_s$ are themselves vectors, not numbers. To be specific, they are column vectors. Thus we line up column vectors a like $[l_1 l_2 \ldots l_s]$ we now get an $m\times s$ matrix, as each $l_i$ is a vector in dimension $m$. If we are stacking the $u_i$ (which are column vectors in dimension $n$) vertically we would get an very tall skinny $(sn)\times 1$ matrix. We don't want that, so we transpose them, turning each into row vectors and then stack them. Thus we get that $U$ is an $s\times n$ matrix.

$\endgroup$

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.