Show $\Vert L \Vert_\infty \le n$

49 Views Asked by At

Let $A \in \mathbb R^{n,n} $ invertible, $b \in \mathbb R^n$ and suppose to solve $Ax=b$ with $LR$ decomposition with pivoting. Why does $\Vert L \Vert_\infty \le n$ always hold?

$$\Vert L\Vert_\infty=\max_{1\le i\le n}\sum_{j=1}^n \vert l_{ij}\vert$$ I do not see how to continue here. Help is highly needed and appreciated!

1

There are 1 best solutions below

0
On BEST ANSWER

Assuming your pivoting scheme picks the largest element every time (which is more or less the case when numerical stability is important), then w.l.o.g. you can assume that $|a_{jj}^{(j-1)}| \geq |a_{ij}^{(j-1)}|, \forall i$. The notation followed here is the one mentioned in Wikipedia's article.

By the definition of $l_{ij}$, you can see that

$$ \begin{align} l_{ij} &= -\frac{a^{(j-1)}_{ij}}{a^{(j-1)}_{jj}} \Rightarrow \\ |l_{ij}| &= \frac{|a_{ij}^{(j-1)}|}{|a_{jj}^{(j-1)}|} \leq 1 \end{align} $$

Therefore, $\sum_{j=1}^n |l_{ij}| \leq \sum_{j=1}^n 1 = n, \forall i$ which implies that $|| L ||_{\infty} \leq n$.