Contents

MIT 18.06(Linear Algebra) L4 Note

1. Mind Map

https://raw.githubusercontent.com/loss4wang/wx_imagehost/main/LA_Lecture_4.png

2. Reading Notes (2.6)

2.6 Elimination = Factorization : A = LU

$(E_{32}E_{31}E_{21})A = U \quad\text{becomes } A = (E_{21}^{-1}E_{31}^{-1}E_{32}^{-1})U \text{which is } A = LU.$

Explanation and Examples

Special Pattern

$A = \begin{bmatrix} 1 & 1 & 0 & 0 \cr 1 & 2 & 1 & 0 \cr 0 & 1 & 2 & 1 \cr 0 & 0 & 1 & 2 \end{bmatrix} = LU \begin{bmatrix} 1 & & & \cr 1 & 1 & & \cr 0 & 1 & 1 & \cr 0 & 0 & 1 & 1 \end{bmatrix} \begin{bmatrix} 1 & 1 & 0 & 0 \cr & 1 & 1 & 0 \cr & & 1 & 1 \cr & & & 1 \end{bmatrix}$

Assume no row exchanges, we can predict zeros in L and U:

  • When a row of A starts with zeros, so does that row of L.
  • When a column of A starts with zeros, so does that column of U.

Key reason why A equals LU

(Row 3 of A)= $l_{31}$(Row 1 of U) + $l_{32}$(Row 2 of U) + 1(Row 3 of U).

This is exactly row 3 of A = LU. That row of L holds $l_{31}$, $l_{32}$ , 1. All rows look like this, whatever the size of A. With no row exchanges, we have A= LU.

Better balance from LDU

A = LU is “unsymmetric”, because U has the pivots on its diagonal where L has 1’s. Divide U by a diagonal matrix D that contains the pivots. That leaves a new triangular matrix with l’s on the diagonal.

$$ LU =\begin{bmatrix} 1 & 0 \cr 3 & 1 \end{bmatrix} \begin{bmatrix} 2 & 8 \cr 0 & 5 \end{bmatrix} \text{splits further into } LDU= \begin{bmatrix} 1 & 0 \cr 3 & 1 \end{bmatrix} \begin{bmatrix} 2 & 0 \cr 0 & 5 \end{bmatrix} \begin{bmatrix} 1 & 4 \cr 0 & 1 \end{bmatrix} $$

One Square System = Two Triangular Systems

Matrix L contains memory of Gaussian elimination.

How do we use it in solving $Ax = b$?

  • Forward (elimination) and backward (substitution)
    • System $Ax=b$ is factored into two triangular systems
    • Solve $Lc = b$
    • and then solve $Ux=c$

Example:

$$ Ax = b \quad \begin{cases}u + 2v &= 5 \cr 4v + 9v &=21 \end{cases} $$

Forward elimination

$$ Ux = c \quad \begin{cases}u + 2v &= 5 \cr v &=1 \end{cases} $$

$Lc =b$ The lower triangular system $\begin{bmatrix} 1 & 0 \cr 4 & 1 \end{bmatrix} \begin{bmatrix} c_1 \cr c_2 \end{bmatrix} =\begin{bmatrix} 5 \cr 21 \end{bmatrix}$ gave $c =\begin{bmatrix} 5 \cr 1 \end{bmatrix}$

$Ux=c$ The upper triangular system $\begin{bmatrix} 1 & 2 \cr 0 & 1 \end{bmatrix} \begin{bmatrix} x_1 \cr x_2 \end{bmatrix} =\begin{bmatrix} 5 \cr 1 \end{bmatrix}$ gives $x = \begin{bmatrix} 3 \cr 1 \end{bmatrix}$