Understanding the Cholesky decomposition

491 Views Asked by At

I'm attempting to understand the Cholesky decomposition via the following site: http://en.wikipedia.org/wiki/Cholesky_decomposition

If I have a matrix, say

$$A = \begin{bmatrix} 2 & -1 & 0\\ -1 & 2 & -1\\ 0 & -1 & 2\end{bmatrix},$$ then I'd like to use the Cholesky Algorithm to find a matrix $D$ such that $A = LDL^{T}$.

I left blank the parts of this example that I do not understand how to find.

The Cholesky Algorithm for this example follows:

Skipping a few steps since we don't need to derive how they get $L$, we know that $L = L_{1}L_{2}L_{3}$ since $n=3$.

For $i=1$, $$L_1 = \begin{bmatrix} I_{1-1} & 0 & 0\\ 0 & \sqrt{a_{1,1}} & 0\\ 0 & b_1/\sqrt{a_{1,1}} & I_{3-1}\end{bmatrix} = \begin{bmatrix} 0 & 0 & 0 & 0\\ 0 & \sqrt{2} & 0 & 0\\ 0 & b_1/\sqrt{2} & 1 & 0\\ 0 & 0 & 0 & 1\end{bmatrix}$$

For $i=2$, $$L_ = \begin{bmatrix} I_{2-1} & 0 & 0\\ 0 & \sqrt{a_{2,2}} & 0\\ 0 & b_2/\sqrt{a_{2,2}} & I_{3-2}\end{bmatrix} = \begin{bmatrix} 1 & 0 & 0 \\ 0 & \sqrt{2} & 0\\ 0 & b_2/\sqrt{2} & 1\end{bmatrix}$$

For $i=n = 3$, $$L_3 = \begin{bmatrix} I_{3-1} & 0 & 0\\ 0 & \sqrt{a_{3,3}} & 0\\ 0 & b_3/\sqrt{a_{3,3}} & I_{3-3}\end{bmatrix} = \begin{bmatrix} 1 & 0 & 0 & 0\\ 0 & 1 & 0 & 0\\ 0 & 0 & \sqrt{2} & 0\\ 0 & 0 & b_3/\sqrt{2} & 0\end{bmatrix}$$.

Thus, \begin{align} L &:= L_1\cdot L_2\cdot L_3\\\notag &= \begin{bmatrix} 0 & 0 & 0 & 0\\ 0 & \sqrt{2} & 0 & 0\\ 0 & b_1/\sqrt{2} & 1 & 0\\ 0 & 0 & 0 & 1\end{bmatrix} \cdot \begin{bmatrix} 1 & 0 & 0 \\ 0 & \sqrt{2} & 0\\ 0 & b_2/\sqrt{2} & 1\end{bmatrix} \cdot \begin{bmatrix} 1 & 0 & 0 & 0\\ 0 & 1 & 0 & 0\\ 0 & 0 & \sqrt{2} & 0\\ 0 & 0 & b_3/\sqrt{2} & 0\end{bmatrix}\\\notag &= ? \end{align}

I guess $b_i$ and $b_i^*$ come from $B$. But how do I find the non-Hermitian matrix $B$ to finish this process?

1

There are 1 best solutions below

0
On

Wikipedia’s probably not the ideal place to learn about the Cholesky decomposition or of the various algorithms that can be used to generate the decomposition. I’d recommend reading a textbook on numerical linear algebra or general book on numerical mathematics. That being said, I’ve gone ahead and fixed up your steps in the outer-product Cholesky algorithm example that you’ve chosen to follow.

For the record, I’m using the notation from the following snapshot of the Wikipedia article on the Cholesky decomposition:

https://en.wikipedia.org/w/index.php?title=Cholesky_decomposition&oldid=711015429

Before I run through the algorithm, however, note that $I_i$ denotes the $i \times i$ identity matrix. $I_0$, in particular, refers to a $0 \times 0$ matrix, so when it occurs in an algorithm step it gets omitted from the matrix block partition. Also, all matrices $A^{(i)}$ and $L_i$ should have the same size as the original matrix $A$. Last, all elements $a_{ii}$, $b_i$, and $B$ are pulled directly from matrix $A^{(i)}$ using the partitioning described at the Wikipedia article.

$A = A^{\left( 1 \right)} = \left( {\begin{array}{*{20}c} 2 & { - 1} & 0 \\ { - 1} & 2 & { - 1} \\ 0 & { - 1} & 2 \end{array}} \right)$, so $a_{11} = 2$, $b_1^T = \left( {\begin{array}{*{20}c} { - 1} & 0\end{array}} \right)$, and $B^{(1)} = \left( {\begin{array}{*{20}c} 2 & { - 1} \\ { - 1} & 2 \end{array}} \right)$.

Therefore, $L_1 = \left( {\begin{array}{*{20}c} {\sqrt 2 } & 0 & 0 \\ {\frac{{ - 1}}{{\sqrt 2 }}} & 1 & 0 \\ 0 & 0 & 1\end{array}} \right)$ and $A^{\left( 2 \right)} = \left( {\begin{array}{*{20}c} 1 & 0 & 0 \\ 0 & {\frac{3}{2}} & { - 1} \\ 0 & { - 1} & 2\end{array}} \right)$.

Continuing on, we see that $a_{22} = \frac{3}{2}$, $b_2 = -1$, and $B^{(2)} = 2$, so

$L_2 = \left( {\begin{array}{*{20}c} 1 & 0 & 0 \\ 0 & {\sqrt {\frac{3}{2}} } & 0 \\ 0 & { - \sqrt {\frac{2}{3}} } & 1\end{array}} \right)$ and $A^{\left( 3 \right)} = \left( {\begin{array}{*{20}c} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & {\frac{4}{3}}\end{array}} \right)$.

From $A^{\left( 3 \right)}$ we immediately write down $L_3 = \left( {\begin{array}{*{20}c} 1 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & {\sqrt {\frac{4}{3}} }\end{array}} \right)$.

Therefore, $L = L_1 L_2 L_3 = \left( {\begin{array}{*{20}c} {\sqrt 2 } & 0 & 0 \\ {\frac{{ - 1}}{{\sqrt 2 }}} & {\sqrt {\frac{3}{2}} } & 0 \\ 0 & { - \sqrt {\frac{2}{3}} } & {\sqrt {\frac{4}{3}} }\end{array}} \right)$ which you can confirm satisfies $LL^T = A$.