Example problem of [[../02 Areas/Math/Singular Value Decomposition|Singular Value Decomposition]].
$\large
\begin{align}
A &= \mat{
1.1&-0.2&1.5\\
-2.3&-1.4&0.5
}\\
&= \underbrace{U}_{{2\times 2}} \underbrace{\Sigma}_{{2\times 3}} \underbrace{V^{\intercal}}_{3\times 3}
\end{align}
$
$ \large \begin{align}
G =A^\intercal A &= \mat{
\frac{3}{2} & 3 & \frac{1}{2}\\
3&2&-1 \\
\frac{1}{2} & -1 & \frac{5}{2}
} \\
\det(G-tI) &= -t^3 + 11t^2 - 24t\\
&\therefore \begin{matrix}
\lambda_{1} = 6 & \lambda_{2}=3 & \lambda_{3}=0
\end{matrix}
\end{align}
$
$\large \begin{align}
E_{8} &= \op{Nul}(G-8I) \\
&\sim \begin{cases}
x_{1}-2x_{2}&=0\\
x_{3}&= 0 \\
0 &= 0 \\
\end{cases}\\
E_{8} &= \op{span}\left\{ \mat{2\\1\\0} \right\}
\\
\\
E_{3} &= \op{Nul}(G-3I) \\
&= \op{span}\left\{
\mat{
\frac{1}{5} \\
-\frac{2}{5} \\
1 }
\right\}
\\
\\
E_{0} &= \op{Nul}(G-0I) \\
&= \op{span}\left\{
\mat{
-1\\2\\1
}
\right\}
\end{align}
$
$\huge\begin{align*}
\hat v_{1} &= \mat{
\frac{2}{\sqrt{ 5 }} \\
\frac{1}{\sqrt{ 5 }}\\
0
}\\
\hat v_{2} &= \mat{
\frac{1}{\sqrt{ 30 }} \\
\frac{-2}{\sqrt{ 30 }} \\
\frac{5}{\sqrt{ 30 }} \\
}\\
\hat v_{2} &= \mat{
\frac{-1}{\sqrt{ 6 }} \\
\frac{2}{\sqrt{ 6 }} \\
\frac{1}{\sqrt{ 6 }} \\
}\\
\end{align*}
$
$\huge
V = \mat{
\frac{2}{\sqrt{ 5 }} &
\frac{1}{\sqrt{ 30 }} &
\frac{-1}{\sqrt{ 6 }} \\
\frac{1}{\sqrt{ 5 }}&
\frac{-2}{\sqrt{ 30 }} &
\frac{2}{\sqrt{ 6 }} \\
0
& \frac{5}{\sqrt{ 30 }}
&\frac{1}{\sqrt{ 6 }} \\
}
$
$\huge
\begin{align}
\sigma_{1} &= \sqrt{ \lambda_{1} } \\
\sigma_{2} &= \sqrt{ \lambda_{2} } \\
\Sigma &= \mat{
\sqrt{ 8 } & 0 & 0 \\
0 & \sqrt{ 3 } &0
}
\end{align}
$
$\large \begin{align*}
\frac{1}{\sigma_{1}} A \hat v_{1} &= \hat u_{1} =
\mat{
\frac{1}{\sqrt{ 10 }}\\
-\frac{3}{\sqrt{ 10 }}
}
\\
\frac{1}{\sigma_{2}} A \hat v_{2} &= \hat u_{2} =
\mat{
\frac{3}{\sqrt{ 10 }} \\
\frac{1}{\sqrt{ 10 }}
} \\
U &= \mat{
\frac{1}{\sqrt{ 10 }} & \frac{3}{\sqrt{ 10 }}\\
-\frac{3}{\sqrt{ 10 }} & \frac{1}{\sqrt{ 10 }}
}
\end{align*}
$
$\large
A =
\mat{
\frac{1}{\sqrt{ 10 }} & \frac{3}{\sqrt{ 10 }}\\
-\frac{3}{\sqrt{ 10 }} & \frac{1}{\sqrt{ 10 }}
}\mat{
\sqrt{ 8 } & 0 & 0 \\
0 & \sqrt{ 3 } &0
}\mat{
\frac{2}{\sqrt{ 5 }} &
\frac{1}{\sqrt{ 30 }} &
\frac{-1}{\sqrt{ 6 }} \\
\frac{1}{\sqrt{ 5 }}&
\frac{-2}{\sqrt{ 30 }} &
\frac{2}{\sqrt{ 6 }} \\
0
& \frac{5}{\sqrt{ 30 }}
&\frac{1}{\sqrt{ 6 }} \\
}
$
$\frac{1}{\sqrt{ 5 }}\mat{2\\1\\0}$ is the [[Unit Vector]] in $\R^3$ such that $\lvert \lvert A\hat v_{1} \rvert \rvert = \sqrt{ 8 }$ is maximal, $\hat u_{1} \parallel A \hat{v}_{1}$
$\frac{1}{\sqrt{ 30 }}\mat{1\\-1\\5}$ is the [[Unit Vector]] in $\R^3$ such that $\lvert \lvert A\hat v_{2} \rvert \rvert = \sqrt{ 3 }$ is maximal, $\hat u_{2} \parallel A \hat{v}_{2}$
$\frac{1}{\sqrt{ 6 }}\mat{-1\\2\\1}$ is the [[Unit Vector]] in $\R^3$ such that $\lvert \lvert A\hat v_{3} \rvert \rvert = 0$ is minimal
[[../02 Areas/Math/Singular Value Decomposition Truncation|Singular Value Decomposition Truncation]]:
$\op{rank}(A) = 2$, find the best fitting [[../02 Areas/Math/Rank|Rank]] $1$ [[../02 Areas/Math/Matrix|Matrix]] to $A$.
$ \begin{align}
A^\dagger &= U \mat{ \sqrt{ 8 } &0&0 \\0&0&0 }V^\intercal \\
&= \mat{
\frac{1}{\sqrt{ 10 }} & \frac{3}{\sqrt{ 10 }}\\
-\frac{3}{\sqrt{ 10 }} & \frac{1}{\sqrt{ 10 }}
}\mat{
\sqrt{ 8 } & 0 & 0 \\
0 & 0 &0
}\mat{
\frac{2}{\sqrt{ 5 }} &
\frac{1}{\sqrt{ 30 }} &
\frac{-1}{\sqrt{ 6 }} \\
\frac{1}{\sqrt{ 5 }}&
\frac{-2}{\sqrt{ 30 }} &
\frac{2}{\sqrt{ 6 }} \\
0
& \frac{5}{\sqrt{ 30 }}
&\frac{1}{\sqrt{ 6 }} \\
} \\
&= \mat{
0.8 & 0.4 & 0 \\
-2.4 & -1.2 & 0
}
\end{align}
$