The power method is an [[Iterative Method]] to find the [[Dominating Eigenvalue]] of a [[Matrix]].
Let $A$ be an $n\times n$ [[Square Matrix]] that is [[Eigendecomposition|Diagonalizable]] with [[Eigenvector|Eigenvectors]] $\set{\vec\lambda_{1},\vec\lambda_{1},\dots}$ forming a basis of $n$, with respect to [[Eigenvalue|Eigenvalues]] $\set{\lambda_{1},\lambda_{2},\dots}$, assuming the eigenvalues are ordered.
$ \huge |\lambda_{1}| \geq |\lambda_{2}| \geq \cdots \geq |\lambda_{n}| $
We start with a non zero initial guess $\vec x^{(0)}$.
$ \huge \vec x^{(k+1)} = \frac{1}{\norm{ A\vec x^{(k)} } } A \vec x^{(k)} $
Note this method applies for any [[Norm]].
$\huge \begin{align}
\vec x^{(0)} &= c_{1} \vec \lambda_{1} + c_{2} \vec \lambda_{2} + \cdots \\
\text{{with }} & c_{1},\dots,c_{n} \in \R \\
\end{align} $
$
\huge \begin{align}
A^{k} \vec x^{(0)} &= c_{1} A^{k}\vec \lambda_{1} + \cdots + c_{n} A^{k}\vec \lambda_{n} \\
&= c_{1} \lambda_{1} \vec \lambda_{1} + \cdots + c_{n} \lambda_{n} \vec \lambda_{n} \\
\text{if } c_{1} \neq 0 &\wedge |\lambda_{1}| > |\lambda_{2}| \\
A^{k} \vec x^{(0)} &= \lambda_{1}^{k} \pa{
c_{1} \vec \lambda_{1} + \frac{\lambda_{2}^{k}}{\lambda_{1}^{k}} c_{2} \vec \lambda_{2} + \cdots + \frac{\lambda_{n}^{k}}{\lambda_{1}^{k}} c_{n} \vec \lambda_{n}
}
\end{align}
$
Because $\lambda_1$ is dominant, all coefficients of the form $\frac{\lambda_{n}^{k}}{\lambda_{1}^{k}}$ will approach $0$ as ${ n \to \infty }$.
If a given [[Eigenvalue]] $\lambda$ has [[Multiplicity]] higher than one, then that the method works - but this method struggles if two [[Eigenvalue|Eigenvalues]] have the same [[Absolute Value]]. A common appearance of this is having two eigenvalues that are [[Complex Conjugate|complex conjugates]] of one another.
This method will often fail to converge if $\lambda_{1} \in \C$ and $\lambda_{2}=\lambda_{1}^{*}$.
>[!example]
>An example of what this sequence will do to a matrix whose dominating eigenvalues are conjugates.
>![[Pasted image 20260323155613.png|500]]