Foundations — Diagonalization — conditions, procedure

This page assumes you know nothing about the notation in the parent note. We build every symbol, one at a time, each resting on the one before. If a line uses a symbol we have not yet earned, that is a bug — flag it. Let's start from an arrow.

1. A vector — the arrow

The picture: look at the figure. The arrow from the origin to the point $(3,1)$ is the vector $\left(\begin{array}{c}3\\ 1\end{array}\right)$. The number $x_1=3$ is its shadow on the horizontal axis; $x_2=1$ is its shadow on the vertical axis.

Why the topic needs it: everything a matrix does, it does to arrows. Diagonalization is a statement about which arrows behave simply, so arrows are our first object. See Change of Basis for how the same arrow can be described by different number-pairs when you tilt your graph paper.

2. A matrix — the machine

How the machine eats an arrow (matrix–vector product): $A\mathbf{x}=\left(\begin{array}{cc}4& 1\\ 2& 3\end{array}\right)\left(\begin{array}{c}{x}_1\\ x_2\end{array}\right)=\left(\begin{array}{c}4x_1+1x_2\\ 2x_1+3x_2\end{array}\right).$

The picture: the grid on the left is normal graph paper. On the right the machine $A$ has warped it — squares become slanted parallelograms. Most arrows changed both length and direction.

Why the topic needs it: the whole subject is "understanding what this machine does." A general machine twists and stretches at once — hard. Diagonalization finds a viewpoint where it only stretches.

3. The eigen-arrow — the special direction

The picture: the orange arrow is an ordinary arrow — after $A$ it swings to a new direction (the faded orange). The blue arrow is an eigenvector — after $A$ it lands on its own line, only longer. That "stays on its own line" is the entire meaning of $A\mathbf{x}=\lambda \mathbf{x}$.

• If $\lambda >1$: the eigen-arrow grows.
• If $0<\lambda <1$: it shrinks.
• If $\lambda =1$: it is untouched (fixed).
• If $\lambda <0$: it flips to the opposite side of the origin, then scales.
• If $\lambda =0$: the arrow is crushed to the origin (this direction is the null direction).

Why the topic needs it: these are exactly the "special arrows" the core idea talks about. Diagonalization = collecting enough of them to rebuild space. Full detail lives in Eigenvalues and Eigenvectors.

4. $\lambda I$, $A-\lambda I$, and $\det$ — how we find eigen-arrows

Why we subtract $\lambda I$. Rewrite the eigen-equation moving everything to one side: $A\mathbf{x}=\lambda \mathbf{x}⟹A\mathbf{x}-\lambda \mathbf{x}=0⟹(A-\lambda I)\mathbf{x}=0.$ The middle step needs $\lambda \mathbf{x}=\lambda I\mathbf{x}$ — that's why $I$ appears: you cannot subtract a plain number $\lambda$ from a matrix $A$, but you can subtract the matrix $\lambda I$. It converts the number into a legal machine.

Why the topic needs it: this is Step 1 of the whole procedure. The left side is a polynomial in $\lambda$ — the Characteristic Polynomial. Its roots are the stretch factors.

5. Null space, kernel, and $\ker$ — where eigenvectors live

So the eigenvectors for a given $\lambda$ are precisely the nonzero arrows in $\ker (A-\lambda I)$. That is Step 2: solve $(A-\lambda I)\mathbf{x}=0$.

Why the topic needs it: the parent's "geometric multiplicity" $g(\lambda )=\dim \ker (A-\lambda I)$ is literally "how many independent eigen-arrows this eigenvalue supplies." No kernel, no eigenvectors, no diagonalization.

6. Linear independence — "no arrow is redundant"

Why the topic needs it: the parent's main theorem says $A$ is diagonalizable iff it has $n$ linearly independent eigenvectors. Now you know why that word is load-bearing.

7. Invertible, $P^{-1}$, and $\text{diag}$ — assembling the answer

Why the topic needs it: the goal $A=PD{P}^{-1}$ reads as: "undo the eigen-basis change ($P$), do the pure stretch ($D$), then apply the eigen-basis change ($P$)." Wait — careful with direction. $P$ has eigenvectors as columns, so $P$ builds an arrow from its eigen-coordinates. Reading $A\mathbf{x}=PD{P}^{-1}\mathbf{x}$ right-to-left: $P^{-1}$ writes $\mathbf{x}$ in eigen-coordinates, $D$ stretches each, $P$ rebuilds the ordinary arrow. The power payoff — $A^{100}=P{D}^{100}{P}^{-1}$ — is why anyone cares; see Matrix Powers and Exponentials.

8. Similarity — the relationship $A=PD{P}^{-1}$

Diagonalization is the special case of similarity where the second matrix $D$ happens to be diagonal. When no diagonal $D$ works, the closest we can get is the Jordan Normal Form; when $A$ is symmetric, Spectral Theorem guarantees a beautiful diagonalization.

How the foundations feed the topic

Equipment checklist

Read each cloze, answer out loud, then reveal.