Foundations — Symmetric matrices — spectral theorem (real eigenvalues, orthogonal eigenvectors)

This page assumes nothing. Before you can read a single line of the Spectral Theorem, you need a small toolkit of symbols. We build each one from a picture, in the order they lean on each other.

1. A vector — an arrow with an address

Picture an arrow starting at the origin (the point $(0,0)$) and ending at the address $(x_1,x_2)$. The numbers are the arrow.

Why the topic needs it: the whole theorem is about how a matrix moves arrows around, and about which special arrows keep their direction.

2. A matrix — a machine that eats arrows

$A=\left(\begin{array}{cc}a& b\\ c& d\end{array}\right),A\mathbf{x}=\left(\begin{array}{c}a{x}_1+b{x}_2\\ c{x}_1+d{x}_2\end{array}\right).$

Read the rule as: each row of $A$ is a recipe. Row 1 says "$a$ of the first coordinate plus $b$ of the second gives the new first coordinate." In $n$ dimensions the same recipe runs over $n$ terms per row.

Why the topic needs it: $A$ is the star of the show. The theorem describes exactly what symmetric $A$ does to space.

3. Scaling a vector — $\lambda \mathbf{x}$ and $c\mathbf{x}$

Picture: $c\mathbf{x}$ is the same arrow, its length multiplied by $c$. If $c=2$ it doubles; if $c=-1$ it flips to point the opposite way; if $c=0$ it collapses to the origin.

Why the topic needs it: the eigen-equation $A\mathbf{x}=\lambda \mathbf{x}$ is a scaling statement, and $A-\lambda I$ subtracts a scaled identity — we cannot read either without knowing what "scale a vector/matrix" means.

4. Transpose $A^{\top }$ — flip across the diagonal

${\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)}^{\top }=\left(\begin{array}{cc}a& c\\ b& d\end{array}\right).$

The $b$ and $c$ trade places; $a$ and $d$ sit on the mirror line and stay put. (In $n$ dimensions the same swap happens for every off-diagonal pair.)

Why the topic needs it: the entire definition of "symmetric" is phrased with the transpose.

5. Symmetric matrix — $A=A^{\top }$

$\left(\begin{array}{cc}2& 1\\ 1& 2\end{array}\right)={\left(\begin{array}{cc}2& 1\\ 1& 2\end{array}\right)}^{\top }✓$

Why the topic needs it: this single equation $A=A^{\top }$ is the hypothesis that unlocks real eigenvalues and perpendicular axes.

6. The identity $I$ and the zero vector $0$

The zero vector $0=\left(\begin{array}{c}0\\ 0\end{array}\right)$ is the arrow of no length, sitting at the origin. When we search for eigenvectors we always demand $\mathbf{x}\ne 0$, because the zero arrow trivially satisfies everything and tells us nothing.

Why the topic needs it: the phrase $A-\lambda I$ (below) subtracts a scaled identity $\lambda I$ (scalar multiplication from §3), and the eigenvector hunt excludes $0$.

7. Eigenvalue $\lambda$ and eigenvector — the arrows that keep their direction

• If $\lambda =2$, the arrow doubles in length, same direction.
• If $\lambda =1$, it doesn't move at all.
• If $\lambda =-1$, it flips to point backwards (same line, opposite way).
• If $\lambda =0$, the arrow is crushed onto the origin.

For a full treatment see Eigenvalues and Eigenvectors. The expression $A-\lambda I$ finds these: solving $\det (A-\lambda I)=0$ gives the $\lambda$'s.

Why the topic needs it: the theorem is a statement about the eigenvalues and eigenvectors of a symmetric matrix.

8. Dot product ${\mathbf{u}}^{\top }\mathbf{v}$ — the perpendicularity detector

The magic fact: ${\mathbf{u}}^{\top }\mathbf{v}=0$ exactly when the two arrows meet at $90^{\circ }$ — they are orthogonal (a fancy word for "perpendicular").

Why the topic needs it: "eigenvectors of distinct eigenvalues are orthogonal" is a statement about the dot product being zero. Without this tool the theorem's second promise is unreadable.

9. Matrix inverse $A^{-1}$ — the "undo" machine

Picture: if $A$ stretches and rotates the rubber sheet, $A^{-1}$ un-stretches and un-rotates it back to the original grid. (Not every matrix has an inverse — one that crushes space flat cannot be undone — but the matrices we build below always do.)

Why the topic needs it: the theorem writes $A=Q\Lambda Q^{\top }$, and the reason it can use $Q^{\top }$ where you'd expect $Q^{-1}$ is the surprise identity $Q^{-1}=Q^{\top }$. You can't appreciate that gift without knowing what $Q^{-1}$ means.

10. Orthonormal, and the matrix $Q$

"Length 1" means we normalize: divide a vector by its length $\sqrt{u_1^2+u_2^2}$ (in ${\mathbb{R}}^n$, the square root of the dot product ${\mathbf{u}}^{\top }\mathbf{u}$). For example $(1,1)$ has length $\sqrt{2}$, so its normalized version is $\frac{1}{\sqrt{2}}(1,1)$.

Why the topic needs it: $Q$ holds the perpendicular skeleton directions as unit columns; it is the coordinate frame in which $A$ looks diagonal.

11. Diagonal matrix $\Lambda$ — pure stretching

This is the simplest possible machine: scale the first axis by ${\lambda }_1$, the second by ${\lambda }_2$, done. The diagonal entries are precisely the eigenvalues.

Why the topic needs it: the theorem's punchline $A=Q\Lambda Q^{\top }$ says "rotate into the good frame ($Q^{\top }$), do pure stretching ($\Lambda$), rotate back ($Q$)." Understanding Diagonalization fully unpacks this.

12. Determinant $\det$ — the eigenvalue-finder

For a $2\times 2$ matrix $\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)$, the determinant is $ad-bc$.

Why the topic needs it: every worked example finds eigenvalues by solving $\det (A-\lambda I)=0$.

How these feed the theorem

Reading the map top-to-bottom retraces this whole page: we start from real numbers and arrows, learn to scale and flip them, and end at the theorem's decomposition. Each box names a concept in the plain words we built above (no new notation).

The theorem sits at the bottom: symmetry plus perpendicular eigen-directions plus pure stretching combine into $A=Q\Lambda Q^{\top }$.

Equipment checklist

Test yourself — cover the right side and answer aloud.