Complex eigenvalues — rotation-scaling interpretation

WHY complex eigenvalues appear

For a $2\times 2$ matrix $A=\begin{pmatrix} a & b\\ c & d\end{pmatrix}$ :

$\det(A-\lambda I)=(a-\lambda)(d-\lambda)-bc = \lambda^2 - (a+d)\lambda + (ad-bc)$

So $\lambda^2 - (\operatorname{tr}A)\lambda + \det A = 0$ , giving

$\lambda = \frac{\operatorname{tr}A \pm \sqrt{(\operatorname{tr}A)^2 - 4\det A}}{2}.$

WHAT decides complex? The discriminant $(\operatorname{tr}A)^2 - 4\det A$ . If it is negative, the roots are a complex conjugate pair $\lambda = \alpha \pm i\beta$ with $\beta \neq 0$ .

HOW a complex eigenvalue encodes rotation-scaling

Take $\lambda = \alpha - i\beta$ (pick this sign by convention) with eigenvector $v = x + i y$ (where $x,y$ are real vectors). Then $Av = \lambda v$ .

Let's expand it. $A(x+iy) = (\alpha - i\beta)(x+iy)$ .

$A x + i A y = (\alpha x + \beta y) + i(\alpha y - \beta x)$

Matching real and imaginary parts (both $x,y$ real, $A$ real, so $Ax,Ay$ real):

$Ax = \alpha x + \beta y, \qquad Ay = -\beta x + \alpha y.$

Now read those two equations as columns. Build the real matrix $P = \begin{pmatrix} x & y\end{pmatrix}$ (columns $x$ and $y$ ). Then

$A P = P \begin{pmatrix} \alpha & -\beta \\ \beta & \alpha\end{pmatrix}.$

Check: $AP$ has columns $Ax$ and $Ay$ . $P\begin{pmatrix}\alpha & -\beta\\ \beta & \alpha\end{pmatrix}$ has columns $\alpha x + \beta y$ and $-\beta x + \alpha y$ . ✓ matches above.

So $A = PCP^{-1}$ with

WHY this factoring works: Every matrix of the form $\begin{pmatrix}\alpha & -\beta\\ \beta & \alpha\end{pmatrix}$ factors as scale $\times$ pure rotation, because $\alpha = r\cos\theta,\ \beta = r\sin\theta$ . So in the coordinate frame given by $P$ (basis vectors $x,y$ ), $A$ acts as "rotate by $\theta$ , then magnify by $r$ ."

Worked Example 1 — pure rotation

$A=\begin{pmatrix} 0 & -1 \\ 1 & 0\end{pmatrix}$ (the $90°$ rotation matrix).

$\operatorname{tr}A=0,\ \det A = 1$ . Why? trace $=a+d=0$ , det $=ad-bc=0-(-1)=1$ .
$\lambda^2 - 0\cdot\lambda + 1=0 \Rightarrow \lambda = \pm i$ . Why complex? discriminant $0-4=-4<0$ .
$r=|\lambda|=1$ , $\theta=\arg(i)...$ taking $\lambda=-i$ gives $\alpha=0,\beta=1$ , $\theta=\operatorname{atan2}(1,0)=90°$ .
Interpretation: scale $1$ , rotate $90°$ . Exactly the rotation we started with. ✓

Worked Example 2 — spiral

$A=\begin{pmatrix} 1 & -1 \\ 1 & 1\end{pmatrix}$ .

$\operatorname{tr}=2,\ \det = 1\cdot1-(-1)(1)=2$ . Why? standard formulas.
$\lambda = \frac{2\pm\sqrt{4-8}}{2} = 1\pm i$ . Why this? plug into quadratic formula; $\sqrt{-4}=2i$ .
Take $\lambda = 1-i$ : $\alpha=1,\beta=1$ , so $r=\sqrt{2}\approx1.41$ , $\theta=\operatorname{atan2}(1,1)=45°$ .
Interpretation: each application rotates $45°$ and magnifies by $\sqrt2$ → an outward spiral. Why outward? $r>1$ .

Worked Example 3 — finding $P$

Same $A=\begin{pmatrix} 1 & -1 \\ 1 & 1\end{pmatrix}$ , $\lambda=1-i$ . Solve $(A-\lambda I)v=0$ : $A-\lambda I=\begin{pmatrix} i & -1 \\ 1 & i\end{pmatrix}.$ Row 1: $i v_1 - v_2 = 0 \Rightarrow v_2 = i v_1$ . Take $v_1=1$ : $v=\begin{pmatrix}1\\ i\end{pmatrix}$ .

Why pick the eigenvector for $\lambda=\alpha-i\beta$ ? Convention so that $C$ comes out with $+\beta$ in the lower-left; the other conjugate just gives the mirror.
Split: $v = \underbrace{\begin{pmatrix}1\\0\end{pmatrix}}_{x} + i\underbrace{\begin{pmatrix}0\\1\end{pmatrix}}_{y}$ , so $P=\begin{pmatrix}1&0\\0&1\end{pmatrix}=I$ .
Then $C = P^{-1}AP = A = \begin{pmatrix}1&-1\\1&1\end{pmatrix}=\sqrt2\begin{pmatrix}\cos45° & -\sin45°\\ \sin45° & \cos45°\end{pmatrix}$ . ✓ Already in rotation-scaling form (makes sense — $A$ was symmetric in that special pattern).

Common mistake Steel-manned errors

Mistake 1: "Complex eigenvalues mean no real eigenvectors, so $A$ isn't diagonalizable / is useless over $\mathbb{R}$ ." Why it feels right: over $\mathbb{R}$ there really is no eigen-line. Fix: over $\mathbb{C}$ it IS diagonalizable; over $\mathbb{R}$ you get the next best thing — block form $\begin{pmatrix}\alpha&-\beta\\\beta&\alpha\end{pmatrix}$ , the rotation-scaling. Fully usable.

Mistake 2: Mixing up which is rotation vs scaling. Students write $r=\alpha$ , $\theta=\beta$ . Why it feels right: $\alpha,\beta$ are right there. Fix: $r=\sqrt{\alpha^2+\beta^2}=|\lambda|$ (modulus = scale), $\theta=\arg\lambda$ (angle = rotation). $\alpha,\beta$ are Cartesian coords of $\lambda$ , not polar.

Mistake 3: Sign of $\theta$ / spiral direction. Choosing eigenvalue $\alpha+i\beta$ vs $\alpha-i\beta$ flips the sign of $\beta$ in $C$ , reversing rotation sense. Fix: it's only a relabeling of the basis ( $x\leftrightarrow y$ orientation); pick one convention ( $\lambda=\alpha-i\beta$ ) and stay consistent.

Mistake 4: Thinking $r>1$ always means "grows." True for iterating $A^n$ (discrete). For the ODE $\dot v = Av$ , growth is governed by $\operatorname{Re}(\lambda)=\alpha$ , not $r$ . Fix: discrete map → use $|\lambda|$ ; continuous flow → use $\operatorname{Re}\lambda$ .

Recall Feynman: explain to a 12-year-old

Imagine spinning a record player. Every dot on the record moves in a circle — none of them stays pointing in the same direction. A "stretch matrix" would push dots straight out along fixed lines, but a "spinner" has no fixed line at all. When math tries to find a fixed direction for a spinner, it can't find a real one, so it answers with "imaginary" numbers. Those imaginary numbers secretly carry two pieces of news: how fast it spins (the angle) and how much it grows or shrinks each turn (the size). So a complex eigenvalue is just a spin-and-zoom instruction in disguise.

Active Recall

When does a real 2×2 matrix have complex eigenvalues?

When the discriminant

(\operatorname{tr}A)^2-4\det A<0

, i.e. the characteristic quadratic has no real roots.

Why do complex eigenvalues come in conjugate pairs?

The characteristic polynomial has real coefficients, and complex roots of a real polynomial always occur as conjugate pairs

\alpha\pm i\beta

For

\lambda=\alpha-i\beta

, what is the scale factor of the rotation-scaling?

r=|\lambda|=\sqrt{\alpha^2+\beta^2}

For

\lambda=\alpha-i\beta

, what is the rotation angle?

\theta=\arg\lambda=\operatorname{atan2}(\beta,\alpha)

What real normal form is

A

similar to?

A=PCP^{-1}

with

C=\begin{pmatrix}\alpha&-\beta\\\beta&\alpha\end{pmatrix}=r\begin{pmatrix}\cos\theta&-\sin\theta\\\sin\theta&\cos\theta\end{pmatrix}

How do you build

P

from a complex eigenvector

v=x+iy

P=\begin{pmatrix}x & y\end{pmatrix}

— real part

x

as first column, imaginary part

y

as second column.

Derive the two real equations from

Av=\lambda v

with

\lambda=\alpha-i\beta,\ v=x+iy

Ax=\alpha x+\beta y

and

Ay=-\beta x+\alpha y

(match real & imaginary parts).

Discrete iteration

A^n

: when does the trajectory spiral outward?

When

r=|\lambda|>1

. Inward if

r<1

, closed ellipse if

r=1

Continuous system

\dot v=Av

: what controls growth/decay?

\operatorname{Re}(\lambda)=\alpha

(not the modulus):

\alpha>0

grows,

\alpha<0

decays,

\alpha=0

neutral oscillation.

Eigenvalues of the 90° rotation matrix

\begin{pmatrix}0&-1\\1&0\end{pmatrix}

\pm i

(so

r=1,\ \theta=90°

, a pure rotation).

Trace and determinant in terms of eigenvalues for 2×2?

\operatorname{tr}A=\lambda_1+\lambda_2=2\alpha

and

\det A=\lambda_1\lambda_2=\alpha^2+\beta^2=r^2

Connections

Eigenvalues and Eigenvectors — the real-eigenvalue baseline this generalizes.
Characteristic Polynomial — source of the discriminant test.
Rotation Matrices — the $r=1$ special case.
Diagonalization — over $\mathbb{C}$ these are diagonalizable; over $\mathbb{R}$ you get block form.
Complex Numbers - Polar Form — modulus/argument = scale/angle.
Linear Dynamical Systems / Phase Portraits — spirals vs centers from $\operatorname{Re}\lambda$ .
Trace and Determinant — quick $\alpha=\tfrac{\operatorname{tr}}{2}$ , $r^2=\det$ .

Concept Map

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho, jab ek real $2\times2$ matrix ke eigenvalues complex aate hain, to ghabrao mat — ye matrix actually plane ko ghuma (rotate) raha hai, aur saath mein thoda bada/chhota (scale) kar raha hai. Soch ke dekho: agar koi matrix sirf ghuma raha hai (jaise record player), to koi bhi arrow apni direction mein same nahi rehta — sab ghoom jaate hain. Isliye koi real eigen-direction milti hi nahi, aur algebra apna jawab "imaginary" numbers ke through deti hai.

Complex eigenvalue $\lambda=\alpha\pm i\beta$ ko ek point ki tarah plane par socho. Iska modulus $r=|\lambda|=\sqrt{\alpha^2+\beta^2}$ batata hai ki har step mein kitna scale ho raha hai, aur argument (angle) $\theta=\arg\lambda$ batata hai ki kitne degree rotate ho raha hai. Yaad rakhne ka mantra: MAS — Modulus = Amount of Scaling, Argument = Spin.

Algebra ka trick simple hai: $Av=\lambda v$ likho jahan $v=x+iy$ (real part $x$ , imaginary part $y$ ). Real aur imaginary parts ko alag-alag match karo, to do real equations milti hain, aur matrix $A$ ban jaata hai $PCP^{-1}$ , jahan $C=\begin{pmatrix}\alpha&-\beta\\\beta&\alpha\end{pmatrix}=r\,(\text{rotation matrix})$ . Matlab $P$ wale naye coordinate frame mein $A$ ka kaam bas "ghumao aur zoom karo" hai.

Ek common confusion: agar $r>1$ to discrete iteration $A^n$ mein spiral bahar jaati hai (outward), $r<1$ to andar (inward). Lekin agar continuous system $\dot v=Av$ ho, to growth $\operatorname{Re}(\lambda)=\alpha$ se decide hoti hai, $r$ se nahi. Ye difference exam mein bahut kaam aata hai!

Complex eigenvalues — rotation-scaling interpretation

WHY complex eigenvalues appear

HOW a complex eigenvalue encodes rotation-scaling

Worked Example 1 — pure rotation

Worked Example 2 — spiral

Worked Example 3 — finding $P$

Active Recall

Connections

Concept Map

Hinglish (regional understanding)

Go deeper — visual, from zero

Test yourself — Linear Algebra (Full)

Connections