Linear AlgebraEigenvalues and Eigenvectors6 of 7
Eigenvalues and Eigenvectors · 6

Complex Eigenvalues and Rotation

To see what repeated multiplication by a matrix does to a vector, we look for its eigenvalues and eigenvectors, though sometimes a matrix of real numbers gives us complex eigenvalues.

1

Why Complex Eigenvalues Mean Rotation

When a matrix has real eigenvalues, its eigenvectors stay on their own lines, and diagonalizing it as A=PDP−1A = PDP^{-1} turns the matrix into simple scaling along those lines. We want to find the eigenvalues of

A=[1−21−1]\displaystyle A = \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}

to see where the vector [10]\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} lands after multiplying by AA a hundred times.

Why Complex Eigenvalues Mean Rotation
A=[1−21−1]\displaystyle A = \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}
A100[10]=  ?\displaystyle A^{100} \orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} = \;?
Text solution
Subtract λ\lambda from the diagonal and set the determinant to zeroWe set up the characteristic equation det⁡(A−λI)=0\det(A - \lambda I) = 0 to look for vectors that stay on their own lines:
(1−λ)(−1−λ)−(−2)(1)\displaystyle (1 - \lambda)(-1 - \lambda) - (-2)(1) =0\displaystyle = 0
Expand the determinant and solve for λ\lambdaMultiplying out the terms simplifies the equation to λ2+1=0\lambda^2 + 1 = 0:
−1+λ2+2=0λ2+1=0λ2=−1λ=±i\displaystyle \begin{aligned} -1 + \lambda^2 + 2 &= 0 \\ \lambda^2 + 1 &= 0 \\ \lambda^2 &= -1 \\ \lambda &= \pm i \end{aligned}

So the eigenvalues are λ=±i\lambda = \pm i, the exact same eigenvalues as the pure 90∘90^\circ counterclockwise rotation matrix [0−110]\begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix}, under which no real vector stays on its own line.

2

Rotation-Scaling Matrices

A pure ±90∘\pm 90^\circ rotation gives λ=±i\lambda = \pm i, where the eigenvalue has no real part. Now we want to see what matrix comes from a full complex eigenvalue λ=a+bi\lambda = a + bi, and how to find the scale factor and rotation angle of

[1−111]\displaystyle \begin{bmatrix} 1 & -1 \\ 1 & 1 \end{bmatrix}

A pure counterclockwise rotation by an angle θ\turnangle{\theta} is given by the standard rotation matrix

[cos⁡θ−sin⁡θsin⁡θcos⁡θ]\displaystyle \begin{bmatrix} \cos\turnangle{\theta} & -\sin\turnangle{\theta} \\ \sin\turnangle{\theta} & \cos\turnangle{\theta} \end{bmatrix}
Rotation-Scaling Matrices
(3+4i) v=  ?(3 + 4i)\,\htmlClass{lc-green}{v} = \;?
v\htmlClass{lc-green}{v}
iv\htmlClass{lc-pink}{iv}
×i\times i
aa
bb
r=a2+b2\htmlClass{lc-orange}{r} = \sqrt{a^2 + b^2}
θ\htmlClass{lc-violet}{\theta}

A matrix with eigenvalues λ=a±bi\lambda = a \pm bi that stretches by aa along each axis and by bb rotated 90∘90^\circ counterclockwise factors into a scaling and a pure rotation:

[a−bba]⏟rotation-scaling matrix C=[ab]r⏟scale factora2+b2[cos⁡θ−sin⁡θsin⁡θcos⁡θ]⏟counterclockwise rotationcos⁡θ=a/r,sin⁡θ=b/r\displaystyle \underbrace{\begin{bmatrix} a & -b \\ b & a \end{bmatrix}}_{\text{rotation-scaling matrix } C} = \underbrace{\vphantom{\begin{bmatrix} a \\ b \end{bmatrix}} \scalefactor{r}}_{\substack{\text{scale factor} \\ \sqrt{a^2 + b^2}}} \underbrace{\begin{bmatrix} \cos\turnangle{\theta} & -\sin\turnangle{\theta} \\ \sin\turnangle{\theta} & \cos\turnangle{\theta} \end{bmatrix}}_{\substack{\text{counterclockwise rotation} \\ \cos\turnangle{\theta} = a/\scalefactor{r},\; \sin\turnangle{\theta} = b/\scalefactor{r}}}
Text solution
Match the matrix with [a−bba]\begin{bmatrix} a & -b \\ b & a \end{bmatrix} to read off the legs aa and bbThe diagonal entries give the horizontal leg aa and the bottom-left entry gives the vertical leg bb:
a\displaystyle a =1,b\displaystyle = 1, \qquad b =1\displaystyle = 1
Use the right triangle with legs 11 and 11 to find the hypotenuse r\scalefactor{r} and angle θ\turnangle{\theta}The Pythagorean theorem gives the scale factor r\scalefactor{r}, and the ratios ar\tfrac{a}{\scalefactor{r}} and br\tfrac{b}{\scalefactor{r}} give the angle θ\turnangle{\theta}:
r=12+12=2cos⁡θ=12,sin⁡θ=12  ⟹  θ=π4 (or 45∘)\displaystyle \begin{aligned} \scalefactor{r} &= \sqrt{1^2 + 1^2} = \sqrt{2} \\ \cos\turnangle{\theta} &= \frac{1}{\sqrt{2}}, \\ \sin\turnangle{\theta} &= \frac{1}{\sqrt{2}} \\ &\implies \turnangle{\theta} = \frac{\pi}{4}\text{ (or } 45^\circ\text{)} \end{aligned}
Factor the scale factor r\scalefactor{r} out of the matrixPulling out 2\sqrt{2} leaves the pure 45∘45^\circ rotation matrix:
[1−111]=2[cos⁡(π4)−sin⁡(π4)sin⁡(π4)cos⁡(π4)]\displaystyle \begin{bmatrix} 1 & -1 \\ 1 & 1 \end{bmatrix} = \scalefactor{\sqrt{2}}\begin{bmatrix} \cos(\turnangle{\tfrac{\pi}{4}}) & -\sin(\turnangle{\tfrac{\pi}{4}}) \\ \sin(\turnangle{\tfrac{\pi}{4}}) & \cos(\turnangle{\tfrac{\pi}{4}}) \end{bmatrix}

So each multiplication by [1−111]\begin{bmatrix} 1 & -1 \\ 1 & 1 \end{bmatrix} rotates a vector 45∘45^\circ counterclockwise and stretches its length by 2\sqrt{2}.

When you multiply by C=[a−bba]C = \begin{bmatrix} a & -b \\ b & a \end{bmatrix} over and over, the path traces a circle when r=1\scalefactor{r} = 1, spirals outward when r>1\scalefactor{r} > 1, spirals inward when r<1\scalefactor{r} < 1, and stays on a straight line when b=0b = 0.

Where the eigenvalue sits sets both the rotation angle and the scale factor of the matrix, turning a vector’s repeated path into a circle, a spiral, or a straight line.
3

Rotation on a Tilted Ellipse

A matrix of the form [a−bba]\begin{bmatrix} a & -b \\ b & a \end{bmatrix} has equal diagonal entries and rotates and scales vectors on the standard grid, so its paths are circles or spirals around the origin. Our starting matrix A=[1−21−1]A = \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix} has different diagonal entries (11 and −1-1), yet its eigenvalues are still λ=±i\lambda = \pm i (r=1,θ=90∘\scalefactor{r} = 1, \turnangle{\theta} = 90^\circ). Let's find what AA actually does to [10]\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} over repeated steps and compute

A100[10]forA=[1−21−1]\displaystyle A^{100}\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} \quad \text{for} \quad A = \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}
Rotation on a Tilted Ellipse
[a−bba]\begin{bmatrix} a & -b \\ b & a \end{bmatrix}
A=[1−21−1]A = \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}
[10]\htmlClass{lc-blue}{\begin{bmatrix} 1 \\ 0 \end{bmatrix}}
Text solution
Multiply [10]\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} by AA four times until it returns to the startMultiplying the vector by AA one step at a time shows a four-step loop:
A[10]=[1−21−1][10]=[11]A2[10]=[1−21−1][11]=[1−21−1]=[−10]A3[10]=[1−21−1][−10]=[−1−1]A4[10]=[1−21−1][−1−1]=[−1+2−1+1]=[10]\displaystyle \begin{aligned} A\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} &= \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} \\ &= \orbit{\begin{bmatrix} 1 \\ 1 \end{bmatrix}} \\ A^2\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} &= \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}\orbit{\begin{bmatrix} 1 \\ 1 \end{bmatrix}} \\ &= \begin{bmatrix} 1 - 2 \\ 1 - 1 \end{bmatrix} = \orbit{\begin{bmatrix} -1 \\ 0 \end{bmatrix}} \\ A^3\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} &= \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}\orbit{\begin{bmatrix} -1 \\ 0 \end{bmatrix}} \\ &= \orbit{\begin{bmatrix} -1 \\ -1 \end{bmatrix}} \\ A^4\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} &= \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}\orbit{\begin{bmatrix} -1 \\ -1 \end{bmatrix}} \\ &= \begin{bmatrix} -1 + 2 \\ -1 + 1 \end{bmatrix} = \orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} \end{aligned}
Use the four-step loop to evaluate the 100th powerBecause 100100 is 2525 full cycles of 44, the vector goes around the loop 2525 times:
A100[10]=(A4)25[10]=[10]\displaystyle A^{100}\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} = (A^4)^{25}\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} = \orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}}

So after 100 multiplications, the vector lands right back at [10]\orbit{\begin{bmatrix} 1 \\ 0 \end{bmatrix}}, having gone around a tilted ellipse 25 times.

[0−110]\begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix}
[10]\htmlClass{lc-blue}{\begin{bmatrix} 1 \\ 0 \end{bmatrix}}
90∘90^\circ
A=[1−21−1]A = \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}
[10]\htmlClass{lc-blue}{\begin{bmatrix} 1 \\ 0 \end{bmatrix}}
[11]\htmlClass{lc-blue}{\begin{bmatrix} 1 \\ 1 \end{bmatrix}}
[−10]\htmlClass{lc-blue}{\begin{bmatrix} -1 \\ 0 \end{bmatrix}}
[−1−1]\htmlClass{lc-blue}{\begin{bmatrix} -1 \\ -1 \end{bmatrix}}
A\htmlClass{lc-violet}{A}
4

Complex Eigenvectors and the Tilted Basis

Instead of multiplying vectors out step by step to see the tilted ellipse, we want a direct algebraic way to find the basis vectors of the tilted grid where

A=[1−21−1]\displaystyle A = \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}

rotates space.

Complex Eigenvectors and the Tilted Basis
A=[1−21−1],    λ=i ⁣:    v=  ?A = \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix},\;\; \lambda = i\!:\;\; v = \;?
Text solution
Substitute λ=i\lambda = i into Av=λvAv = \lambda v and read off the simpler rowOnce an eigenvalue of a 2×22 \times 2 matrix is plugged in, one row is always a multiple of the other, so the bottom row of [1−21−1][xy]=i[xy]\begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix}\begin{bmatrix} x \\ y \end{bmatrix} = i\begin{bmatrix} x \\ y \end{bmatrix} gives the whole relationship:
1x−1y=iyx=y+iyx=(1+i)y\displaystyle \begin{aligned} 1x - 1y &= iy \\ x &= y + iy \\ x &= (1 + i)y \end{aligned}
Choose a nonzero value for the free variable yy and split vv into real and imaginary partsSetting y=1y = 1 gives x=1+ix = 1 + i, which splits into two real vectors u\blocktwo{u} and w\blockone{w}:
v=[1+i1]=[11]⏟real part u+i[10]⏟imaginary part w\displaystyle \begin{aligned} v &= \begin{bmatrix} 1 + i \\ 1 \end{bmatrix} \\ &= \underbrace{\blocktwo{\begin{bmatrix} 1 \\ 1 \end{bmatrix}}}_{\text{real part } \blocktwo{u}} + i\underbrace{\blockone{\begin{bmatrix} 1 \\ 0 \end{bmatrix}}}_{\text{imaginary part } \blockone{w}} \end{aligned}
Flip the sign of ii to get the conjugate eigenvector for λˉ=−i\bar{\lambda} = -iBecause AA is real, taking the complex conjugate of Av=λvAv = \lambda v gives Avˉ=λˉvˉA\bar{v} = \bar{\lambda}\bar{v} without any extra algebra:
vˉ=u−iw⏟conjugate eigenvector=[11]−i[10]=[1−i1]\displaystyle \begin{aligned} \bar{v} &= \underbrace{\blocktwo{u} - i\blockone{w}}_{\text{conjugate eigenvector}} \\ &= \blocktwo{\begin{bmatrix} 1 \\ 1 \end{bmatrix}} - i\blockone{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} = \begin{bmatrix} 1 - i \\ 1 \end{bmatrix} \end{aligned}

Notice that both u=[11]\blocktwo{u} = \blocktwo{\begin{bmatrix} 1 \\ 1 \end{bmatrix}} and w=[10]\blockone{w} = \blockone{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} are completely real vectors (w\blockone{w} is the vector multiplied by ii, without the ii inside it). Expanding A(u+iw)=i(u+iw)=−w+iuA(\blocktwo{u} + i\blockone{w}) = i(\blocktwo{u} + i\blockone{w}) = -\blockone{w} + i\blocktwo{u} and matching the real and imaginary parts on each side shows what AA does to them:

Au\displaystyle A\blocktwo{u} =−w,Aw\displaystyle = -\blockone{w}, \qquad A\blockone{w} =u\displaystyle = \blocktwo{u}
w\htmlClass{lc-green}{w}
u\htmlClass{lc-pink}{u}
45∘45^\circ
A\htmlClass{lc-violet}{A}
−w-\htmlClass{lc-green}{w}
A\htmlClass{lc-violet}{A}

So u\blocktwo{u} and w\blockone{w} are the basis vectors of the skewed grid for the tilted ellipse, sitting one quarter-turn apart under AA.

Picking any other nonzero complex number for the free variable yy in x=(1+i)yx = (1 + i)y just gives another valid pair of basis vectors u\blocktwo{u} and w\blockone{w} on the same family of tilted ellipses, still one quarter-turn apart (Aw=uA\blockone{w} = \blocktwo{u} and Au=−wA\blocktwo{u} = -\blockone{w}). Because the ellipse is tilted, u\blocktwo{u} and w\blockone{w} aren't perpendicular unless they happen to line up with its main axes.

Every nonzero choice of the complex free variable y gives you another valid pair of basis vectors u and w on the same family of tilted ellipses, always one quarter-turn apart when multiplied by A.
5

Factoring A=PCP−1A = PCP^{-1}

Now we can put the rotation-scaling matrix CC and the tilted basis vectors u\blocktwo{u} and w\blockone{w} together to factor any real 2×22 \times 2 matrix with complex eigenvalues as A=PCP−1A = PCP^{-1}, the complex version of A=PDP−1A = PDP^{-1}. Let's factor

A=[3−211]\displaystyle A = \begin{bmatrix} 3 & -2 \\ 1 & 1 \end{bmatrix}
Factoring A = PCP⁻¹
[3−211]=P C P−1\displaystyle \begin{bmatrix} 3 & -2 \\ 1 & 1 \end{bmatrix} = P\,C\,P^{-1}
P\displaystyle P =  ?C\displaystyle = \;? \qquad C =  ?\displaystyle = \;?

For any eigenvalue λ=a+bi\lambda = a + bi with eigenvector v=u+iwv = \blocktwo{u} + i\blockone{w}, expanding A(u+iw)=(a+bi)(u+iw)=(au−bw)+i(bu+aw)A(\blocktwo{u} + i\blockone{w}) = (a + bi)(\blocktwo{u} + i\blockone{w}) \\ = (a\blocktwo{u} - b\blockone{w}) + i(b\blocktwo{u} + a\blockone{w}) and matching real and imaginary parts tells us where AA sends both basis vectors:

Au\displaystyle A\blocktwo{u} =au−bw,Aw\displaystyle = a\blocktwo{u} - b\blockone{w}, \qquad A\blockone{w} =bu+aw\displaystyle = b\blocktwo{u} + a\blockone{w}

Reading the weights on u\blocktwo{u} and w\blockone{w} directly from those two equations gives the two valid ways to pair PP and CC:

P=[∣∣uw∣∣],C=[ab−ba]⏟u in column 1 puts −b in column 1orP=[∣∣wu∣∣],C=[a−bba]⏟u in column 2 puts −b in column 2\displaystyle \begin{gathered} \underbrace{P = \begin{bmatrix} | & | \\ \blocktwo{u} & \blockone{w} \\ | & | \end{bmatrix}, \quad C = \begin{bmatrix} a & b \\ -b & a \end{bmatrix}}_{\blocktwo{u} \text{ in column 1 puts } -b \text{ in column 1}} \\ \text{or} \\ \underbrace{P = \begin{bmatrix} | & | \\ \blockone{w} & \blocktwo{u} \\ | & | \end{bmatrix}, \quad C = \begin{bmatrix} a & -b \\ b & a \end{bmatrix}}_{\blocktwo{u} \text{ in column 2 puts } -b \text{ in column 2}} \end{gathered}
Text solution
Set det⁡(A−λI)=0\det(A - \lambda I) = 0 and solve for the complex eigenvaluesSubtracting λ\lambda from the diagonal gives a quadratic equation that we can solve by completing the square:
(3−λ)(1−λ)−(−2)(1)=0λ2−4λ+5=0(λ−2)2+1=0λ=2±i\displaystyle \begin{aligned} (3 - \lambda)(1 - \lambda) - (-2)(1) &= 0 \\ \lambda^2 - 4\lambda + 5 &= 0 \\ (\lambda - 2)^2 + 1 &= 0 \\ \lambda &= 2 \pm i \end{aligned}
Plug λ=2+i\lambda = 2 + i into Av=λvAv = \lambda v and split the eigenvector into u\blocktwo{u} and w\blockone{w}Using the bottom row of [3−211][xy]=(2+i)[xy]\begin{bmatrix} 3 & -2 \\ 1 & 1 \end{bmatrix}\begin{bmatrix} x \\ y \end{bmatrix} = (2 + i)\begin{bmatrix} x \\ y \end{bmatrix} and picking y=1y = 1 gives u\blocktwo{u} and w\blockone{w}:
1x+1y=(2+i)yx=(1+i)yv=[1+i1]=[11]⏟u+i[10]⏟w\displaystyle \begin{aligned} 1x + 1y &= (2 + i)y \\ x &= (1 + i)y \\ v &= \begin{bmatrix} 1 + i \\ 1 \end{bmatrix} = \underbrace{\blocktwo{\begin{bmatrix} 1 \\ 1 \end{bmatrix}}}_{\blocktwo{u}} + i\underbrace{\blockone{\begin{bmatrix} 1 \\ 0 \end{bmatrix}}}_{\blockone{w}} \end{aligned}
Build PP, CC, and P−1P^{-1} with matching column orderPutting u\blocktwo{u} in the first column and w\blockone{w} in the second column of PP puts −b=−1-b = -1 in the first column (bottom-left) of CC for a=2,b=1a = 2, b = 1:
P=[1110],C=[21−12],P−1=1−1[0−1−11]=[011−1]\displaystyle \begin{aligned} P &= \begin{bmatrix} \blocktwo{1} & \blockone{1} \\ \blocktwo{1} & \blockone{0} \end{bmatrix}, \\ C &= \begin{bmatrix} 2 & 1 \\ -1 & 2 \end{bmatrix}, \\ P^{-1} &= \frac{1}{-1}\begin{bmatrix} 0 & -1 \\ -1 & 1 \end{bmatrix} = \begin{bmatrix} 0 & 1 \\ 1 & -1 \end{bmatrix} \end{aligned}
A=[1110][21−12][011−1]\displaystyle A = \begin{bmatrix} \blocktwo{1} & \blockone{1} \\ \blocktwo{1} & \blockone{0} \end{bmatrix}\begin{bmatrix} 2 & 1 \\ -1 & 2 \end{bmatrix}\begin{bmatrix} 0 & 1 \\ 1 & -1 \end{bmatrix}

(Swapping the columns to P=[wu]=[1101]P = \begin{bmatrix} \blockone{w} & \blocktwo{u} \end{bmatrix} = \begin{bmatrix} \blockone{1} & \blocktwo{1} \\ \blockone{0} & \blocktwo{1} \end{bmatrix} puts −1-1 in the top-right instead, giving C=[2−112]C = \begin{bmatrix} 2 & -1 \\ 1 & 2 \end{bmatrix} and P−1=[1−101]P^{-1} = \begin{bmatrix} 1 & -1 \\ 0 & 1 \end{bmatrix}.)

In the tilted basis PP, each multiplication by AA scales vectors by r=22+12=5\scalefactor{r} = \sqrt{2^2 + 1^2} = \sqrt{5} and rotates them by θ\turnangle{\theta} (where cos⁡θ=25\cos\turnangle{\theta} = \tfrac{2}{\sqrt{5}} and sin⁡θ=15\sin\turnangle{\theta} = \tfrac{1}{\sqrt{5}}), spiraling them outward.

Changing the basis vectors u\blocktwo{u} and w\blockone{w} in PP tilts and stretches the coordinate grid that the path follows, while CC keeps the same scale factor r\scalefactor{r} and rotation angle θ\turnangle{\theta} at each step.

Changing the basis vectors tilts and stretches the grid the spiral follows, without changing its scale factor or rotation angle at each step.

Key concepts

When you find the eigenvalues of a real 2×22 \times 2 matrix from det⁡(A−λI)=0\det(A - \lambda I) = 0, whether the eigenvalues are real or complex tells you whether the matrix scales along fixed lines or rotates, and whether to factor it as A=PDP−1A = PDP^{-1} or A=PCP−1A = PCP^{-1}. If the eigenvalues are real and distinct, such as λ=2,3\lambda = 2, 3 for [2003]\begin{bmatrix} 2 & 0 \\ 0 & 3 \end{bmatrix}, there is no rotation at all. Do not use CC; diagonalize as A=PDP−1A = PDP^{-1}, where the real eigenvectors stay on their own lines and DD scales along them.

When det⁡(A−λI)=0\det(A - \lambda I) = 0 gives a complex conjugate pair λ=a±bi\lambda = a \pm bi, the matrix is rotating space. In both cases below, the real part aa and imaginary part bb are the horizontal and vertical legs of a right triangle that set the scale factor and rotation angle at each step:

r=a2+b2,cos⁡θ=ar,sin⁡θ=br\displaystyle \begin{aligned} \scalefactor{r} &= \sqrt{a^2 + b^2}, \\ \cos\turnangle{\theta} &= \frac{a}{\scalefactor{r}}, \\ \sin\turnangle{\theta} &= \frac{b}{\scalefactor{r}} \end{aligned}

If the matrix already has equal diagonal entries and opposite off-diagonal entries like C=[3−443]C = \begin{bmatrix} 3 & -4 \\ 4 & 3 \end{bmatrix} (λ=3±4i\lambda = 3 \pm 4i), it is already rotating on the standard grid, so no basis matrix PP is needed; just factor out the hypotenuse r=5\scalefactor{r} = 5:

[a−bba]=r[cos⁡θ−sin⁡θsin⁡θcos⁡θ]\displaystyle \begin{bmatrix} a & -b \\ b & a \end{bmatrix} = \scalefactor{r}\begin{bmatrix} \cos\turnangle{\theta} & -\sin\turnangle{\theta} \\ \sin\turnangle{\theta} & \cos\turnangle{\theta} \end{bmatrix}

If the matrix does not have the [a−bba]\begin{bmatrix} a & -b \\ b & a \end{bmatrix} pattern, like A=[3−211]A = \begin{bmatrix} 3 & -2 \\ 1 & 1 \end{bmatrix} (λ=2±i\lambda = 2 \pm i), it is rotating on a skewed grid, so factor it as A=PCP−1A = PCP^{-1}. Solve Av=λvAv = \lambda v for just one eigenvalue (λ=2+i\lambda = 2 + i) using the simpler row to get v=u+iw=[11]+i[10]v = \blocktwo{u} + i\blockone{w} = \blocktwo{\begin{bmatrix} 1 \\ 1 \end{bmatrix}} + i\blockone{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} (the conjugate eigenvalue λˉ=2−i\bar{\lambda} = 2 - i automatically has eigenvector vˉ=u−iw\bar{v} = \blocktwo{u} - i\blockone{w}). Put the real vectors u\blocktwo{u} and w\blockone{w} into the columns of PP, and match the position of −b-b in CC to the column of u\blocktwo{u} using Au=au−bwA\blocktwo{u} = a\blocktwo{u} - b\blockone{w} and Aw=bu+awA\blockone{w} = b\blocktwo{u} + a\blockone{w}:

P=[uw]  ⟹  C=[ab−ba],P=[wu]  ⟹  C=[a−bba]\displaystyle \begin{aligned} P = \begin{bmatrix} \blocktwo{u} & \blockone{w} \end{bmatrix} &\implies C = \begin{bmatrix} a & b \\ -b & a \end{bmatrix}, \\ P = \begin{bmatrix} \blockone{w} & \blocktwo{u} \end{bmatrix} &\implies C = \begin{bmatrix} a & -b \\ b & a \end{bmatrix} \end{aligned}
6

Practice

1

Find the complex eigenvalues λ\lambda, the scale factor r\scalefactor{r}, and the counterclockwise rotation angle θ\turnangle{\theta} of the matrix C=[3−113]C = \begin{bmatrix} \sqrt{3} & -1 \\ 1 & \sqrt{3} \end{bmatrix}, and write CC in the factored form r[cos⁡θ−sin⁡θsin⁡θcos⁡θ]\scalefactor{r}\begin{bmatrix} \cos\turnangle{\theta} & -\sin\turnangle{\theta} \\ \sin\turnangle{\theta} & \cos\turnangle{\theta} \end{bmatrix}.

[3−113]\begin{bmatrix} \sqrt{3} & -1 \\ 1 & \sqrt{3} \end{bmatrix} has the form [a−bba]\begin{bmatrix} a & -b \\ b & a \end{bmatrix}: right triangle with horizontal leg a=3a = \sqrt{3} and vertical leg b=1b = 1, hypotenuse r=?\scalefactor{r} = \text{?}

Matching C=[3−113]C = \begin{bmatrix} \sqrt{3} & -1 \\ 1 & \sqrt{3} \end{bmatrix} with [a−bba]\begin{bmatrix} a & -b \\ b & a \end{bmatrix} gives the horizontal and vertical legs and the complex eigenvalues:
a\displaystyle a =3,b\displaystyle = \sqrt{3}, \qquad b =1,λ\displaystyle = 1, \qquad \lambda =3±i\displaystyle = \sqrt{3} \pm i
(Watch out: the top-right entry of [a−bba]\begin{bmatrix} a & -b \\ b & a \end{bmatrix} is −b-b, so b=1b = 1, not −1-1. Reading b=−1b = -1 from the top-right gives sin⁡θ=−12\sin\turnangle{\theta} = -\tfrac{1}{2} and the wrong clockwise angle θ=−π6\turnangle{\theta} = -\tfrac{\pi}{6}.)
Find the scale factor r\scalefactor{r} from the hypotenuse of the right triangle:
r\displaystyle \scalefactor{r} =(3)2+12\displaystyle = \sqrt{(\sqrt{3})^2 + 1^2} =3+1\displaystyle = \sqrt{3 + 1} =4\displaystyle = \sqrt{4} =2\displaystyle = 2
Use the legs over the hypotenuse to find the rotation angle θ\turnangle{\theta}:
cos⁡θ=32,sin⁡θ=12  ⟹  θ=π6 radians (or 30∘)\displaystyle \begin{aligned} \cos\turnangle{\theta} &= \frac{\sqrt{3}}{2}, \\ \sin\turnangle{\theta} &= \frac{1}{2} \\ &\implies \turnangle{\theta} = \frac{\pi}{6}\text{ radians (or } 30^\circ\text{)} \end{aligned}
Factor out r=2\scalefactor{r} = 2:
C=2[cos⁡(π6)−sin⁡(π6)sin⁡(π6)cos⁡(π6)]\displaystyle C = \scalefactor{2}\begin{bmatrix} \cos(\turnangle{\tfrac{\pi}{6}}) & -\sin(\turnangle{\tfrac{\pi}{6}}) \\ \sin(\turnangle{\tfrac{\pi}{6}}) & \cos(\turnangle{\tfrac{\pi}{6}}) \end{bmatrix}
2

A real 2×22 \times 2 matrix AA has eigenvalue λ=3+4i\lambda = 3 + 4i with complex eigenvector v=[1+2i1]=u+iwv = \begin{bmatrix} 1 + 2i \\ 1 \end{bmatrix} = \blocktwo{u} + i\blockone{w}. Without finding AA itself: (a) write the other eigenvalue of AA and its eigenvector, (b) express AuA\blocktwo{u} and AwA\blockone{w} in terms of u\blocktwo{u} and w\blockone{w}, and (c) write PP and CC in A=PCP−1A = PCP^{-1} both when P=[uw]P = \begin{bmatrix} \blocktwo{u} & \blockone{w} \end{bmatrix} and when P=[wu]P = \begin{bmatrix} \blockone{w} & \blocktwo{u} \end{bmatrix}.

Split vv into two completely real vectors u\blocktwo{u} and w\blockone{w} first, then expand A(u+iw)=(3+4i)(u+iw)A(\blocktwo{u} + i\blockone{w}) = (3 + 4i)(\blocktwo{u} + i\blockone{w}) and match the real and imaginary parts.

Splitting vv into its real and imaginary parts gives:
v=[1+2i1]=[11]+i[20]  ⟹  u=[11],w=[20]\displaystyle \begin{aligned} v &= \begin{bmatrix} 1 + 2i \\ 1 \end{bmatrix} = \blocktwo{\begin{bmatrix} 1 \\ 1 \end{bmatrix}} + i\blockone{\begin{bmatrix} 2 \\ 0 \end{bmatrix}} \\ &\implies \blocktwo{u} = \blocktwo{\begin{bmatrix} 1 \\ 1 \end{bmatrix}}, \quad \blockone{w} = \blockone{\begin{bmatrix} 2 \\ 0 \end{bmatrix}} \end{aligned}
(a)Take the complex conjugate of λ\lambda and vv by flipping the sign of ii:
λˉ=3−4i,vˉ=u−iw=[1−2i1]\displaystyle \bar{\lambda} = 3 - 4i, \qquad \bar{v} = \blocktwo{u} - i\blockone{w} = \begin{bmatrix} 1 - 2i \\ 1 \end{bmatrix}
(b)Expand Av=λvAv = \lambda v using i2=−1i^2 = -1:
A(u+iw)=(3+4i)(u+iw)Au+i(Aw)=3u+3iw+4iu+4i2wAu+i(Aw)=(3u−4w)+i(4u+3w)\displaystyle \begin{aligned} A(\blocktwo{u} + i\blockone{w}) &= (3 + 4i)(\blocktwo{u} + i\blockone{w}) \\ A\blocktwo{u} + i(A\blockone{w}) &= 3\blocktwo{u} + 3i\blockone{w} + 4i\blocktwo{u} \\ &\quad + 4i^2\blockone{w} \\ A\blocktwo{u} + i(A\blockone{w}) &= (3\blocktwo{u} - 4\blockone{w}) \\ &\quad + i(4\blocktwo{u} + 3\blockone{w}) \end{aligned}
(Watch out: (4i)(iw)=4i2w=−4w(4i)(i\blockone{w}) = 4i^2\blockone{w} = -4\blockone{w}; forgetting i2=−1i^2 = -1 gives the wrong real part Au=3u+4wA\blocktwo{u} = 3\blocktwo{u} + 4\blockone{w}.)
Matching the real and imaginary parts on both sides gives:
Au\displaystyle A\blocktwo{u} =3u−4w,Aw\displaystyle = 3\blocktwo{u} - 4\blockone{w}, \qquad A\blockone{w} =4u+3w\displaystyle = 4\blocktwo{u} + 3\blockone{w}
(c)Reading the columns of CC from Au=3u−4wA\blocktwo{u} = 3\blocktwo{u} - 4\blockone{w} and Aw=4u+3wA\blockone{w} = 4\blocktwo{u} + 3\blockone{w} for each column order of PP (putting C=[3−443]C = \begin{bmatrix} 3 & -4 \\ 4 & 3 \end{bmatrix} with P=[uw]P = \begin{bmatrix} \blocktwo{u} & \blockone{w} \end{bmatrix} puts −4-4 in the wrong column, matching Aw=−4u+3wA\blockone{w} = -4\blocktwo{u} + 3\blockone{w} instead):
When P=[uw]=[1210],C=[34−43]\displaystyle \begin{aligned} &\text{When } P = \begin{bmatrix} \blocktwo{u} & \blockone{w} \end{bmatrix} = \begin{bmatrix} \blocktwo{1} & \blockone{2} \\ \blocktwo{1} & \blockone{0} \end{bmatrix}, \\ &C = \begin{bmatrix} 3 & 4 \\ -4 & 3 \end{bmatrix} \end{aligned}
When P=[wu]=[2101],C=[3−443]\displaystyle \begin{aligned} &\text{When } P = \begin{bmatrix} \blockone{w} & \blocktwo{u} \end{bmatrix} = \begin{bmatrix} \blockone{2} & \blocktwo{1} \\ \blockone{0} & \blocktwo{1} \end{bmatrix}, \\ &C = \begin{bmatrix} 3 & -4 \\ 4 & 3 \end{bmatrix} \end{aligned}
3

Factor the matrix A=[3−51−1]A = \begin{bmatrix} 3 & -5 \\ 1 & -1 \end{bmatrix} as A=PCP−1A = PCP^{-1}, giving the matrices PP, CC, and P−1P^{-1} explicitly.

After finding λ=a±bi\lambda = a \pm bi from det⁡(A−λI)=0\det(A - \lambda I) = 0, use the bottom row of Av=λvAv = \lambda v to express xx in terms of yy.

Set the determinant of A−λIA - \lambda I to zero to find the eigenvalues:
(3−λ)(−1−λ)−(−5)(1)=0λ2−2λ−3+5=0λ2−2λ+2=0(λ−1)2+1=0λ=1±i\displaystyle \begin{aligned} (3 - \lambda)(-1 - \lambda) - (-5)(1) &= 0 \\ \lambda^2 - 2\lambda - 3 + 5 &= 0 \\ \lambda^2 - 2\lambda + 2 &= 0 \\ (\lambda - 1)^2 + 1 &= 0 \\ \lambda &= 1 \pm i \end{aligned}
So a=1a = 1 and b=1b = 1. For λ=1+i\lambda = 1 + i, write out the bottom row of [3−51−1][xy]=(1+i)[xy]\begin{bmatrix} 3 & -5 \\ 1 & -1 \end{bmatrix}\begin{bmatrix} x \\ y \end{bmatrix} = (1 + i)\begin{bmatrix} x \\ y \end{bmatrix}:
x−y=(1+i)yx=y+(1+i)yx=(2+i)y\displaystyle \begin{aligned} x - y &= (1 + i)y \\ x &= y + (1 + i)y \\ x &= (2 + i)y \end{aligned}
(Watch out: when adding yy to (1+i)y(1 + i)y, the real part increases from 11 to 22; dropping the −y-y term on the left without adding it on the right gives x=(1+i)yx = (1 + i)y and the wrong basis vector u=[11]\blocktwo{u} = \begin{bmatrix} 1 \\ 1 \end{bmatrix}.)
Setting y=1y = 1 gives x=2+ix = 2 + i, which splits into u\blocktwo{u} and w\blockone{w}:
v=[2+i1]=[21]+i[10]  ⟹  u=[21],w=[10]\displaystyle \begin{aligned} v &= \begin{bmatrix} 2 + i \\ 1 \end{bmatrix} = \blocktwo{\begin{bmatrix} 2 \\ 1 \end{bmatrix}} + i\blockone{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} \\ &\implies \blocktwo{u} = \blocktwo{\begin{bmatrix} 2 \\ 1 \end{bmatrix}}, \quad \blockone{w} = \blockone{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} \end{aligned}
Using P=[uw]P = \begin{bmatrix} \blocktwo{u} & \blockone{w} \end{bmatrix} (and remembering to divide by det⁡(P)=2(0)−1(1)=−1\det(P) = 2(0) - 1(1) = -1 when inverting PP, rather than forgetting to divide and leaving [0−1−12]\begin{bmatrix} 0 & -1 \\ -1 & 2 \end{bmatrix}):
P=[2110],C=[11−11],P−1=[011−2]\displaystyle \begin{aligned} P &= \begin{bmatrix} \blocktwo{2} & \blockone{1} \\ \blocktwo{1} & \blockone{0} \end{bmatrix}, \\ C &= \begin{bmatrix} 1 & 1 \\ -1 & 1 \end{bmatrix}, \\ P^{-1} &= \begin{bmatrix} 0 & 1 \\ 1 & -2 \end{bmatrix} \end{aligned}
(Or, using P=[wu]P = \begin{bmatrix} \blockone{w} & \blocktwo{u} \end{bmatrix}: P=[1201]P = \begin{bmatrix} \blockone{1} & \blocktwo{2} \\ \blockone{0} & \blocktwo{1} \end{bmatrix}, C=[1−111]C = \begin{bmatrix} 1 & -1 \\ 1 & 1 \end{bmatrix}, and P−1=[1−201]P^{-1} = \begin{bmatrix} 1 & -2 \\ 0 & 1 \end{bmatrix}.) Starting from λ=1−i\lambda = 1 - i instead gives v=[2−i1]v = \begin{bmatrix} 2 - i \\ 1 \end{bmatrix}, so u=[21]\blocktwo{u} = \blocktwo{\begin{bmatrix} 2 \\ 1 \end{bmatrix}} and w=[−10]\blockone{w} = \blockone{\begin{bmatrix} -1 \\ 0 \end{bmatrix}}, and P=[uw]=[2−110]P = \begin{bmatrix} \blocktwo{u} & \blockone{w} \end{bmatrix} = \begin{bmatrix} \blocktwo{2} & \blockone{-1} \\ \blocktwo{1} & \blockone{0} \end{bmatrix} gives C=[1−111]C = \begin{bmatrix} 1 & -1 \\ 1 & 1 \end{bmatrix} and P−1=[01−12]P^{-1} = \begin{bmatrix} 0 & 1 \\ -1 & 2 \end{bmatrix}, an equally valid factorization.
4

When finding the eigenvector of A=[1−21−1]A = \begin{bmatrix} 1 & -2 \\ 1 & -1 \end{bmatrix} for λ=i\lambda = i from x=(1+i)yx = (1 + i)y, one student picks y=1y = 1 to get v1=u1+iw1v_1 = \blocktwo{u_1} + i\blockone{w_1} and a second student picks y=iy = i to get v2=u2+iw2v_2 = \blocktwo{u_2} + i\blockone{w_2}. Find u2\blocktwo{u_2} and w2\blockone{w_2}, and explain how the two pairs of real basis vectors are related on the tilted ellipse.

Substitute y=iy = i into both entries of v2=[(1+i)yy]v_2 = \begin{bmatrix} (1 + i)y \\ y \end{bmatrix} and expand (1+i)i(1 + i)i using i2=−1i^2 = -1 before splitting v2v_2 into its real and imaginary parts.

For y=1y = 1, the first eigenvector splits into:
v1=[1+i1]=[11]+i[10]  ⟹  u1=[11],w1=[10]\displaystyle \begin{aligned} v_1 &= \begin{bmatrix} 1 + i \\ 1 \end{bmatrix} = \blocktwo{\begin{bmatrix} 1 \\ 1 \end{bmatrix}} + i\blockone{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} \\ &\implies \blocktwo{u_1} = \blocktwo{\begin{bmatrix} 1 \\ 1 \end{bmatrix}}, \quad \blockone{w_1} = \blockone{\begin{bmatrix} 1 \\ 0 \end{bmatrix}} \end{aligned}
For y=iy = i, substitute y=iy = i into both entries of [xy]\begin{bmatrix} x \\ y \end{bmatrix}:
v2=[(1+i)ii]=[i+i2i]=[−1+ii]\displaystyle v_2 = \begin{bmatrix} (1 + i)i \\ i \end{bmatrix} = \begin{bmatrix} i + i^2 \\ i \end{bmatrix} = \begin{bmatrix} -1 + i \\ i \end{bmatrix}
(Watch out: the bottom entry of v2v_2 is y=iy = i, not 11. Leaving the bottom entry as 11 gives [−1+i1]\begin{bmatrix} -1 + i \\ 1 \end{bmatrix} and the wrong pair u2=[−11],w2=[10]\blocktwo{u_2} = \begin{bmatrix} -1 \\ 1 \end{bmatrix}, \blockone{w_2} = \begin{bmatrix} 1 \\ 0 \end{bmatrix}.)
Split v2v_2 into its real and imaginary parts:
v2=[−10]+i[11]  ⟹  u2=[−10],w2=[11]\displaystyle \begin{aligned} v_2 &= \blocktwo{\begin{bmatrix} -1 \\ 0 \end{bmatrix}} + i\blockone{\begin{bmatrix} 1 \\ 1 \end{bmatrix}} \\ &\implies \blocktwo{u_2} = \blocktwo{\begin{bmatrix} -1 \\ 0 \end{bmatrix}}, \quad \blockone{w_2} = \blockone{\begin{bmatrix} 1 \\ 1 \end{bmatrix}} \end{aligned}
Because v2=iv1v_2 = iv_1, multiplying the eigenvector by ii rotates both basis vectors forward by one 90∘90^\circ step along the exact same tilted ellipse: w2=u1\blockone{w_2} = \blocktwo{u_1} and u2=−w1\blocktwo{u_2} = -\blockone{w_1}.
5

For the matrix A=[2−210]A = \begin{bmatrix} 2 & -2 \\ 1 & 0 \end{bmatrix}, find its complex eigenvalues, determine the scale factor r\scalefactor{r} and rotation angle θ\turnangle{\theta} in the tilted basis, and use them to compute A8[35]A^8\orbit{\begin{bmatrix} 3 \\ 5 \end{bmatrix}}.

Once you have r\scalefactor{r} and θ\turnangle{\theta} from λ=a+bi\lambda = a + bi, find the total rotation angle 8θ\turnangle{8\theta} and total scaling r8\scalefactor{r^8} in C8C^8 before deciding whether you need to compute PP.

Find the eigenvalues from det⁡(A−λI)=0\det(A - \lambda I) = 0:
(2−λ)(0−λ)−(−2)(1)=0λ2−2λ+2=0(λ−1)2+1=0λ=1±i\displaystyle \begin{aligned} (2 - \lambda)(0 - \lambda) - (-2)(1) &= 0 \\ \lambda^2 - 2\lambda + 2 &= 0 \\ (\lambda - 1)^2 + 1 &= 0 \\ \lambda &= 1 \pm i \end{aligned}
Read off a=1a = 1 and b=1b = 1 to find the scale factor r\scalefactor{r} and rotation angle θ\turnangle{\theta} in the tilted basis:
r=12+12=2,cos⁡θ=12,sin⁡θ=12  ⟹  θ=π4 (or 45∘)\displaystyle \begin{aligned} \scalefactor{r} &= \sqrt{1^2 + 1^2} = \sqrt{2}, \\ \cos\turnangle{\theta} &= \frac{1}{\sqrt{2}}, \\ \sin\turnangle{\theta} &= \frac{1}{\sqrt{2}} \\ &\implies \turnangle{\theta} = \frac{\pi}{4}\text{ (or } 45^\circ\text{)} \end{aligned}
Over 88 steps, the total rotation angle and total scale factor are:
8θ=8×45∘=360∘,r8=(2)8=24=16\displaystyle \begin{aligned} \turnangle{8\theta} &= 8 \times 45^\circ = 360^\circ, \\ \scalefactor{r^8} &= (\sqrt{2})^8 = 2^4 = 16 \end{aligned}
(Watch out: r=12+12=2\scalefactor{r} = \sqrt{1^2 + 1^2} = \sqrt{2}, not 22; forgetting the square root gives 28=2562^8 = 256 and the wrong vector [7681280]\begin{bmatrix} 768 \\ 1280 \end{bmatrix}.)
Because a 360∘360^\circ rotation brings every vector all the way back to its starting direction, C8=16IC^8 = 16I, which makes PP and P−1P^{-1} cancel out:
A8=PC8P−1=P(16I)P−1=16IA8[35]=16[35]=[4880]\displaystyle \begin{aligned} & A^8 = PC^8P^{-1} = P(16I)P^{-1} \\ &\quad = 16I \\ & A^8\orbit{\begin{bmatrix} 3 \\ 5 \end{bmatrix}} = 16\orbit{\begin{bmatrix} 3 \\ 5 \end{bmatrix}} = \begin{bmatrix} 48 \\ 80 \end{bmatrix} \end{aligned}