Linear AlgebraDeeper Theory6 of 6
Deeper Theory · 6

The Matrix Exponential and Systems of Differential Equations

We want to solve a system of differential equations where the rates of change of x\state{x} and y\state{y} depend on both variables at once:

dxdt=x+2ydydt=3x+2y\displaystyle \begin{aligned} \frac{d\state{x}}{dt} &= \state{x} + 2\state{y} \\ \frac{d\state{y}}{dt} &= 3\state{x} + 2\state{y} \end{aligned}

Because y\state{y} is in the equation for dxdt\frac{d\state{x}}{dt} and x\state{x} is in the equation for dydt\frac{d\state{y}}{dt}, we cannot solve for either variable on its own. Putting the two variables into a vector turns the coupled system into a single matrix equation that looks just like dxdt=3x\frac{d\state{x}}{dt} = \matA{3}\state{x}.

Writing a coupled system as dvdt=Av\frac{dv}{dt} = Av
dxdt=3x\dfrac{d\state{x}}{dt} = \matA{3}\state{x}
x(t)=  ?\state{x(t)} = \;?
x\htmlClass{lc-blue}{x}
3x\htmlClass{lc-violet}{3x}
0\htmlClass{lc-mute}{0}
1\htmlClass{lc-mute}{1}
2\htmlClass{lc-mute}{2}
3\htmlClass{lc-mute}{3}
4\htmlClass{lc-mute}{4}
5\htmlClass{lc-mute}{5}
6\htmlClass{lc-mute}{6}
7\htmlClass{lc-mute}{7}

Writing the coordinates as a vector v\state{v} and the coefficients as a matrix A\matA{A} puts the entire system into one equation:

ddt[xy]⏟v=[1232]⏟A[xy]⏟v  ⟹  dvdt=Av\displaystyle \frac{d}{dt}\underbrace{\state{\begin{bmatrix} x \\ y \end{bmatrix}}}_{\state{v}} = \underbrace{\matA{\begin{bmatrix} 1 & 2 \\ 3 & 2 \end{bmatrix}}}_{\matA{A}}\underbrace{\state{\begin{bmatrix} x \\ y \end{bmatrix}}}_{\state{v}} \quad \implies \quad \frac{d\state{v}}{dt} = \matA{A}\state{v}

Here dvdt\frac{d\state{v}}{dt} just means taking the derivative of each entry in the vector, [dxdtdydt]\begin{bmatrix} \tfrac{d\state{x}}{dt} \\ \tfrac{d\state{y}}{dt} \end{bmatrix}. Just as dxdt=3x\frac{d\state{x}}{dt} = \matA{3}\state{x} has the solution x(t)=e3tx0\state{x(t)} = e^{\matA{3}t}\state{x_0}, this matrix equation has a solution of the form v(t)=eAtv0\state{v(t)} = e^{\matA{A}t}\state{v_0}.

1

Defining the matrix exponential eAte^{At}

The matrix exponential eAte^{\matA{A}t} is the 2×22 \times 2 matrix that plays the role of e3te^{\matA{3}t}, giving the solution v(t)=eAtv0\state{v(t)} = e^{\matA{A}t}\state{v_0} to dvdt=Av\frac{d\state{v}}{dt} = \matA{A}\state{v} from any starting vector v0\state{v_0}.

To see how a matrix can go into an exponent, remember the Taylor series for e3te^{\matA{3}t}, whose derivative is 3\matA{3} times itself:

e3t\displaystyle e^{\matA{3}t} =1+3t+(3t)22!+(3t)33!+⋯\displaystyle = 1 + \matA{3}t + \frac{(\matA{3}t)^2}{2!} + \frac{(\matA{3}t)^3}{3!} + \cdots

Because this infinite series only uses powers, division by numbers, and addition, we can plug a square matrix A\matA{A} into the exact same formula.

Defining the matrix exponential eAte^{At}
dxdt=3xdvdt=Avx(t)=e3tx0v(t)=eAtv0\displaystyle \begin{alignedat}{2} \frac{d\state{x}}{dt} &= \matA{3}\state{x} &\qquad \frac{d\state{v}}{dt} &= \matA{A}\state{v} \\[6pt] \state{x(t)} &= e^{\matA{3}t}\state{x_0} &\qquad \state{v(t)} &= e^{\matA{A}t}\state{v_0} \end{alignedat}
eAt\displaystyle e^{\matA{A}t} =  ?\displaystyle = \;?

Replacing the number 11 with the identity matrix II and 3\matA{3} with A\matA{A} defines the matrix exponential:

eAt⏟matrix exponential=[1001]⏟I+At+A2t22!+A3t33!+⋯\displaystyle \underbrace{e^{\matA{A}t}}_{\text{matrix exponential}} = \underbrace{\begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix}}_{I} + \matA{A}t + \frac{\matA{A}^2t^2}{2!} + \frac{\matA{A}^3t^3}{3!} + \cdots

Just like e3te^{\matA{3}t}, eAte^{\matA{A}t} has derivative ddt(eAt)=AeAt\frac{d}{dt}(e^{\matA{A}t}) = \matA{A}e^{\matA{A}t} and starts at eA⋅0=Ie^{\matA{A}\cdot 0} = I when t=0t = 0. That makes the solution to dvdt=Av\frac{d\state{v}}{dt} = \matA{A}\state{v} starting at v0\state{v_0}:

v(t)\displaystyle \state{v(t)} =eAtv0\displaystyle = e^{\matA{A}t}\state{v_0}

For a general matrix A\matA{A}, adding up I+At+A2t22!+⋯I + \matA{A}t + \frac{\matA{A}^2t^2}{2!} + \cdots by hand means summing four infinite series at once, one for each entry of the 2×22 \times 2 matrix. When the matrix is diagonal, however, x\state{x} and y\state{y} never mix, and raising the matrix to powers only raises its diagonal entries to powers:

eDt=exp⁡([4t00−t])=[e4t00e−t]\displaystyle e^{Dt} = \exp\left(\begin{bmatrix} \vone{4t} & 0 \\ 0 & \vtwo{-t} \end{bmatrix}\right) = \begin{bmatrix} \vone{e^{4t}} & 0 \\ 0 & \vtwo{e^{-t}} \end{bmatrix}
2

Solving a system with eigenvectors

For a coupled matrix, the off-diagonal numbers mix x\state{x} and y\state{y} together, so we cannot just put each entry into an exponent. We want to solve our original coupled system from the starting point (5,5)\state{(5, 5)}:

ddt[xy]=[1232][xy],[x(0)y(0)]=[55]\displaystyle \frac{d}{dt}\state{\begin{bmatrix} x \\ y \end{bmatrix}} = \matA{\begin{bmatrix} 1 & 2 \\ 3 & 2 \end{bmatrix}}\state{\begin{bmatrix} x \\ y \end{bmatrix}}, \quad \state{\begin{bmatrix} x(0) \\ y(0) \end{bmatrix}} = \state{\begin{bmatrix} 5 \\ 5 \end{bmatrix}}

Geometrically, your position v\state{v} has velocity Av\matA{A}\state{v}. At most points, Av\matA{A}\state{v} points in a new direction and pushes the particle off the line through v\state{v}, but along an eigenvector line (Av=λv\matA{A}\state{v} = \lambda \state{v}) the velocity stays on that line and the motion simplifies to dvdt=λv\frac{d\state{v}}{dt} = \lambda \state{v}.

Solving a system with eigenvectors
ddt[xy]=[1232][xy]\dfrac{d}{dt}\state{\begin{bmatrix} x \\ y \end{bmatrix}} = \matA{\begin{bmatrix} 1 & 2 \\ 3 & 2 \end{bmatrix}}\state{\begin{bmatrix} x \\ y \end{bmatrix}}
[x(t)y(t)]=  ?\state{\begin{bmatrix} x(t) \\ y(t) \end{bmatrix}} = \;?
[55]\htmlClass{lc-blue}{\begin{bmatrix} 5 \\ 5 \end{bmatrix}}
Av\htmlClass{lc-violet}{Av}
Text solution
Find the eigenvalues of A\matA{A} to get the growth ratesSetting det⁡(A−λI)=0\det(\matA{A} - \lambda I) = 0 gives the growth rates λ\lambda for the system.
det⁡[1−λ232−λ]=(1−λ)(2−λ)−6=λ2−3λ−4=(λ−4)(λ+1)=0  ⟹  λ1=4,    λ2=−1\displaystyle \begin{aligned} & \det\begin{bmatrix} 1 - \lambda & 2 \\ 3 & 2 - \lambda \end{bmatrix} \\ &\quad = (1 - \lambda)(2 - \lambda) - 6 \\ &\quad = \lambda^2 - 3\lambda - 4 \\ &\quad = \vone{(\lambda - 4)}\vtwo{(\lambda + 1)} = 0 \\ &\quad \implies \vone{\lambda_1 = 4}, \;\; \vtwo{\lambda_2 = -1} \end{aligned}
Find the eigenvectors to get the straight-line directionsPlugging each eigenvalue into (A−λI)v=0(\matA{A} - \lambda I)v = 0 gives the two lines where the variables do not mix.
For λ1=4:[−323−2][xy]=[00]  ⟹  3x=2y  ⟹  v1=[23]For λ2=−1:[2233][xy]=[00]  ⟹  x=−y  ⟹  v2=[1−1]\displaystyle \begin{aligned} & \text{For } \vone{\lambda_1 = 4}: \\ &\quad \begin{bmatrix} -3 & 2 \\ 3 & -2 \end{bmatrix}\begin{bmatrix} x \\ y \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \end{bmatrix} \\ &\quad \implies 3x = 2y \\ &\quad \implies \vone{v_1 = \begin{bmatrix} 2 \\ 3 \end{bmatrix}} \\[10pt] & \text{For } \vtwo{\lambda_2 = -1}: \\ &\quad \begin{bmatrix} 2 & 2 \\ 3 & 3 \end{bmatrix}\begin{bmatrix} x \\ y \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \end{bmatrix} \\ &\quad \implies x = -y \\ &\quad \implies \vtwo{v_2 = \begin{bmatrix} 1 \\ -1 \end{bmatrix}} \end{aligned}
Split the starting point into eigenvector piecesBecause v1\vone{v_1} and v2\vtwo{v_2} point in different directions, we can write [55]\state{\begin{bmatrix} 5 \\ 5 \end{bmatrix}} as a combination of them by adding the two row equations.
c1[23]+c2[1−1]=[55]  ⟹  {2c1+c2=53c1−c2=5  ⟹  5c1=10  ⟹  c1=2,    c2=1\displaystyle \vone{c_1\begin{bmatrix} 2 \\ 3 \end{bmatrix}} + \vtwo{c_2\begin{bmatrix} 1 \\ -1 \end{bmatrix}} = \state{\begin{bmatrix} 5 \\ 5 \end{bmatrix}} \implies \left\{\begin{aligned} 2\vone{c_1} + \vtwo{c_2} &= \state{5} \\ 3\vone{c_1} - \vtwo{c_2} &= \state{5} \end{aligned}\right. \implies 5\vone{c_1} = 10 \implies \vone{c_1 = 2}, \;\; \vtwo{c_2 = 1}
Grow each eigenvector piece by its own eλte^{\lambda t}Letting the v1\vone{v_1} piece grow by e4t\vone{e^{4t}} and the v2\vtwo{v_2} piece shrink by e−t\vtwo{e^{-t}} gives the path at any time tt.
[x(t)y(t)]=2e4t[23]+1e−t[1−1]=[4e4t+e−t6e4t−e−t]\displaystyle \begin{aligned} \state{\begin{bmatrix} x(t) \\ y(t) \end{bmatrix}} &= \vone{2e^{4t}\begin{bmatrix} 2 \\ 3 \end{bmatrix}} + \vtwo{1e^{-t}\begin{bmatrix} 1 \\ -1 \end{bmatrix}} \\ &= \begin{bmatrix} \vone{4e^{4t}} + \vtwo{e^{-t}} \\ \vone{6e^{4t}} - \vtwo{e^{-t}} \end{bmatrix} \end{aligned}

As tt grows, the e−t\vtwo{e^{-t}} term shrinks to zero, so the particle moves closer to the v1=[23]\vone{v_1 = \begin{bmatrix} 2 \\ 3 \end{bmatrix}} line as its v2=[1−1]\vtwo{v_2 = \begin{bmatrix} 1 \\ -1 \end{bmatrix}} piece shrinks, and moves outward along v1\vone{v_1}, growing like e4t\vone{e^{4t}}.

Starting directly on either eigenvector line keeps the particle on that line forever, while any other starting point bends toward the v1\vone{v_1} line as its v2\vtwo{v_2} piece shrinks by e−t\vtwo{e^{-t}}, and moves outward along v1\vone{v_1} as its v1\vone{v_1} piece grows by e4t\vone{e^{4t}}. A starting point close to the v2\vtwo{v_2} line first comes in toward the origin along v2\vtwo{v_2} before it turns out along v1\vone{v_1}.

Every starting point splits into a shrinking piece along one eigenvector line and a growing piece along the other, so as the shrinking piece dies out, the path bends toward the growing line and heads out along it.
3

Computing eAte^{At} by diagonalization

Splitting a starting vector into eigenvectors solved the system for one initial point, but we can also put those same three steps into a single formula to compute the matrix eAte^{\matA{A}t} itself for A=[1232]\matA{A} = \matA{\begin{bmatrix} 1 & 2 \\ 3 & 2 \end{bmatrix}}.

When A\matA{A} has enough independent eigenvectors to make an invertible matrix PP, we can factor A=PDP−1\matA{A} = PDP^{-1} and pull PP and P−1P^{-1} outside the Taylor series for eAte^{\matA{A}t}.

Computing eAte^{At} by diagonalization
A=[1232]\matA{A} = \matA{\begin{bmatrix} 1 & 2 \\ 3 & 2 \end{bmatrix}}
eAt=  ?e^{\matA{A}t} = \;?
v1=[23]\vone{v_1 = \begin{bmatrix} 2 \\ 3 \end{bmatrix}}
λ1=4\vone{\lambda_1 = 4}
v2=[1−1]\vtwo{v_2 = \begin{bmatrix} 1 \\ -1 \end{bmatrix}}
λ2=−1\vtwo{\lambda_2 = -1}

Reading from right to left when multiplying a starting vector v0\state{v_0}, factoring PP and P−1P^{-1} out of the series turns eAte^{\matA{A}t} into our three-step eigenvector method:

eAt\displaystyle e^{\matA{A}t} =P⏟back to(x,y)eDt⏟grow alongeigenvectorsP−1⏟into eigenvectorcoordinates\displaystyle = \underbrace{P}_{\substack{\text{back to} \\ (x, y)}} \underbrace{e^{Dt}}_{\substack{\text{grow along} \\ \text{eigenvectors}}} \underbrace{P^{-1}}_{\substack{\text{into eigenvector} \\ \text{coordinates}}}
Text solution
Set up PP, eDte^{Dt}, and P−1P^{-1} from the eigenvectors and eigenvaluesPut the eigenvectors v1=[23]\vone{v_1 = \begin{bmatrix} 2 \\ 3 \end{bmatrix}} and v2=[1−1]\vtwo{v_2 = \begin{bmatrix} 1 \\ -1 \end{bmatrix}} into the columns of PP and match that column order with λ1=4\vone{\lambda_1 = 4} and λ2=−1\vtwo{\lambda_2 = -1} on the diagonal of eDte^{Dt}.
P=[213−1],eDt=[e4t00e−t],P−1=1−5[−1−1−32]=15[113−2]\displaystyle \begin{aligned} P &= \begin{bmatrix} \vone{2} & \vtwo{1} \\ \vone{3} & \vtwo{-1} \end{bmatrix}, \\ e^{Dt} &= \begin{bmatrix} \vone{e^{4t}} & 0 \\ 0 & \vtwo{e^{-t}} \end{bmatrix}, \\ P^{-1} &= \frac{1}{-5}\begin{bmatrix} -1 & -1 \\ -3 & 2 \end{bmatrix} \\ &= \frac{1}{5}\begin{bmatrix} 1 & 1 \\ 3 & -2 \end{bmatrix} \end{aligned}
Multiply PeDtP−1Pe^{Dt}P^{-1} out in orderMultiplying the first two matrices multiplies each eigenvector column by its exponential before P−1P^{-1} combines them.
eAt=[213−1][e4t00e−t]15[113−2]=15[2e4te−t3e4t−e−t][113−2]=15[2e4t+3e−t2e4t−2e−t3e4t−3e−t3e4t+2e−t]\displaystyle \begin{aligned} e^{\matA{A}t} &= \begin{bmatrix} \vone{2} & \vtwo{1} \\ \vone{3} & \vtwo{-1} \end{bmatrix}\begin{bmatrix} \vone{e^{4t}} & 0 \\ 0 & \vtwo{e^{-t}} \end{bmatrix}\frac{1}{5}\begin{bmatrix} 1 & 1 \\ 3 & -2 \end{bmatrix} \\ &= \frac{1}{5}\begin{bmatrix} \vone{2e^{4t}} & \vtwo{e^{-t}} \\ \vone{3e^{4t}} & \vtwo{-e^{-t}} \end{bmatrix}\begin{bmatrix} 1 & 1 \\ 3 & -2 \end{bmatrix} \\ &= \frac{1}{5}\begin{bmatrix} \vone{2e^{4t}} + \vtwo{3e^{-t}} & \vone{2e^{4t}} - \vtwo{2e^{-t}} \\ \vone{3e^{4t}} - \vtwo{3e^{-t}} & \vone{3e^{4t}} + \vtwo{2e^{-t}} \end{bmatrix} \end{aligned}

Plugging t=0t = 0 into your final matrix is the quickest way to catch arithmetic mistakes: since e0=1e^0 = 1, eA⋅0e^{\matA{A}\cdot 0} must simplify to the identity matrix [1001]\begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix}. Also keep in mind that eAt=PeDtP−1e^{\matA{A}t} = Pe^{Dt}P^{-1} requires A\matA{A} to have enough independent eigenvectors to make PP invertible; if a matrix is missing an eigenvector, P−1P^{-1} does not exist even though the infinite series eAte^{\matA{A}t} still does.

4

Complex eigenvalues and spiraling solutions

When a system rotates vectors in the plane, there are no real straight-line eigenvectors, and det⁡(A−λI)=0\det(\matA{A} - \lambda I) = 0 gives complex eigenvalues λ=a±bi\lambda = a \pm bi. We want to find the real path (x(t),y(t))\state{(x(t), y(t))} for a system with rotation:

ddt[xy]=[0−21−2][xy],[x(0)y(0)]=[31]\displaystyle \frac{d}{dt}\state{\begin{bmatrix} x \\ y \end{bmatrix}} = \matA{\begin{bmatrix} 0 & -2 \\ 1 & -2 \end{bmatrix}}\state{\begin{bmatrix} x \\ y \end{bmatrix}}, \quad \state{\begin{bmatrix} x(0) \\ y(0) \end{bmatrix}} = \state{\begin{bmatrix} 3 \\ 1 \end{bmatrix}}

To turn a complex eigenvalue λ=a+bi\lambda = a + bi into real functions of time, we use Euler's formula to split the exponent into a real growth factor and a rotation:

e(a+bi)t\displaystyle e^{(a + bi)t} =eateibt\displaystyle = e^{at}e^{ibt} =eat⏟size(cos⁡bt+isin⁡bt)⏟rotation\displaystyle = \underbrace{e^{at}}_{\text{size}}\underbrace{(\cos bt + i\sin bt)}_{\text{rotation}}
Complex eigenvalues and spiraling solutions
ddt[xy]=[0−21−2][xy]\dfrac{d}{dt}\state{\begin{bmatrix} x \\ y \end{bmatrix}} = \matA{\begin{bmatrix} 0 & -2 \\ 1 & -2 \end{bmatrix}}\state{\begin{bmatrix} x \\ y \end{bmatrix}}
[x(t)y(t)]=  ?\state{\begin{bmatrix} x(t) \\ y(t) \end{bmatrix}} = \;?
[31]\htmlClass{lc-blue}{\begin{bmatrix} 3 \\ 1 \end{bmatrix}}
Av\htmlClass{lc-violet}{Av}
Text solution
Find the complex eigenvalues and one eigenvectorWe only need the eigenvector for one of the two complex conjugate eigenvalues, so we plug λ=−1+i\lambda = -1 + i into (A−λI)v=0(\matA{A} - \lambda I)v = 0 and use the simpler bottom row.
det⁡[−λ−21−2−λ]=λ2+2λ+2=(λ+1)2+1=0  ⟹  λ=−1±i\displaystyle \det\begin{bmatrix} -\lambda & -2 \\ 1 & -2 - \lambda \end{bmatrix} = \lambda^2 + 2\lambda + 2 = (\lambda + 1)^2 + 1 = 0 \implies \lambda = -1 \pm i
For λ=−1+i:[1−i−21−1−i][xy]=[00]  ⟹  x=(1+i)y  ⟹  v=[1+i1]\displaystyle \text{For } \lambda = -1 + i:\quad \begin{bmatrix} 1 - i & -2 \\ 1 & -1 - i \end{bmatrix}\begin{bmatrix} x \\ y \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \end{bmatrix} \implies x = (1 + i)y \implies v = \begin{bmatrix} 1 + i \\ 1 \end{bmatrix}
Expand eλtve^{\lambda t}v and split into real and imaginary partsMultiplying (cos⁡t+isin⁡t)(\cos t + i\sin t) into the vector using i2=−1i^2 = -1 gives two independent real solutions, u(t)\realsol{u(t)} and w(t)\imagsol{w(t)}.
e(−1+i)t[1+i1]=e−t(cos⁡t+isin⁡t)[1+i1]=e−t[(cos⁡t−sin⁡t)+i(cos⁡t+sin⁡t)cos⁡t+isin⁡t]=e−t[cos⁡t−sin⁡tcos⁡t]⏟u(t) (real part)+i e−t[cos⁡t+sin⁡tsin⁡t]⏟w(t) (imaginary part)\displaystyle \begin{aligned} & e^{(-1 + i)t}\begin{bmatrix} 1 + i \\ 1 \end{bmatrix} \\ &\quad = e^{-t}(\cos t + i\sin t)\begin{bmatrix} 1 + i \\ 1 \end{bmatrix} \\ &\quad = e^{-t}\begin{bmatrix} \begin{gathered}\realsol{(\cos t - \sin t)} \\ + i\imagsol{(\cos t + \sin t)}\end{gathered} \\[8pt] \realsol{\cos t} + i\imagsol{\sin t} \end{bmatrix} \\ &\quad = \underbrace{\realsol{e^{-t}\begin{bmatrix} \cos t - \sin t \\ \cos t \end{bmatrix}}}_{\realsol{u(t)}\text{ (real part)}} \\ &\qquad + i\,\underbrace{\imagsol{e^{-t}\begin{bmatrix} \cos t + \sin t \\ \sin t \end{bmatrix}}}_{\imagsol{w(t)}\text{ (imaginary part)}} \end{aligned}
Combine the two real solutions to match the starting pointSetting [x(t)y(t)]=c1u(t)+c2w(t)\state{\begin{bmatrix} x(t) \\ y(t) \end{bmatrix}} = \realsol{c_1 u(t)} + \imagsol{c_2 w(t)} and plugging in t=0t = 0 (where e0=1e^0 = 1, cos⁡0=1\cos 0 = 1, and sin⁡0=0\sin 0 = 0) gives c1\realsol{c_1} and c2\imagsol{c_2}.
c1[11]+c2[10]=[31]  ⟹  c1=1,    c2=2\displaystyle \realsol{c_1\begin{bmatrix} 1 \\ 1 \end{bmatrix}} + \imagsol{c_2\begin{bmatrix} 1 \\ 0 \end{bmatrix}} = \state{\begin{bmatrix} 3 \\ 1 \end{bmatrix}} \implies \realsol{c_1 = 1}, \;\; \imagsol{c_2 = 2}
[x(t)y(t)]=1e−t[cos⁡t−sin⁡tcos⁡t]+2e−t[cos⁡t+sin⁡tsin⁡t]=e−t[3cos⁡t+sin⁡tcos⁡t+2sin⁡t]\displaystyle \begin{aligned} \state{\begin{bmatrix} x(t) \\ y(t) \end{bmatrix}} &= \realsol{1e^{-t}\begin{bmatrix} \cos t - \sin t \\ \cos t \end{bmatrix}} \\ &\quad + \imagsol{2e^{-t}\begin{bmatrix} \cos t + \sin t \\ \sin t \end{bmatrix}} \\ &= e^{-t}\begin{bmatrix} 3\cos t + \sin t \\ \cos t + 2\sin t \end{bmatrix} \end{aligned}

Starting at (3,1)\state{(3, 1)}, the cosine and sine terms rotate the particle around the origin while e−te^{-t} shrinks its distance to (0,0)(0, 0).

5

Long-term behavior from the eigenvalues

Often we don't need the full formula for x(t)\state{x(t)} and y(t)\state{y(t)}. We just want to look at the eigenvalues of A\matA{A} and know what happens to the system as t→∞t \to \infty.

In discrete systems vk+1=Avk\state{v_{k+1}} = \matA{A}\state{v_k}, each step multiplies by λ\lambda, so λk\lambda^k shrinks or grows without bound depending on whether the magnitude ∣λ∣|\lambda| is smaller or larger than 11. In continuous differential equations dvdt=Av\frac{d\state{v}}{dt} = \matA{A}\state{v}, λ\lambda is in the exponent of eλte^{\lambda t}, which changes the condition we check.

Long-term behavior from the eigenvalues
dvdt=Av\dfrac{d\state{v}}{dt} = \matA{A}\state{v}
t→∞:v(t)→  ?t \to \infty\text{:}\quad \state{v(t)} \to \;?
[31]\htmlClass{lc-blue}{\begin{bmatrix} 3 \\ 1 \end{bmatrix}}
v(t)\htmlClass{lc-blue}{v(t)}

Because e(a+bi)t=eat(cos⁡bt+isin⁡bt)e^{(a+bi)t} = e^{at}(\cos bt + i\sin bt), the imaginary part bibi only brings in cos⁡bt\cos bt and sin⁡bt\sin bt, which keep repeating, so it makes the path go around the origin without making it grow or shrink over time. All growth or decay comes from eate^{at}, which shrinks when the real part a=Re(λ)a = \text{Re}(\lambda) is negative and grows when Re(λ)\text{Re}(\lambda) is positive:

Re(λ)<0: all paths decay to (0,0)\text{Re}(\lambda) < 0\text{: all paths decay to }(0, 0)
λ=−2, −5\lambda = -2,\ -5
λ=−1±i\lambda = -1 \pm i
Re(λ)>0: non-zero paths blow up\text{Re}(\lambda) > 0\text{: non-zero paths blow up}
λ=2, 4\lambda = 2,\ 4
λ=1±i\lambda = 1 \pm i
Mixed signs\text{Mixed signs}
(λ1>0, λ2<0):(\htmlClass{lc-green}{\lambda_1 > 0},\ \htmlClass{lc-pink}{\lambda_2 < 0})\text{:}
saddle\text{saddle}
λ=±bi,\lambda = \pm bi,
Re(λ)=0:\text{Re}(\lambda) = 0\text{:}
closed loops\text{closed loops}
λ1=4, λ2=−1\htmlClass{lc-green}{\lambda_1 = 4},\ \htmlClass{lc-pink}{\lambda_2 = -1}
λ=±2i\lambda = \pm 2i
v1\htmlClass{lc-green}{v_1}
v2\htmlClass{lc-pink}{v_2}

In the complex plane, crossing the vertical line Re(λ)=0\text{Re}(\lambda) = 0 changes inward spirals (Re(λ)<0\text{Re}(\lambda) < 0) into closed orbits (Re(λ)=0\text{Re}(\lambda) = 0) and then outward spirals (Re(λ)>0\text{Re}(\lambda) > 0). The imaginary part bb sets the rotation speed, and when b=0b = 0 on the real axis, the paths do not rotate at all.

Which side of the vertical line a complex eigenvalue sits on sets whether solution paths shrink, orbit, or grow, while its height only sets their spin speed.

Key concepts

Every coupled linear system dvdt=Av\frac{d\state{v}}{dt} = \matA{A}\state{v} starting at v0\state{v_0} has the solution v(t)=eAtv0\state{v(t)} = e^{\matA{A}t}\state{v_0}, where eAt=I+At+A2t22!+A3t33!+⋯e^{\matA{A}t} = I + \matA{A}t + \frac{\matA{A}^2t^2}{2!} + \frac{\matA{A}^3t^3}{3!} + \cdots. How you compute that solution depends on the matrix A\matA{A}:

When not to find eigenvectors

If the matrix is already diagonal, D=[400−1]D = \begin{bmatrix} \vone{4} & 0 \\ 0 & \vtwo{-1} \end{bmatrix}, the variables never mix, so just put the diagonal entries into exponents: eDt=[e4t00e−t]e^{Dt} = \begin{bmatrix} \vone{e^{4t}} & 0 \\ 0 & \vtwo{e^{-t}} \end{bmatrix}. If A\matA{A} is missing an eigenvector (such as A=[0400]\matA{A} = \matA{\begin{bmatrix} 0 & 4 \\ 0 & 0 \end{bmatrix}}), P−1P^{-1} does not exist, so compute A2,A3,…\matA{A}^2, \matA{A}^3, \dots and plug them straight into the Taylor series I+At+A2t22!+⋯I + \matA{A}t + \frac{\matA{A}^2t^2}{2!} + \cdots.

Distinct real eigenvalues (λ1,λ2\vone{\lambda_1}, \vtwo{\lambda_2})

To solve for one starting point v0\state{v_0}, split v0=c1v1+c2v2\state{v_0} = \vone{c_1 v_1} + \vtwo{c_2 v_2} along the eigenvectors and multiply each piece by eλte^{\lambda t}, giving v(t)=c1eλ1tv1+c2eλ2tv2\state{v(t)} = \vone{c_1 e^{\lambda_1 t}v_1} + \vtwo{c_2 e^{\lambda_2 t}v_2}. To find the full matrix eAte^{\matA{A}t}, use eAt=PeDtP−1e^{\matA{A}t} = Pe^{Dt}P^{-1} and check that t=0t = 0 gives II.

Complex eigenvalues (λ=a±bi\lambda = a \pm bi)

Find one eigenvector vv for λ=a+bi\lambda = a + bi, expand eat(cos⁡bt+isin⁡bt)v=u(t)+i w(t)e^{at}(\cos bt + i\sin bt)v = \realsol{u(t)} + i\,\imagsol{w(t)}, and use the real and imaginary parts as your two real solutions: [x(t)y(t)]=c1u(t)+c2w(t)\state{\begin{bmatrix} x(t) \\ y(t) \end{bmatrix}} = \realsol{c_1 u(t)} + \imagsol{c_2 w(t)}.

Long-term behavior (t→∞t \to \infty)

Look at the sign of Re(λ)\text{Re}(\lambda) compared to 00 (not ∣λ∣|\lambda| compared to 11). When both Re(λ)<0\text{Re}(\lambda) < 0, paths decay to [00]\begin{bmatrix} 0 \\ 0 \end{bmatrix}; when both Re(λ)>0\text{Re}(\lambda) > 0, paths grow without bound; opposite signs (λ1>0,λ2<0\vone{\lambda_1 > 0}, \vtwo{\lambda_2 < 0}) form a saddle that pulls paths in along v2\vtwo{v_2} and sends them out along v1\vone{v_1}; and pure imaginary eigenvalues λ=±bi\lambda = \pm bi (Re(λ)=0\text{Re}(\lambda) = 0) give closed loops.

6

Practice

1

Find the matrix exponential eDte^{Dt} for D=[−2005]D = \begin{bmatrix} \vone{-2} & 0 \\ 0 & \vtwo{5} \end{bmatrix}, and use it to solve ddt[xy]=[−2005][xy]\frac{d}{dt}\state{\begin{bmatrix} x \\ y \end{bmatrix}} = \begin{bmatrix} \vone{-2} & 0 \\ 0 & \vtwo{5} \end{bmatrix}\state{\begin{bmatrix} x \\ y \end{bmatrix}} with starting point [x(0)y(0)]=[4−3]\state{\begin{bmatrix} x(0) \\ y(0) \end{bmatrix}} = \state{\begin{bmatrix} 4 \\ -3 \end{bmatrix}}.

Look at the off-diagonal zeros: do x\state{x} and y\state{y} mix, and what happens to the zero entries in every term of the Taylor series for eDte^{Dt}?

Because DD is diagonal, each diagonal entry goes into its own exponential while the off-diagonal zeros stay 00 in every power of DD:
eDt=[e−2t00e5t]\displaystyle e^{Dt} = \begin{bmatrix} \vone{e^{-2t}} & 0 \\ 0 & \vtwo{e^{5t}} \end{bmatrix}
(Watch out not to turn the off-diagonal zeros into e0=1e^0 = 1, which would turn eDte^{Dt} into [e−2t11e5t]\begin{bmatrix} e^{-2t} & 1 \\ 1 & e^{5t} \end{bmatrix} and give [4e−2t−34−3e5t]\begin{bmatrix} 4e^{-2t} - 3 \\ 4 - 3e^{5t} \end{bmatrix}.) Multiplying eDte^{Dt} by the starting point [4−3]\state{\begin{bmatrix} 4 \\ -3 \end{bmatrix}} gives:
[x(t)y(t)]=[e−2t00e5t][4−3]=[4e−2t−3e5t]\displaystyle \state{\begin{bmatrix} x(t) \\ y(t) \end{bmatrix}} = \begin{bmatrix} \vone{e^{-2t}} & 0 \\ 0 & \vtwo{e^{5t}} \end{bmatrix}\state{\begin{bmatrix} 4 \\ -3 \end{bmatrix}} = \begin{bmatrix} \vone{4e^{-2t}} \\ \vtwo{-3e^{5t}} \end{bmatrix}
So x(t)=4e−2t\state{x(t)} = \vone{4e^{-2t}} and y(t)=−3e5t\state{y(t)} = \vtwo{-3e^{5t}}.
2

Solve the coupled system dxdt=2x+3y\frac{d\state{x}}{dt} = 2\state{x} + 3\state{y}, dydt=4x+y\frac{d\state{y}}{dt} = 4\state{x} + \state{y} with starting point [x(0)y(0)]=[5−2]\state{\begin{bmatrix} x(0) \\ y(0) \end{bmatrix}} = \state{\begin{bmatrix} 5 \\ -2 \end{bmatrix}} by finding the eigenvalues and eigenvectors of the matrix.

Write the system with A=[2341]\matA{A} = \matA{\begin{bmatrix} 2 & 3 \\ 4 & 1 \end{bmatrix}}, find the two eigenvector lines where the variables do not mix, and solve c1v1+c2v2=[5−2]\vone{c_1 v_1} + \vtwo{c_2 v_2} = \state{\begin{bmatrix} 5 \\ -2 \end{bmatrix}}.

First find the eigenvalues of A=[2341]\matA{A} = \matA{\begin{bmatrix} 2 & 3 \\ 4 & 1 \end{bmatrix}}:
det⁡[2−λ341−λ]=(2−λ)(1−λ)−12=λ2−3λ−10=(λ−5)(λ+2)=0\displaystyle \begin{aligned} & \det\begin{bmatrix} 2 - \lambda & 3 \\ 4 & 1 - \lambda \end{bmatrix} \\ &\quad = (2 - \lambda)(1 - \lambda) - 12 \\ &\quad = \lambda^2 - 3\lambda - 10 \\ &\quad = \vone{(\lambda - 5)}\vtwo{(\lambda + 2)} = 0 \end{aligned}
So λ1=5\vone{\lambda_1 = 5} and λ2=−2\vtwo{\lambda_2 = -2}. Plugging each into (A−λI)v=0(\matA{A} - \lambda I)v = 0 gives the eigenvectors:
For λ1=5:[−334−4][xy]=[00]  ⟹  x=y  ⟹  v1=[11]For λ2=−2:[4343][xy]=[00]  ⟹  4x=−3y  ⟹  v2=[3−4]\displaystyle \begin{aligned} & \text{For } \vone{\lambda_1 = 5}: \\ &\quad \begin{bmatrix} -3 & 3 \\ 4 & -4 \end{bmatrix}\begin{bmatrix} x \\ y \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \end{bmatrix} \\ &\quad \implies x = y \\ &\quad \implies \vone{v_1 = \begin{bmatrix} 1 \\ 1 \end{bmatrix}} \\[10pt] & \text{For } \vtwo{\lambda_2 = -2}: \\ &\quad \begin{bmatrix} 4 & 3 \\ 4 & 3 \end{bmatrix}\begin{bmatrix} x \\ y \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \end{bmatrix} \\ &\quad \implies 4x = -3y \\ &\quad \implies \vtwo{v_2 = \begin{bmatrix} 3 \\ -4 \end{bmatrix}} \end{aligned}
Now split the starting vector [5−2]\state{\begin{bmatrix} 5 \\ -2 \end{bmatrix}} into pieces along v1\vone{v_1} and v2\vtwo{v_2} (be careful not to use the starting coordinates 5\state{5} and −2\state{-2} directly as c1\vone{c_1} and c2\vtwo{c_2}, which would give x(t)=5e5t−6e−2tx(t) = 5e^{5t} - 6e^{-2t} and y(t)=5e5t+8e−2ty(t) = 5e^{5t} + 8e^{-2t}):
c1[11]+c2[3−4]=[5−2]  ⟹  c1=2,    c2=1\displaystyle \vone{c_1\begin{bmatrix} 1 \\ 1 \end{bmatrix}} + \vtwo{c_2\begin{bmatrix} 3 \\ -4 \end{bmatrix}} = \state{\begin{bmatrix} 5 \\ -2 \end{bmatrix}} \implies \vone{c_1 = 2}, \;\; \vtwo{c_2 = 1}
Multiplying each piece by its own eλte^{\lambda t} gives:
[x(t)y(t)]=2e5t[11]+1e−2t[3−4]=[2e5t+3e−2t2e5t−4e−2t]\displaystyle \begin{aligned} \state{\begin{bmatrix} x(t) \\ y(t) \end{bmatrix}} &= \vone{2e^{5t}\begin{bmatrix} 1 \\ 1 \end{bmatrix}} + \vtwo{1e^{-2t}\begin{bmatrix} 3 \\ -4 \end{bmatrix}} \\ &= \begin{bmatrix} \vone{2e^{5t}} + \vtwo{3e^{-2t}} \\ \vone{2e^{5t}} - \vtwo{4e^{-2t}} \end{bmatrix} \end{aligned}
So x(t)=2e5t+3e−2t\state{x(t)} = \vone{2e^{5t}} + \vtwo{3e^{-2t}} and y(t)=2e5t−4e−2t\state{y(t)} = \vone{2e^{5t}} - \vtwo{4e^{-2t}}.
3

Compute the matrix exponential eAte^{\matA{A}t} for A=[4−211]\matA{A} = \matA{\begin{bmatrix} 4 & -2 \\ 1 & 1 \end{bmatrix}} using eAt=PeDtP−1e^{\matA{A}t} = Pe^{Dt}P^{-1}, and check that your answer gives II at t=0t = 0.

After finding the eigenvalues and eigenvectors, put the eigenvectors into the columns of PP in the same order as the eigenvalues in eDte^{Dt}.

First find the eigenvalues and eigenvectors of A\matA{A}:
det⁡[4−λ−211−λ]=(4−λ)(1−λ)+2=λ2−5λ+6=(λ−3)(λ−2)=0\displaystyle \begin{aligned} & \det\begin{bmatrix} 4 - \lambda & -2 \\ 1 & 1 - \lambda \end{bmatrix} \\ &\quad = (4 - \lambda)(1 - \lambda) + 2 \\ &\quad = \lambda^2 - 5\lambda + 6 \\ &\quad = \vone{(\lambda - 3)}\vtwo{(\lambda - 2)} = 0 \end{aligned}
For λ1=3\vone{\lambda_1 = 3}, [1−21−2][xy]=[00]\begin{bmatrix} 1 & -2 \\ 1 & -2 \end{bmatrix}\begin{bmatrix} x \\ y \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \end{bmatrix} gives v1=[21]\vone{v_1 = \begin{bmatrix} 2 \\ 1 \end{bmatrix}}. For λ2=2\vtwo{\lambda_2 = 2}, [2−21−1][xy]=[00]\begin{bmatrix} 2 & -2 \\ 1 & -1 \end{bmatrix}\begin{bmatrix} x \\ y \end{bmatrix} = \begin{bmatrix} 0 \\ 0 \end{bmatrix} gives v2=[11]\vtwo{v_2 = \begin{bmatrix} 1 \\ 1 \end{bmatrix}}. This gives PP, DD, and P−1P^{-1}:
P=[2111],D=[3002],P−1=12(1)−1(1)[1−1−12]=[1−1−12]\displaystyle \begin{aligned} P &= \begin{bmatrix} \vone{2} & \vtwo{1} \\ \vone{1} & \vtwo{1} \end{bmatrix}, \\ D &= \begin{bmatrix} \vone{3} & 0 \\ 0 & \vtwo{2} \end{bmatrix}, \\ P^{-1} &= \frac{1}{2(1) - 1(1)}\begin{bmatrix} 1 & -1 \\ -1 & 2 \end{bmatrix} \\ &= \begin{bmatrix} 1 & -1 \\ -1 & 2 \end{bmatrix} \end{aligned}
(Make sure to swap the diagonal entries when inverting PP; forgetting to swap 22 and 11 gives [2−1−11]\begin{bmatrix} 2 & -1 \\ -1 & 1 \end{bmatrix}, which leads to [4e3t−e2t−2e3t+e2t2e3t−e2t−e3t+e2t]\begin{bmatrix} 4e^{3t} - e^{2t} & -2e^{3t} + e^{2t} \\ 2e^{3t} - e^{2t} & -e^{3t} + e^{2t} \end{bmatrix} and fails the t=0t = 0 check by giving [3−110]\begin{bmatrix} 3 & -1 \\ 1 & 0 \end{bmatrix}.) Multiplying PeDtP−1Pe^{Dt}P^{-1} gives:
eAt=[2111][e3t00e2t][1−1−12]=[2e3te2te3te2t][1−1−12]=[2e3t−e2t−2e3t+2e2te3t−e2t−e3t+2e2t]\displaystyle \begin{aligned} e^{\matA{A}t} &= \begin{bmatrix} \vone{2} & \vtwo{1} \\ \vone{1} & \vtwo{1} \end{bmatrix}\begin{bmatrix} \vone{e^{3t}} & 0 \\ 0 & \vtwo{e^{2t}} \end{bmatrix}\begin{bmatrix} 1 & -1 \\ -1 & 2 \end{bmatrix} \\ &= \begin{bmatrix} \vone{2e^{3t}} & \vtwo{e^{2t}} \\ \vone{e^{3t}} & \vtwo{e^{2t}} \end{bmatrix}\begin{bmatrix} 1 & -1 \\ -1 & 2 \end{bmatrix} \\ &= \begin{bmatrix} \vone{2e^{3t}} - \vtwo{e^{2t}} & \vone{-2e^{3t}} + \vtwo{2e^{2t}} \\ \vone{e^{3t}} - \vtwo{e^{2t}} & \vone{-e^{3t}} + \vtwo{2e^{2t}} \end{bmatrix} \end{aligned}
At t=0t = 0, this equals [2−1−2+21−1−1+2]=[1001]=I\begin{bmatrix} 2 - 1 & -2 + 2 \\ 1 - 1 & -1 + 2 \end{bmatrix} = \begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix} = I.
4

The matrix A=[0400]\matA{A} = \matA{\begin{bmatrix} 0 & 4 \\ 0 & 0 \end{bmatrix}} has only one eigenvector direction, so it cannot be diagonalized as PDP−1PDP^{-1}. Compute A2\matA{A}^2, then use the Taylor series eAt=I+At+A2t22!+A3t33!+⋯e^{\matA{A}t} = I + \matA{A}t + \frac{\matA{A}^2t^2}{2!} + \frac{\matA{A}^3t^3}{3!} + \cdots directly to find eAte^{\matA{A}t} and check that ddt(eAt)=AeAt\frac{d}{dt}(e^{\matA{A}t}) = \matA{A}e^{\matA{A}t}.

Multiply A\matA{A} by itself first. How many terms of the infinite Taylor series are non-zero?

Multiplying A\matA{A} by itself gives:
A2=[0400][0400]=[0000]\displaystyle \matA{A}^2 = \matA{\begin{bmatrix} 0 & 4 \\ 0 & 0 \end{bmatrix}}\matA{\begin{bmatrix} 0 & 4 \\ 0 & 0 \end{bmatrix}} = \begin{bmatrix} 0 & 0 \\ 0 & 0 \end{bmatrix}
Because A2\matA{A}^2 is the zero matrix, every higher power Ak\matA{A}^k for k≥2k \ge 2 is also [0000]\begin{bmatrix} 0 & 0 \\ 0 & 0 \end{bmatrix}. All terms from A2t22!\frac{\matA{A}^2t^2}{2!} onward are zero, leaving only the first two terms (don't forget to include the identity matrix I=[1001]I = \begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix} at the start, or you would get [04t00]\begin{bmatrix} 0 & 4t \\ 0 & 0 \end{bmatrix}):
eAt=I+At=[1001]+[04t00]=[14t01]\displaystyle e^{\matA{A}t} = I + \matA{A}t = \begin{bmatrix} 1 & 0 \\ 0 & 1 \end{bmatrix} + \begin{bmatrix} 0 & 4t \\ 0 & 0 \end{bmatrix} = \begin{bmatrix} 1 & 4t \\ 0 & 1 \end{bmatrix}
Taking the derivative of each entry gives ddt(eAt)=[0400]\frac{d}{dt}(e^{\matA{A}t}) = \begin{bmatrix} 0 & 4 \\ 0 & 0 \end{bmatrix}, which matches AeAt=[0400][14t01]=[0400]\matA{A}e^{\matA{A}t} = \matA{\begin{bmatrix} 0 & 4 \\ 0 & 0 \end{bmatrix}}\begin{bmatrix} 1 & 4t \\ 0 & 1 \end{bmatrix} = \begin{bmatrix} 0 & 4 \\ 0 & 0 \end{bmatrix}.
5

Solve ddt[xy]=[2−510][xy]\frac{d}{dt}\state{\begin{bmatrix} x \\ y \end{bmatrix}} = \matA{\begin{bmatrix} 2 & -5 \\ 1 & 0 \end{bmatrix}}\state{\begin{bmatrix} x \\ y \end{bmatrix}} with starting point [x(0)y(0)]=[51]\state{\begin{bmatrix} x(0) \\ y(0) \end{bmatrix}} = \state{\begin{bmatrix} 5 \\ 1 \end{bmatrix}}, and describe the path as tt increases.

Find one eigenvector vv for λ=1+2i\lambda = 1 + 2i, and expand et(cos⁡2t+isin⁡2t)ve^t(\cos 2t + i\sin 2t)v using i2=−1i^2 = -1 to separate the real part u(t)\realsol{u(t)} and imaginary part w(t)\imagsol{w(t)}.

First find the eigenvalues of A=[2−510]\matA{A} = \matA{\begin{bmatrix} 2 & -5 \\ 1 & 0 \end{bmatrix}}:
det⁡[2−λ−51−λ]=λ2−2λ+5=(λ−1)2+4=0  ⟹  λ=1±2i\displaystyle \det\begin{bmatrix} 2 - \lambda & -5 \\ 1 & -\lambda \end{bmatrix} = \lambda^2 - 2\lambda + 5 = (\lambda - 1)^2 + 4 = 0 \implies \lambda = 1 \pm 2i
For λ=1+2i\lambda = 1 + 2i, the bottom row of (A−λI)v=0(\matA{A} - \lambda I)v = 0 is 1x+(−1−2i)y=01x + (-1 - 2i)y = 0, so x=(1+2i)yx = (1 + 2i)y, giving v=[1+2i1]v = \begin{bmatrix} 1 + 2i \\ 1 \end{bmatrix}. Expand e(1+2i)tve^{(1 + 2i)t}v using Euler's formula (watch the sign on 2i2sin⁡2t=−2sin⁡2t2i^2\sin 2t = -2\sin 2t in the top entry; writing +2sin⁡2t+2\sin 2t instead leads to x(t)=et(5cos⁡2t+4sin⁡2t)x(t) = e^t(5\cos 2t + 4\sin 2t)):
e(1+2i)t[1+2i1]=et(cos⁡2t+isin⁡2t)[1+2i1]=et[cos⁡2t−2sin⁡2tcos⁡2t]⏟u(t)+i et[2cos⁡2t+sin⁡2tsin⁡2t]⏟w(t)\displaystyle \begin{aligned} & e^{(1 + 2i)t}\begin{bmatrix} 1 + 2i \\ 1 \end{bmatrix} \\ & = e^t(\cos 2t + i\sin 2t)\begin{bmatrix} 1 + 2i \\ 1 \end{bmatrix} \\ & = \underbrace{\realsol{e^t\begin{bmatrix} \cos 2t - 2\sin 2t \\ \cos 2t \end{bmatrix}}}_{\realsol{u(t)}} \\ & + i\,\underbrace{\imagsol{e^t\begin{bmatrix} 2\cos 2t + \sin 2t \\ \sin 2t \end{bmatrix}}}_{\imagsol{w(t)}} \end{aligned}
Matching c1u(0)+c2w(0)=[51]\realsol{c_1 u(0)} + \imagsol{c_2 w(0)} = \state{\begin{bmatrix} 5 \\ 1 \end{bmatrix}} at t=0t = 0 gives:
c1[11]+c2[20]=[51]  ⟹  c1=1,    c2=2\displaystyle \realsol{c_1\begin{bmatrix} 1 \\ 1 \end{bmatrix}} + \imagsol{c_2\begin{bmatrix} 2 \\ 0 \end{bmatrix}} = \state{\begin{bmatrix} 5 \\ 1 \end{bmatrix}} \implies \realsol{c_1 = 1}, \;\; \imagsol{c_2 = 2}
Combining 1u(t)+2w(t)\realsol{1u(t)} + \imagsol{2w(t)} gives the solution:
[x(t)y(t)]=et[5cos⁡2tcos⁡2t+2sin⁡2t]\displaystyle \state{\begin{bmatrix} x(t) \\ y(t) \end{bmatrix}} = e^t\begin{bmatrix} 5\cos 2t \\ \cos 2t + 2\sin 2t \end{bmatrix}
The path starts at (5,1)\state{(5, 1)} and spirals outward away from the origin because Re(λ)=1>0\text{Re}(\lambda) = 1 > 0.
6

For each matrix A\matA{A} below, determine the long-term behavior of solutions to the continuous system dvdt=Av\frac{d\state{v}}{dt} = \matA{A}\state{v} as t→∞t \to \infty and compare it with the discrete system vk+1=Avk\state{v_{k+1}} = \matA{A}\state{v_k} as k→∞k \to \infty: (a) A\matA{A} has eigenvalues λ1=−0.5\vone{\lambda_1 = -0.5} (with eigenvector v1=[12]\vone{v_1 = \begin{bmatrix} 1 \\ 2 \end{bmatrix}}) and λ2=0.5\vtwo{\lambda_2 = 0.5} (with eigenvector v2=[1−1]\vtwo{v_2 = \begin{bmatrix} 1 \\ -1 \end{bmatrix}}). (b) A\matA{A} has pure imaginary eigenvalues λ=±2i\lambda = \pm 2i.

Check how eλte^{\lambda t} behaves (sign of Re(λ)\text{Re}(\lambda) compared to 00) versus how λk\lambda^k behaves (magnitude ∣λ∣|\lambda| compared to 11).

(a) In continuous time, λ1=−0.5<0\vone{\lambda_1 = -0.5} < 0 and λ2=0.5>0\vtwo{\lambda_2 = 0.5} > 0 have opposite signs (don't use the discrete test ∣0.5∣<1|\vtwo{0.5}| < 1 to conclude that the λ2=0.5\vtwo{\lambda_2 = 0.5} piece decays; in e0.5t\vtwo{e^{0.5t}}, a positive exponent grows). The system forms a saddle: starting points on the v1=[12]\vone{v_1 = \begin{bmatrix} 1 \\ 2 \end{bmatrix}} line decay to [00]\begin{bmatrix} 0 \\ 0 \end{bmatrix} like e−0.5t\vone{e^{-0.5t}}, while every other starting point has a piece along v2=[1−1]\vtwo{v_2 = \begin{bmatrix} 1 \\ -1 \end{bmatrix}} that grows like e0.5t\vtwo{e^{0.5t}} without bound. In discrete time, both eigenvalues have magnitude ∣λ∣=0.5<1|\lambda| = 0.5 < 1, so (−0.5)k\vone{(-0.5)^k} and (0.5)k\vtwo{(0.5)^k} both go to 00 and every starting point decays to [00]\begin{bmatrix} 0 \\ 0 \end{bmatrix}.
(b) In continuous time, Re(λ)=0\text{Re}(\lambda) = 0, so the size factor is e0t=1e^{0t} = 1 and non-zero solutions orbit the origin in closed loops forever without growing or shrinking. In discrete time, ∣±2i∣=2>1|\pm 2i| = 2 > 1, so each eigenvector piece is multiplied by a number of magnitude 22 at every step, the pieces grow like 2k2^k, and every non-zero starting point grows without bound as k→∞k \to \infty.