Eigenvalue Method for Linear Systems

The eigenvalue method is the computational core of stability analysis throughout applied mathematics and engineering. In classical mechanics, small oscillations about equilibrium are governed by linear systems whose eigenvalues determine normal-mode frequencies; a purely imaginary eigenvalue signals oscillation, while a real positive eigenvalue signals instability. In electrical engineering, RLC circuit networks reduce to exactly this framework, with eigenvalues predicting whether transients decay, oscillate, or blow up. In population biology and epidemiology, linearizing a nonlinear model near an equilibrium (the technique behind Phase Portraits for Linear Systems) produces a Jacobian matrix whose eigenvalues classify the equilibrium as stable, unstable, or a saddle. In Markov chain theory and numerical analysis, the same eigen-decomposition idea underlies the convergence rate of iterative algorithms and the long-run behavior of continuous-time Markov processes. In control theory, placing eigenvalues of a closed-loop system matrix in the left half-plane (via feedback) is the literal engineering goal of pole placement, a direct descendant of this method. Even in cryptography and data science, eigen-decompositions of transition or covariance matrices (principal component analysis, spectral clustering) reuse the same linear-algebra machinery that makes this ODE method work.

Ask the tutor: How does the eigenvalue method change when the matrix has a complex conjugate pair of eigenvalues instead of real ones?

Picture the plane $\mathbb{R}^2$ with a matrix $A$ that has two distinct real eigenvalues $\lambda_1 \ne \lambda_2$ and corresponding eigenvectors $\mathbf{v}_1, \mathbf{v}_2$. These two eigenvectors point along two special directions through the origin. If you release a particle exactly on the line through $\mathbf{v}_1$, the vector field $A\mathbf{x}$ at that point is a multiple of $\mathbf{v}_1$ itself — the flow never pushes the particle off the line. It simply scales the position by $e^{\lambda_1 t}$ as time passes: motion along that line is one-dimensional exponential growth or decay in disguise. The same holds for $\mathbf{v}_2$ with rate $\lambda_2$. For a general initial condition not on either eigenline, decompose it as $\mathbf{x}(0) = c_1\mathbf{v}_1 + c_2\mathbf{v}_2$ — possible precisely because two linearly independent vectors form a basis of $\mathbb{R}^2$. Linearity of the system then guarantees that the solution is just the superposition of the two eigen-solutions, each evolving independently along its own line:

The eigenvectors thus act as natural axes for the dynamics — the coordinate system in which the tangled system becomes two unrelated exponential clocks running at different rates $\lambda_1$ and $\lambda_2$.

Example 1 — an asymmetric $2\times 2$ system. Solve

Step 1: Eigenvalues. Solve $\det(A - \lambda I) = 0$:

Factor: $(\lambda - 5)(\lambda + 1) = 0$, so $\lambda_1 = 5$, $\lambda_2 = -1$. These are real and distinct, so the theorem applies. Step 2: Eigenvectors. For $\lambda_1 = 5$: solve $(A - 5I)\mathbf{v} = 0$, i.e.

Take $\mathbf{v}_1 = \begin{pmatrix}1\\1\end{pmatrix}$. For $\lambda_2 = -1$: solve $(A+I)\mathbf{v}=0$, i.e.

Take $\mathbf{v}_2 = \begin{pmatrix}-2\\1\end{pmatrix}$. Step 3: General solution. By the theorem,

Step 4: Apply the initial condition. At $t=0$:

Subtracting the equations: $-3c_2 = 4$, so $c_2 = -\tfrac{4}{3}$, and then $c_1 = 1 - c_2 = 1 + \tfrac43 = \tfrac73$. Step 5: Final answer.

Check at $t=0$: $\frac73\begin{pmatrix}1\\1\end{pmatrix} - \frac43\begin{pmatrix}-2\\1\end{pmatrix} = \begin{pmatrix}\frac73+\frac83\\\frac73-\frac43\end{pmatrix} = \begin{pmatrix}5\\1\end{pmatrix}$, matching $\mathbf{x}(0)$. Since $\lambda_1 = 5 > 0$ dominates as $t\to\infty$, almost every trajectory is eventually swept toward the direction of $\mathbf{v}_1$: an unstable node.

Example 2 — a $3\times 3$ system with an initial condition off any eigenline. Solve $\mathbf{x}' = A\mathbf{x}$ with

Step 1: Eigenvalues. Since the first row is $(3,0,0)$, expand the determinant along the first row:

The $2\times 2$ block is upper triangular, so its determinant is $(1-\lambda)(2-\lambda)$. Hence

giving eigenvalues $\lambda_1 = 3$, $\lambda_2 = 1$, $\lambda_3 = 2$: three distinct real values. Step 2: Eigenvectors.

For $\lambda_1 = 3$: solve $(A-3I)\mathbf{v}=\mathbf 0$:

Row 3 gives $v_1 = -v_3$. Row 2: $v_1 - 2v_2+2v_3 = 0 \Rightarrow -v_3 - 2v_2+2v_3=0 \Rightarrow v_3 = 2v_2$. Let $v_2=1$: $v_3=2$, $v_1=-2$. So $\mathbf{v}_1 = \begin{pmatrix}-2\\1\\2\end{pmatrix}$. Check: $A\mathbf{v}_1 = \begin{pmatrix}3(-2)\\-2+1+4\\2+0+4\end{pmatrix}=\begin{pmatrix}-6\\3\\6\end{pmatrix} = 3\begin{pmatrix}-2\\1\\2\end{pmatrix}$. ✓.

For $\lambda_2 = 1$: solve $(A-I)\mathbf{v}=\mathbf 0$:

Row 1: $2v_1=0\Rightarrow v_1=0$. Row 2: $v_1+2v_3=0 \Rightarrow v_3=0$. $v_2$ is free. Take $\mathbf{v}_2 = \begin{pmatrix}0\\1\\0\end{pmatrix}$. Check: $A\mathbf{v}_2 = \begin{pmatrix}0\\1\\0\end{pmatrix} = 1\cdot\mathbf v_2$. ✓.

For $\lambda_3 = 2$: solve $(A-2I)\mathbf{v}=\mathbf 0$:

Row 1: $v_1=0$. Row 2: $-v_2+2v_3=0 \Rightarrow v_2=2v_3$. Let $v_3=1$: $v_2=2$. So $\mathbf{v}_3 = \begin{pmatrix}0\\2\\1\end{pmatrix}$. Check: $A\mathbf{v}_3=\begin{pmatrix}0\\0+2+2\\0+0+2\end{pmatrix}=\begin{pmatrix}0\\4\\2\end{pmatrix} = 2\begin{pmatrix}0\\2\\1\end{pmatrix}$. ✓. Step 3: General solution.

Step 4: Solve for constants using $\mathbf{x}(0) = (2,0,3)^T$.

From the first equation, $c_1 = -1$. Substitute into the third: $-2 + c_3 = 3 \Rightarrow c_3 = 5$. Substitute $c_1=-1, c_3=5$ into the second: $-1 + c_2 + 10 = 0 \Rightarrow c_2 = -9$. Step 5: Final answer.

Verify at $t=0$: first component $2 + 0 + 0 = 2$; second component $-1-9+10=0$; third component $-2+0+5=3$. Matches $\mathbf{x}(0)$. As $t\to\infty$, the $e^{3t}$ term dominates, so the trajectory asymptotically aligns with $\mathbf{v}_1$.