Differential Equations › Section E · Systems of differential equations › Topic E.5
Real eigenvalues
When the matrix of a system has two different real eigenvalues, the general solution is just two exponentials, each riding along its own eigenvector. This topic solves such systems, fits initial conditions, and turns the formulas into phase portraits of saddles and nodes.
Two eigenvalues, two solutions
Suppose the 2 × 2 matrix \(A\) has real eigenvalues \(\lambda_1 \ne \lambda_2\) with eigenvectors \(\mathbf{v}_1\) and \(\mathbf{v}_2\). "Solutions to systems" showed that \(e^{\lambda_1 t}\mathbf{v}_1\) and \(e^{\lambda_2 t}\mathbf{v}_2\) are solutions. Eigenvectors for different eigenvalues are independent, so at \(t = 0\) the Wronskian \(\det[\mathbf{v}_1 \; \mathbf{v}_2]\) is non-zero and the two form a fundamental set. The general solution is therefore
The same holds for an \(n \times n\) matrix with \(n\) distinct real eigenvalues: \(n\) such terms.
The method
1. Find the eigenvalues from \(\lambda^2 - (\operatorname{tr} A)\lambda + \det A = 0\). 2. For each eigenvalue, solve \((A - \lambda I)\mathbf{v} = \mathbf{0}\) for an eigenvector. 3. Write the general solution as above. 4. For an initial value problem with \(\mathbf{x}(0) = \mathbf{x}_0\), set \(t = 0\): every exponential equals 1, leaving \(c_1\mathbf{v}_1 + c_2\mathbf{v}_2 = \mathbf{x}_0\), a 2 × 2 linear system for the constants. 5. Check by substituting \(t = 0\) and, if time allows, by differentiating one component.
Worked example: a saddle
Solve \(\mathbf{x}' = A\mathbf{x}\), \(\mathbf{x}(0) = (1, -4)\), for
1. Trace 1 and determinant \(-6 + 4 = -2\) give \(\lambda^2 - \lambda - 2 = (\lambda - 2)(\lambda + 1) = 0\). 2. For 2: \(A - 2I\) has rows (1, -2) and (2, -4), so \(\mathbf{v}_1 = (2, 1)\). For -1: \(A + I\) has rows (4, -2) and (2, -1), so \(\mathbf{v}_2 = (1, 2)\). 3. Solve \(2c_1 + c_2 = 1\) and \(c_1 + 2c_2 = -4\): \(c_1 = 2\) and \(c_2 = -3\). 4. So \(x_1 = 4e^{2t} - 3e^{-t}\) and \(x_2 = 2e^{2t} - 6e^{-t}\). 5. Check: \(x_1' = 8e^{2t} + 3e^{-t}\), and \(3x_1 - 2x_2 = 8e^{2t} + 3e^{-t}\) too.
The saddle's phase portrait
Each eigenvector line carries straight-line trajectories. With \(c_2 = 0\), the solution \(c_1e^{2t}(2, 1)\) moves outwards along the line through (2, 1). With \(c_1 = 0\), the solution \(c_2e^{-t}(1, 2)\) moves inwards along the line through (1, 2). Any other trajectory mixes the two. As \(t \to -\infty\) the \(e^{-t}\) term dominates, so the trajectory arrives roughly parallel to (1, 2); as \(t \to \infty\) the \(e^{2t}\) term dominates, so it leaves roughly parallel to (2, 1). The example's solution, with \(c_1 = 2 > 0\), ends up heading out along the direction of (2, 1). The origin is an unstable saddle.
Worked example: salt in two tanks
In the two-tank model of "Systems of differential equations", \(x_1' = -0.09x_1 + 0.04x_2\) and \(x_2' = 0.09x_1 - 0.09x_2\), in kilograms and minutes. Tank 1 starts with 20 kg of salt and tank 2 with none. 1. Trace -0.18 and determinant \(0.0081 - 0.0036 = 0.0045\) give \((\lambda + 0.03)(\lambda + 0.15) = 0\). 2. For -0.03, the first row of \(A + 0.03I\) is (-0.06, 0.04), giving \(\mathbf{v}_1 = (2, 3)\). For -0.15, the row (0.06, 0.04) gives \(\mathbf{v}_2 = (2, -3)\). 3. Solve \(2c_1 + 2c_2 = 20\) and \(3c_1 - 3c_2 = 0\): \(c_1 = c_2 = 5\). 4. So \(x_1 = 10e^{-0.03t} + 10e^{-0.15t}\) and \(x_2 = 15e^{-0.03t} - 15e^{-0.15t}\).
Reading the tank solution
Both eigenvalues are negative, so all the salt eventually washes out and the origin is an asymptotically stable node. The two terms have clear physical meanings. The fast term, along (2, -3), describes the tanks evening out: after 20 minutes its factor \(e^{-3}\) is about 0.05. The slow term, along (2, 3), describes the evened-out salt draining away, with the amounts settling into the ratio 2 : 3. Tank 2 starts empty, fills and then empties: \(x_2' = 0\) when \(e^{0.12t} = 5\), at \(t = \frac{\ln 5}{0.12} \approx 13.4\) minutes. At that moment both tanks hold about 8.0 kg, as they must, since \(x_2' = 0.09(x_1 - x_2)\).
The node's phase portrait
For a node, both eigenvector lines carry straight-line trajectories, all moving in when the eigenvalues are negative and all moving out when they are positive. Other trajectories curve. Close to the origin they run tangent to the eigenvector whose eigenvalue is smaller in absolute value, because that term changes more slowly and is the last to fade (or the first to appear, for an unstable node). Far from the origin they run nearly parallel to the other eigenvector. In the tank example, trajectories settle onto the origin tangent to the line through (2, 3), and the slow eigenvalue -0.03 is the long-run decay rate.
Larger systems and component form
Nothing in the method depends on the size. For an \(n \times n\) matrix with \(n\) distinct real eigenvalues, find all \(n\) eigenpairs, write \(n\) exponential terms, and solve the \(n \times n\) linear system \(c_1\mathbf{v}_1 + \dots + c_n\mathbf{v}_n = \mathbf{x}_0\). Triangular matrices are the easy case, since their eigenvalues are the diagonal entries. When a problem is stated without matrices, as a list of equations for \(x_1'\) and \(x_2'\), convert it to matrix form, solve, then multiply out the vectors to report each component separately, as in both worked examples.
Common mistakes to avoid
First, pairing an exponential with the wrong eigenvector: \(e^{\lambda_1 t}\) must multiply the eigenvector for \(\lambda_1\). Second, taking the constants straight from \(\mathbf{x}_0\); they solve \(c_1\mathbf{v}_1 + c_2\mathbf{v}_2 = \mathbf{x}_0\). Third, calling a saddle stable because one line of trajectories heads in. Fourth, choosing the wrong tangent direction for a node: near the origin, trajectories follow the eigenvector whose eigenvalue is smaller in absolute value, not the larger. Fifth, using the discrete-time test \(|\lambda| < 1\): here decay needs \(\lambda < 0\), so an eigenvalue of 0.5 means growth.
Key takeaways
Distinct real eigenvalues give the general solution \(c_1e^{\lambda_1 t}\mathbf{v}_1 + c_2e^{\lambda_2 t}\mathbf{v}_2\), with constants from \(c_1\mathbf{v}_1 + c_2\mathbf{v}_2 = \mathbf{x}_0\). Each eigenvector line holds straight-line trajectories, inwards for negative eigenvalues and outwards for positive ones. Opposite signs give an unstable saddle; equal signs give a node, stable if both are negative, with nearby trajectories tangent to the eigenvector of the eigenvalue nearer zero. In applications, each term is a mode with its own rate.
Flashcards
Flashcard
x = c₁e^(λ₁t)v₁ + c₂e^(λ₂t)v₂
Test your understanding
Question 1 of 5
A has eigenvalue 4 with eigenvector (1, 2) and eigenvalue −1 with eigenvector (1, −3). What is the general solution of x′ = Ax?
Question 2 of 5
With eigenvector (1, 1) for λ = 2 and (1, −1) for λ = −2, which solution of x′ = Ax satisfies x(0) = (4, 2)?
Question 3 of 5
A node has eigenvalues −1 and −6. Near the origin, trajectories that are not on an eigenvector line run tangent to which direction?
Question 4 of 5
The solution x = 2e^(3t)(1, 1) − e^(−2t)(1, 4) is followed for large positive t. Its direction approaches which vector?
Question 5 of 5
In the two-tank model, x₂(t) = 15e^(−0.03t) − 15e^(−0.15t) kg. What happens to the salt in tank 2?
