Differential Equations › Section E · Systems of differential equations › Topic E.3
Solutions to systems
A clever guess turns the homogeneous system x′ = Ax into an eigenvalue problem. This topic makes that guess, shows when a handful of solutions is enough to build every solution, and uses the Wronskian to check.
A guess borrowed from one equation
When there is a single unknown, \(x' = ax\) has the solutions \(x = ce^{at}\). For the system \(\mathbf{x}' = A\mathbf{x}\), with \(A\) a constant \(n \times n\) matrix, try the same shape with a constant vector in place of \(c\):
The left side becomes \(\mathbf{x}' = \lambda e^{\lambda t}\mathbf{v}\) and the right side \(A\mathbf{x} = e^{\lambda t}A\mathbf{v}\). Because \(e^{\lambda t}\) is never zero, they agree for all \(t\) exactly when \(A\mathbf{v} = \lambda\mathbf{v}\). So every eigenvalue \(\lambda\) of \(A\) with an eigenvector \(\mathbf{v}\) produces a solution, and finding solutions starts with the eigen-work reviewed in "Review: matrices, vectors, and eigenvalues".
Worked example: two exponential solutions
Find two solutions of \(\mathbf{x}' = A\mathbf{x}\) for
1. The trace is -1 and the determinant is \(-2 - 4 = -6\), so \(\lambda^2 + \lambda - 6 = (\lambda - 2)(\lambda + 3) = 0\). 2. For \(\lambda = 2\), \(A - 2I\) has rows (-4, 1) and (4, -1), so \(v_2 = 4v_1\) and \(\mathbf{v} = (1, 4)\). 3. For \(\lambda = -3\), \(A + 3I\) has rows (1, 1) and (4, 4), so \(\mathbf{v} = (1, -1)\). 4. The solutions are \(\mathbf{x}_1 = e^{2t}(1, 4)\) and \(\mathbf{x}_2 = e^{-3t}(1, -1)\), with vectors read as columns.
Checking a solution directly
Substitution confirms a claimed solution without redoing any eigen-work. For \(\mathbf{x}_1 = e^{2t}(1, 4)\), the derivative is \(e^{2t}(2, 8)\), and \(A(1, 4) = (-2 + 4, 4 + 4) = (2, 8)\), so \(A\mathbf{x}_1 = e^{2t}(2, 8)\) as well. The same test exposes a wrong guess. Try \(e^{2t}(1, 1)\): its derivative is \(e^{2t}(2, 2)\), but \(A(1, 1) = (-1, 5)\). Since (1, 1) is not an eigenvector, the two sides differ and this is not a solution. Make the check a habit; it catches most arithmetic slips in a few seconds.
Superposition
If \(\mathbf{x}_1\) and \(\mathbf{x}_2\) solve a homogeneous system \(\mathbf{x}' = A(t)\mathbf{x}\), so does every combination \(c_1\mathbf{x}_1 + c_2\mathbf{x}_2\). The proof is short:
This is the same principle as in "Basic concepts and superposition" for second-order equations, and it relies on the system being both linear and homogeneous. For a nonhomogeneous system the sum of two solutions carries the forcing term twice, so it is no longer a solution.
Fundamental sets and the Wronskian
For an \(n \times n\) system, take \(n\) solutions and place them as the columns of a matrix \(X(t)\). Their Wronskian is \(W(t) = \det X(t)\). If \(W(t) \ne 0\), the solutions are linearly independent and form a fundamental set, and then every solution of the system is
the general solution. A useful fact: for solutions of the same system on an interval where \(A(t)\) is continuous, \(W\) is either zero at every point or at none, so testing one convenient value of \(t\) is enough. For constant \(A\) there is even a formula, \(W(t) = W(0)e^{(\operatorname{tr} A)t}\).
Worked example: a fundamental set and an initial value
Use the two solutions found above to solve \(\mathbf{x}' = A\mathbf{x}\), \(\mathbf{x}(0) = (3, 2)\). 1. The solution matrix is
2. \(W(t) = -e^{-t} - 4e^{-t} = -5e^{-t}\), never zero, so this is a fundamental set. It agrees with \(W(0)e^{(\operatorname{tr} A)t}\), since \(\operatorname{tr} A = -1\). 3. At \(t = 0\): \(c_1 + c_2 = 3\) and \(4c_1 - c_2 = 2\), so \(c_1 = 1\) and \(c_2 = 2\). 4. Hence \(x_1 = e^{2t} + 2e^{-3t}\) and \(x_2 = 4e^{2t} - 2e^{-3t}\). 5. Check: at \(t = 0\) these give 3 and 2.
Why a non-zero Wronskian matters
Matching an initial condition \(\mathbf{x}(t_0) = \mathbf{x}_0\) means solving the linear system \(X(t_0)\mathbf{c} = \mathbf{x}_0\) for the constants. By the independence and inverse facts from the review topic, this has exactly one solution for every \(\mathbf{x}_0\) precisely when \(\det X(t_0) = W(t_0) \ne 0\). So a fundamental set is exactly a collection of solutions flexible enough to fit any starting point. If \(W = 0\), one solution is a combination of the others, and some initial conditions cannot be matched.
The link to second-order equations
Convert \(y'' + y' - 6y = 0\) with \(x_1 = y\), \(x_2 = y'\): the matrix has rows (0, 1) and (6, -1). Its characteristic equation, \(\lambda^2 + \lambda - 6 = 0\), is exactly the characteristic equation \(r^2 + r - 6 = 0\) of "Real, distinct roots", so the eigenvalues are 2 and -3. For each \(\lambda\), the first row of \(A - \lambda I\) is \((-\lambda, 1)\), so an eigenvector is \((1, \lambda)\). The solution \(e^{\lambda t}(1, \lambda)\) therefore lists \(y = e^{\lambda t}\) and its derivative, and the matrix Wronskian equals the Wronskian of "Fundamental sets and the Wronskian".
Three cases ahead
For a 2 × 2 system with real entries, the characteristic equation is a quadratic, so its roots are real and distinct, a complex conjugate pair, or a repeated real root. Each case gets its own topic. In "Real eigenvalues", two independent eigenvectors give a fundamental set at once. In "Complex eigenvalues", one complex solution is split into two real ones. In "Repeated eigenvalues", there may be only one independent eigenvector, so \(e^{\lambda t}\mathbf{v}\) supplies too few solutions and a second solution of a new form is needed.
Common mistakes to avoid
First, using the guess \(e^{\lambda t}\mathbf{v}\) when \(A\) depends on \(t\); the eigenvalue argument needs a constant matrix. Second, reading the constants straight off \(\mathbf{x}_0\): they come from solving \(X(t_0)\mathbf{c} = \mathbf{x}_0\). Third, concluding that solutions are dependent because \(W(t) \to 0\) as \(t\) grows; only \(W = 0\) at some point signals dependence. Fourth, combining solutions of a nonhomogeneous system as if superposition applied. Fifth, forgetting that each solution is a vector function, so the columns of \(X(t)\) are whole solutions, not single components.
Key takeaways
For constant \(A\), \(e^{\lambda t}\mathbf{v}\) solves \(\mathbf{x}' = A\mathbf{x}\) exactly when \(\lambda\) and \(\mathbf{v}\) are an eigenpair of \(A\). Homogeneous linear systems obey superposition. \(n\) solutions with non-zero Wronskian \(W = \det X\) form a fundamental set, every solution is a combination of them, and the constants come from \(X(t_0)\mathbf{c} = \mathbf{x}_0\). The 2 × 2 case splits into real, complex and repeated eigenvalues.
Flashcards
Flashcard
The trial solution for x′ = Ax with constant A
Test your understanding
Question 1 of 5
For which choice of λ and v is x(t) = e^(λt) v a solution of x′ = Ax, where A is a constant matrix?
Question 2 of 5
Which of these is a solution of x′ = Ax, where A has rows (3, 2) and (2, 3)?
Question 3 of 5
Two solutions of a 2 × 2 system with constant A have Wronskian W(t) = 6e^(−2t). What can you conclude?
Question 4 of 5
Using x₁ = e^(2t)(1, 4) and x₂ = e^(−3t)(1, −1), which constants give x(0) = (2, 3)?
Question 5 of 5
y″ − 4y′ + 3y = 0 is written as x′ = Ax with x₁ = y and x₂ = y′. What are the eigenvalues of A?
