Differential Equations › Section E · Systems of differential equations › Topic E.4
The phase plane
A solution of a two-unknown system is a pair of functions, but it can also be drawn as one curve traced out in the plane. Collect enough of those curves and a single picture shows whether every solution settles down, runs away or circles forever, and the eigenvalues of the matrix predict that picture.
Solutions as moving points
For a 2 × 2 system \(\mathbf{x}' = A\mathbf{x}\), a solution gives a point \((x_1(t), x_2(t))\) for each time \(t\). As \(t\) increases, the point moves and traces a curve in the \(x_1x_2\)-plane, called the phase plane. The curve is the solution's trajectory, and an arrow on it shows the direction of increasing \(t\). Time itself does not appear on the axes; it is a hidden parameter, just as in a parametric curve. A drawing of several representative trajectories is a phase portrait, and it shows the behaviour of every solution at a glance.
The equilibrium at the origin
An equilibrium solution is a constant solution, so \(\mathbf{x}' = \mathbf{0}\) and therefore \(A\mathbf{x} = \mathbf{0}\). The origin always qualifies. If \(\det A \ne 0\), it is the only equilibrium, and this topic assumes so. The key question extends the one asked in "Equilibrium solutions and stability": as \(t\) increases, do trajectories that start near the origin move towards it, away from it, or round it? The answer is a property of the matrix, so it can be read off without solving anything.
Sketching from tangent vectors
Pick a point \(\mathbf{x}\) in the phase plane and compute \(A\mathbf{x}\). That vector is \(\mathbf{x}'\) for the trajectory passing through the point, so it is tangent to the trajectory and points the way the solution moves. Drawing these vectors on a grid of points gives a picture like the direction fields of "Direction fields", and trajectories can then be sketched by following the arrows. The method needs no solving at all, but it is slow by hand. It is most useful for spot checks, such as finding which way solutions turn.
Worked example: reading a saddle from arrows
Sketch trajectories of \(\mathbf{x}' = A\mathbf{x}\) for
1. At (1, 0) the tangent vector is (1, 2), and at (0, 1) it is (2, -2). 2. At (2, 1) it is (4, 2), which is 2 times (2, 1): the arrow points straight outwards along the line through (2, 1), so that line is itself a trajectory leaving the origin. 3. At (1, -2) it is (-3, 6), which is -3 times (1, -2): the arrow points straight back towards the origin along that line. 4. Other trajectories come in roughly along the second line and leave along the first. The origin is a saddle point, and the eigenvalues 2 and -3 confirm it.
Words for stability
The origin is asymptotically stable if every trajectory that starts near it approaches it as \(t \to \infty\). It is stable if trajectories that start near it stay near it, even if they never arrive. It is unstable if some trajectories starting arbitrarily close to it move away. A saddle is unstable even though two trajectories head straight in, because almost every other trajectory eventually leaves. The difference between stable and asymptotically stable matters for centres, met below: their trajectories circle forever without approaching.
Real eigenvalues: nodes and saddles
When the eigenvalues are real and distinct, each eigenvector line holds straight-line trajectories, moving inwards if its eigenvalue is negative and outwards if positive. If both eigenvalues are negative, every trajectory approaches the origin, an asymptotically stable node (the linear algebra course calls it a sink, or attractor). If both are positive, every non-zero trajectory leaves, an unstable node (a source, or repeller). If they have opposite signs, the origin is a saddle, which is unstable. Near a node, trajectories run tangent to the eigenvector whose eigenvalue is smaller in absolute value, because that term decays or grows more slowly. "Real eigenvalues" builds these portraits from the general solution.
Complex eigenvalues: spirals and centres
When the eigenvalues are \(\alpha \pm \beta i\) with \(\beta \ne 0\), trajectories rotate about the origin. If \(\alpha < 0\) they spiral inwards, an asymptotically stable spiral point. If \(\alpha > 0\) they spiral outwards, an unstable spiral point. If \(\alpha = 0\) they are closed curves, ellipses, around a centre, which is stable but not asymptotically stable. To find the direction of rotation, compute \(A\mathbf{x}\) at (1, 0): if the second entry is positive, trajectories cross the positive \(x_1\)-axis moving upwards, so the rotation is anticlockwise; if negative, clockwise. "Complex eigenvalues" develops these cases.
The trace-determinant test
Write \(\tau = \operatorname{tr} A\) and \(\Delta = \det A\). The eigenvalues solve
so they add to \(\tau\) and multiply to \(\Delta\). If \(\Delta < 0\), they are real with opposite signs: a saddle. If \(\Delta > 0\) and \(\tau^2 > 4\Delta\), they are real with the same sign as \(\tau\): a node, stable when \(\tau < 0\). If \(\tau^2 < 4\Delta\), they are complex with real part \(\tau/2\): a spiral, stable when \(\tau < 0\), or a centre when \(\tau = 0\). The borderline \(\tau^2 = 4\Delta\) gives a repeated eigenvalue, treated in "Repeated eigenvalues".
Worked example: three quick classifications
Classify the origin for each matrix, given by its rows. 1. Rows (-3, 2) and (1, -4): \(\tau = -7\), \(\Delta = 12 - 2 = 10\), and \(\tau^2 - 4\Delta = 9 > 0\), so an asymptotically stable node; the eigenvalues are -2 and -5. 2. Rows (2, -5) and (1, -2): \(\tau = 0\) and \(\Delta = -4 + 5 = 1\), so a centre with eigenvalues \(\pm i\). At (1, 0) the tangent vector is (2, 1), pointing upwards, so the rotation is anticlockwise. 3. Rows (1, -4) and (2, 3): \(\tau = 4\), \(\Delta = 3 + 8 = 11\), and \(\tau^2 - 4\Delta = -28 < 0\), so an unstable spiral point, with eigenvalues \(2 \pm i\sqrt{7}\).
Stability in any dimension
The same idea works for \(n \times n\) systems with \(\det A \ne 0\). The origin is asymptotically stable exactly when every eigenvalue of \(A\) has a negative real part, and it is unstable if any eigenvalue has a positive real part. Control engineers use this constantly: a feedback design is chosen so that every eigenvalue of the resulting system matrix lies safely to the left of the imaginary axis, which makes disturbances die away. "Nonlinear systems: equilibria and linearization" applies the same test to equilibria of nonlinear models.
Common mistakes to avoid
First, calling a saddle stable because some trajectories approach it; almost all of them leave. Second, deciding spiral stability from \(\beta\) instead of from the real part \(\alpha\). Third, calling a centre asymptotically stable: its trajectories never get closer to the origin. Fourth, testing only \(\Delta > 0\) and forgetting \(\tau^2 - 4\Delta\), which separates nodes from spirals. Fifth, reading the rotation direction from the first entry of \(A\mathbf{x}\) at (1, 0); on the positive \(x_1\)-axis it is the second entry that says up or down.
Key takeaways
In the phase plane each solution of a 2 × 2 system is a trajectory, and the tangent vector at \(\mathbf{x}\) is \(A\mathbf{x}\). Real eigenvalues of one sign give nodes, opposite signs give saddles, and complex eigenvalues give spirals, or centres when the real part is 0. With \(\tau = \operatorname{tr} A\) and \(\Delta = \det A\): \(\Delta < 0\) means saddle, the sign of \(\tau^2 - 4\Delta\) separates nodes from spirals, and the sign of \(\tau\) decides stability. Asymptotic stability needs every eigenvalue to have a negative real part.
Flashcards
Flashcard
The x₁x₂-plane in which solutions of a 2 × 2 system are drawn
Test your understanding
Question 1 of 5
A 2 × 2 system x′ = Ax has trace 0 and determinant 9. What is the origin?
Question 2 of 5
At the point (1, 0), the system x′ = Ax gives the tangent vector (−2, −3). Which way is the trajectory through (1, 0) moving?
Question 3 of 5
The matrix A has eigenvalues −4 and 1. How do trajectories of x′ = Ax behave near the origin?
Question 4 of 5
For which matrix, given by its rows, is the origin an asymptotically stable spiral point?
Question 5 of 5
A 2 × 2 matrix has trace −6 and determinant 5. What is the origin of x′ = Ax?
