5.17 · AHL · HL onlyPaper 1Paper 2Paper 3

Phase portraits for coupled equations

In a minute

A shows where a pair of linear takes every starting point. The of the system's set the picture: paths in or out, spirals, ellipses or a . You use it to predict long-term behaviour, such as whether a population settles or dies out.

You'll be able to

  • Write a linear system in matrix form, find velocities and gradients, and find its .
  • Find the exact solution when the eigenvalues are real and distinct.
  • Decide from the eigenvalues whether solutions move in or out, spiral, form ellipses, or meet a saddle point.
  • Sketch phase portraits with lines, arrows and a trajectory in every region.
  • Interpret a phase portrait in context: stable populations, which species dies out, and the long-term behaviour.
Before you start3 quick questions on what this builds on
Check yourself
Find the eigenvalues of (4123)\begin{pmatrix}4 & 1\\ 2 & 3\end{pmatrix}.
Check yourself
Solve λ2+2λ+5=0\lambda^{2} + 2\lambda + 5 = 0.
Check yourself
dxdt=3x−y\frac{\mathrm{d}x}{\mathrm{d}t} = 3x - y, dydt=x+2y\frac{\mathrm{d}y}{\mathrm{d}t} = x + 2y. Find dydt\frac{\mathrm{d}y}{\mathrm{d}t} when x=2x = 2 and y=5y = 5.

Revise first: 1.12 Complex numbers in Cartesian form, 1.14 Matrices, determinants and inverses, 1.15 Eigenvalues and eigenvectors, 3.10 Vectors in 2D and 3D, 5.14 Setting up and separating differential equations, 5.15 Slope fields and 5.16 Euler's method

Linear systems and the phase plane

In 5.16 you stepped coupled differential equations forward with Euler's method. Here every rate is a fixed mix of xx and yy, so the system is linear. Time is tt, and aa, bb, cc, dd are constants:

Learn this
(dxdtdydt)=(abcd)(xy)\begin{pmatrix}\dfrac{\mathrm{d}x}{\mathrm{d}t}\\[6pt] \dfrac{\mathrm{d}y}{\mathrm{d}t}\end{pmatrix} = \begin{pmatrix}a & b\\ c & d\end{pmatrix}\begin{pmatrix}x\\ y\end{pmatrix}

This is the pair dxdt=ax+by\frac{\mathrm{d}x}{\mathrm{d}t} = ax + by and dydt=cx+dy\frac{\mathrm{d}y}{\mathrm{d}t} = cx + dy written with a matrix. Call the matrix MM. Its eigenvalues and eigenvectors decide how every solution behaves.

At any point, the two equations give the velocity (dx/dtdy/dt)\begin{pmatrix}\mathrm{d}x/\mathrm{d}t\\ \mathrm{d}y/\mathrm{d}t\end{pmatrix}. It is tangent to the trajectory and points the way it moves. Its gradient is the gradient of the trajectory:

Learn this
dydx=dy/dtdx/dt\frac{\mathrm{d}y}{\mathrm{d}x} = \frac{\mathrm{d}y/\mathrm{d}t}{\mathrm{d}x/\mathrm{d}t}
Worked exampleFor the system dxdt=2x−3y\frac{\mathrm{d}x}{\mathrm{d}t} = 2x - 3y, dydt=x+4y\frac{\mathrm{d}y}{\mathrm{d}t} = x + 4y, find the velocity vector at (1,2)(1, 2). Describe which way the trajectory through (1,2)(1, 2) is moving.

Try it first, then reveal one step at a time.

  1. Substitute x=1x = 1, y=2y = 2: dxdt=2(1)−3(2)=−4\frac{\mathrm{d}x}{\mathrm{d}t} = 2(1) - 3(2) = -4.
Worked exampleThe heating in two connected rooms, A and B, is switched off. Let xx and yy be how many ∘C^{\circ}\text{C} rooms A and B are above the outside temperature, tt hours later. A model is dxdt=−0.3x+0.1y\frac{\mathrm{d}x}{\mathrm{d}t} = -0.3x + 0.1y, dydt=0.2x−0.4y\frac{\mathrm{d}y}{\mathrm{d}t} = 0.2x - 0.4y. At one moment x=10x = 10 and y=4y = 4. Find dxdt\frac{\mathrm{d}x}{\mathrm{d}t} and dydt\frac{\mathrm{d}y}{\mathrm{d}t}, and interpret them. Then find the gradient of the trajectory at (10,4)(10, 4).

Try it first, then reveal one step at a time.

  1. dxdt=−0.3(10)+0.1(4)=−2.6\frac{\mathrm{d}x}{\mathrm{d}t} = -0.3(10) + 0.1(4) = -2.6.
Check yourself
Which point is an equilibrium point of dxdt=5x−2y\frac{\mathrm{d}x}{\mathrm{d}t} = 5x - 2y, dydt=3x+y\frac{\mathrm{d}y}{\mathrm{d}t} = 3x + y?
Check yourself
For the two-room model, find the gradient of the trajectory at (6,9)(6, 9). Give your answer to 3 s.f.

Real eigenvalues: straight lines and exact solutions

Start a solution on the line through the origin in the direction of an eigenvector p\boldsymbol{p}, with eigenvalue λ\lambda. Then Mp=λpM\boldsymbol{p} = \lambda\boldsymbol{p}, so the velocity points along that same line. The solution never leaves the line.

If λ<0\lambda < 0, it moves along the line towards the origin. If λ>0\lambda > 0, it moves away. So each eigenvector line is made of straight-line trajectories.

Every other solution is a mix of the two straight-line motions. When the eigenvalues λ1\lambda_1 and λ2\lambda_2 are real and distinct, with eigenvectors p1\boldsymbol{p}_1 and p2\boldsymbol{p}_2, the exact solution is:

In the formula booklet
x=Aeλ1tp1+Beλ2tp2\boldsymbol{x} = A\mathrm{e}^{\lambda_{1}t}\boldsymbol{p}_{1} + B\mathrm{e}^{\lambda_{2}t}\boldsymbol{p}_{2}

Here x=(xy)\boldsymbol{x} = \begin{pmatrix}x\\ y\end{pmatrix}, and AA and BB are constants. At t=0t = 0 both exponentials are 11, so the starting point gives Ap1+Bp2A\boldsymbol{p}_1 + B\boldsymbol{p}_2. This formula is in the booklet. You only need exact solutions when the eigenvalues are real and distinct.

To sketch the portrait when both eigenvalues have the same sign:

1. Draw both eigenvector lines through the origin.

2. Put arrows on them: towards the origin for a negative eigenvalue, away for a positive one.

3. Draw curved trajectories between the lines. Near the origin they run alongside the line whose eigenvalue is closer to 00, because that term changes most slowly. With both eigenvalues positive, far from the origin they turn parallel to the other line.

Phase portrait with every trajectory moving into the origin. The line y equals x and the line y equals minus 2x are drawn through the origin, with arrows pointing inwards on both. Curved trajectories start near the edges, bend towards the line y equals x and enter the origin along it. x y y = x y = −2x λ = −0.2 on y = x, λ = −0.5 on y = −2x
The two-room model: eigenvalues −0.2-0.2 on y=xy = x and −0.5-0.5 on y=−2xy = -2x. Curves enter the origin along y=xy = x, the line of the eigenvalue closer to 00.
Worked exampleFor the two-room model dxdt=−0.3x+0.1y\frac{\mathrm{d}x}{\mathrm{d}t} = -0.3x + 0.1y, dydt=0.2x−0.4y\frac{\mathrm{d}y}{\mathrm{d}t} = 0.2x - 0.4y, room A starts 12 ∘C12\,^{\circ}\text{C} and room B 3 ∘C3\,^{\circ}\text{C} above the outside temperature. Find the eigenvalues and eigenvectors, write down the , and find AA and BB. Describe what happens in the long term.

Try it first, then reveal one step at a time.

  1. Characteristic equation: (−0.3−λ)(−0.4−λ)−(0.1)(0.2)=λ2+0.7λ+0.1=0(-0.3 - \lambda)(-0.4 - \lambda) - (0.1)(0.2) = \lambda^{2} + 0.7\lambda + 0.1 = 0, so λ=−0.2\lambda = -0.2 or λ=−0.5\lambda = -0.5.
Worked exampleA weed spreads in two neighbouring fields. Let xx and yy be the areas it covers, in m2\text{m}^{2}, in each field after tt months. A model is dxdt=0.3x+0.1y\frac{\mathrm{d}x}{\mathrm{d}t} = 0.3x + 0.1y, dydt=0.1x+0.3y\frac{\mathrm{d}y}{\mathrm{d}t} = 0.1x + 0.3y. Use the eigenvalues to describe the long-term behaviour, and comment on the model.

Try it first, then reveal one step at a time.

  1. (0.3−λ)2−0.01=0(0.3 - \lambda)^{2} - 0.01 = 0, so 0.3−λ=±0.10.3 - \lambda = \pm 0.1 and λ=0.2\lambda = 0.2 or λ=0.4\lambda = 0.4.
Check yourself
Describe the long-term behaviour of dxdt=−2x+y\frac{\mathrm{d}x}{\mathrm{d}t} = -2x + y, dydt=x−2y\frac{\mathrm{d}y}{\mathrm{d}t} = x - 2y.
Check yourself
dxdt=−3x+y\frac{\mathrm{d}x}{\mathrm{d}t} = -3x + y, dydt=2x−2y\frac{\mathrm{d}y}{\mathrm{d}t} = 2x - 2y has eigenvalues −1-1 and −4-4, with eigenvectors (12)\begin{pmatrix}1\\ 2\end{pmatrix} and (1−1)\begin{pmatrix}1\\ -1\end{pmatrix}. The solution is x=Ae−t(12)+Be−4t(1−1)\boldsymbol{x} = A\mathrm{e}^{-t}\begin{pmatrix}1\\ 2\end{pmatrix} + B\mathrm{e}^{-4t}\begin{pmatrix}1\\ -1\end{pmatrix}. It starts at (5,4)(5, 4). Find AA.

Real eigenvalues of opposite signs: saddle points

Now let one eigenvalue be negative and the other positive. Solutions on the negative eigenvalue's line move in to the origin. Solutions on the positive one's line move out.

Any other solution has both terms. The negative eigenvalue's term dies away and the positive one takes over. So the trajectory comes in, turns near the origin and leaves. A saddle point is unstable: a tiny nudge off the incoming line and the solution leaves.

To sketch it, draw both eigenvector lines with arrows, then one trajectory in each of the four regions.

Saddle point at the origin. The line y equals x has arrows pointing away from the origin. The line y equals minus four thirds x has arrows pointing towards the origin. In each of the four regions between the lines, curved trajectories come in alongside the incoming line, turn near the origin and leave alongside the line y equals x. x y y = x y = −4x/3 λ = 5 on y = x (out), λ = −2 on y = −4x/3 (in)
Eigenvalues 55 on y=xy = x (out) and −2-2 on y=−43xy = -\frac{4}{3}x (in): a saddle point.
Worked exampleFor dxdt=2x+3y\frac{\mathrm{d}x}{\mathrm{d}t} = 2x + 3y, dydt=4x+y\frac{\mathrm{d}y}{\mathrm{d}t} = 4x + y, find the eigenvalues and eigenvectors. Use them to describe the phase portrait.

Try it first, then reveal one step at a time.

  1. det⁡(2−λ341−λ)=(2−λ)(1−λ)−12=λ2−3λ−10=0\det\begin{pmatrix}2 - \lambda & 3\\ 4 & 1 - \lambda\end{pmatrix} = (2 - \lambda)(1 - \lambda) - 12 = \lambda^{2} - 3\lambda - 10 = 0.

In a population model the phase portrait only uses x≥0x \ge 0 and y≥0y \ge 0. A trajectory that reaches an axis means one population has reached 00: that species dies out.

First-quadrant phase portrait for two competing species, x and y in thousands. The line y equals x is drawn with arrows pointing towards the origin. Trajectories starting above the line curve up and to the left and reach the y-axis, so species X dies out. Trajectories starting below the line curve down and to the right and reach the x-axis, so species Y dies out. y = x x y X dies out Y dies out
Two competing species: the incoming line y=xy = x splits the portrait into two outcomes.
Worked exampleTwo species of beetle, X and Y, compete for food on an island. Their populations, in thousands, after tt years are xx and yy, with dxdt=0.1x−0.3y\frac{\mathrm{d}x}{\mathrm{d}t} = 0.1x - 0.3y, dydt=−0.3x+0.1y\frac{\mathrm{d}y}{\mathrm{d}t} = -0.3x + 0.1y. At first x=4.5x = 4.5 and y=5y = 5. Use the phase portrait to decide which species dies out. What would happen if both started at 55 thousand?

Try it first, then reveal one step at a time.

  1. (0.1−λ)2−0.09=0(0.1 - \lambda)^{2} - 0.09 = 0, so 0.1−λ=±0.30.1 - \lambda = \pm 0.3: λ=−0.2\lambda = -0.2 or λ=0.4\lambda = 0.4. Opposite signs, so the origin is a saddle point.
Check yourself
Which describes the phase portrait of dxdt=x+2y\frac{\mathrm{d}x}{\mathrm{d}t} = x + 2y, dydt=4x−y\frac{\mathrm{d}y}{\mathrm{d}t} = 4x - y?
Check yourself
The system dxdt=x+y\frac{\mathrm{d}x}{\mathrm{d}t} = x + y, dydt=4x−2y\frac{\mathrm{d}y}{\mathrm{d}t} = 4x - 2y has a saddle point at the origin. Trajectories come in alongside the line y=kxy = kx. Find kk.

Complex eigenvalues: spirals and ellipses

If the characteristic equation has a negative discriminant, the eigenvalues are complex: λ=α±βi\lambda = \alpha \pm \beta\mathrm{i} with β≠0\beta \neq 0. There are no real eigenvector lines, so no trajectory is straight. Every trajectory turns around the origin.

The eigenvalues don't tell you which way the paths turn. Substitute a point, such as (1,0)(1, 0). If dydt>0\frac{\mathrm{d}y}{\mathrm{d}t} > 0 there, the path crosses the positive xx-axis going up, so the motion is anticlockwise. If dydt<0\frac{\mathrm{d}y}{\mathrm{d}t} < 0, it is clockwise.

Two phase portraits side by side. Left, eigenvalues minus 0.2 plus or minus 0.5i: one trajectory starts on the positive x-axis and spirals anticlockwise into the origin. Right, eigenvalues plus or minus root 3 i: two closed ellipses centred on the origin, tilted, each traced anticlockwise. x y λ = −0.2 ± 0.5i anticlockwise spiral inwards x y λ = ±√3 i anticlockwise closed ellipses
Left: real part −0.2<0-0.2 < 0, a spiral inwards. Right: real part 00, closed ellipses.
Worked exampleIn a , xx and yy are how far, in hundreds, the numbers of rabbits and foxes are above their balance levels after tt years. The model is dxdt=−0.2x−0.5y\frac{\mathrm{d}x}{\mathrm{d}t} = -0.2x - 0.5y, dydt=0.5x−0.2y\frac{\mathrm{d}y}{\mathrm{d}t} = 0.5x - 0.2y. Find the eigenvalues, describe the trajectories, and interpret them.

Try it first, then reveal one step at a time.

  1. (−0.2−λ)2+0.25=0(-0.2 - \lambda)^{2} + 0.25 = 0, which is λ2+0.4λ+0.29=0\lambda^{2} + 0.4\lambda + 0.29 = 0.
Worked exampleFor dxdt=x−4y\frac{\mathrm{d}x}{\mathrm{d}t} = x - 4y, dydt=x−y\frac{\mathrm{d}y}{\mathrm{d}t} = x - y, show that the eigenvalues are imaginary. Describe the trajectory through (2,0)(2, 0) and find its gradient there.

Try it first, then reveal one step at a time.

  1. (1−λ)(−1−λ)+4=λ2+3=0(1 - \lambda)(-1 - \lambda) + 4 = \lambda^{2} + 3 = 0, so λ=±3 i\lambda = \pm\sqrt{3}\,\mathrm{i}. The real part is 00: the eigenvalues are imaginary.
Check yourself
The system dxdt=−x+3y\frac{\mathrm{d}x}{\mathrm{d}t} = -x + 3y, dydt=−2x−y\frac{\mathrm{d}y}{\mathrm{d}t} = -2x - y has eigenvalues −1±6 i-1 \pm \sqrt{6}\,\mathrm{i}. Which describes its trajectories?
Check yourself
The eigenvalues of dxdt=0.3x−y\frac{\mathrm{d}x}{\mathrm{d}t} = 0.3x - y, dydt=2x+0.1y\frac{\mathrm{d}y}{\mathrm{d}t} = 2x + 0.1y are complex. Find their real part.

Common mistakes

  • Checking only one rate for an equilibrium point. Both dxdt\frac{\mathrm{d}x}{\mathrm{d}t} and dydt\frac{\mathrm{d}y}{\mathrm{d}t} must be 00 at the same point.
  • Leaving the arrows off the eigenvector lines, or pointing them the wrong way. In for a negative eigenvalue, out for a positive one.
  • Guessing which way a spiral or ellipse turns. Substitute a point such as (1,0)(1, 0) and use the sign of dydt\frac{\mathrm{d}y}{\mathrm{d}t}.
  • Calling every complex case an ellipse. A non-zero real part gives a spiral; only a real part of 00 gives closed paths.
  • Pairing eλ1t\mathrm{e}^{\lambda_1 t} with the eigenvector of λ2\lambda_2 in the exact solution. Keep each eigenvalue with its own eigenvector.
  • Stopping at the eigenvalues when the question asks you to interpret. Say what happens to the quantities in context, such as which species dies out.

Exam tips

  • Sketch a phase portrait: draw both eigenvector lines through the origin, put arrows on them, and draw at least one trajectory with an arrow in every region.
  • Justify with reference to the eigenvalues: say whether they are real or complex and give the sign of the real part. Then name the behaviour. For a direction, show the substitution, such as dydt=0.5>0\frac{\mathrm{d}y}{\mathrm{d}t} = 0.5 > 0 at (1,0)(1, 0).
  • Find the eigenvalues: write det⁡(M−λI)=0\det(M - \lambda I) = 0 and the quadratic before the values. The set-up earns the method mark even when the GDC finds the roots.
  • Hence write down the general solution: use the booklet form with your own eigenvalues and eigenvectors. Any non-zero multiple of an eigenvector is accepted.

Key takeaways

Learn thisFor dxdt=ax+by\frac{\mathrm{d}x}{\mathrm{d}t} = ax + by, dydt=cx+dy\frac{\mathrm{d}y}{\mathrm{d}t} = cx + dy with non-zero eigenvalues, the only equilibrium point is the origin. A trajectory's gradient is dydx=dy/dtdx/dt\frac{\mathrm{d}y}{\mathrm{d}x} = \frac{\mathrm{d}y/\mathrm{d}t}{\mathrm{d}x/\mathrm{d}t}.
In bookletx=Aeλ1tp1+Beλ2tp2(real distinct eigenvalues)\boldsymbol{x} = A\mathrm{e}^{\lambda_{1}t}\boldsymbol{p}_{1} + B\mathrm{e}^{\lambda_{2}t}\boldsymbol{p}_{2}\quad\text{(real distinct eigenvalues)}
Learn thisReal eigenvalues: each eigenvector line through the origin holds straight-line trajectories, moving in for a negative eigenvalue and out for a positive one.
Learn thisFor dxdt=ax+by\frac{\mathrm{d}x}{\mathrm{d}t} = ax + by, dydt=cx+dy\frac{\mathrm{d}y}{\mathrm{d}t} = cx + dy, the origin is a stable equilibrium, with every solution heading into it, when both eigenvalues are negative or have a negative real part. When both are positive or have a positive real part, every solution except the origin itself heads away.
Learn thisReal eigenvalues of opposite signs: the origin is a saddle point. Trajectories come in alongside the negative eigenvalue's line and leave alongside the positive one's.
Learn thisComplex eigenvalues α±βi\alpha \pm \beta\mathrm{i} (α,β≠0\alpha, \beta \neq 0) give spirals. Imaginary eigenvalues ±βi\pm\beta\mathrm{i} (β≠0\beta \neq 0) give closed circles or ellipses. In both cases the motion is anticlockwise if dydt>0\frac{\mathrm{d}y}{\mathrm{d}t} > 0 at (1,0)(1, 0), clockwise if dydt<0\frac{\mathrm{d}y}{\mathrm{d}t} < 0.

Practise this

20 past-paper questions on Coupled differential equations in the AI HL bank, each with a mark scheme and a worked explanation.

Practise Coupled differential equations →