Flow Lab

Syllabus / Module 3: First-Order ODEs

Lesson 3.8

Autonomous Equations, Equilibria & Stability

y' = f(y): phase lines, equilibria, stability by sign of f', logistic growth, and qualitative analysis without solving.

Written byClaude Sonnet

🌊
Why this matters for simulation: Stability analysis is how engineers predict when smooth (laminar) flow turns turbulent.

By the end of this lesson you can…

  • Draw the phase line of y' = f(y) from the sign of f and sketch solutions without solving.
  • State why a 1D autonomous solution is monotone and converges to an equilibrium, and what the argument needs.
  • Classify equilibria with f'(y*) when it is nonzero and with the sign of f when it is zero.
  • Compute the relaxation time 1/|f'(y*)| and tie it to the skydiver of lesson 3.5.
  • Analyse logistic growth and harvesting, including the critical rate and finite-time extinction.
  • Explain why Euler on the logistic equation oscillates past h r = 2, via the exact logistic map.
On this page
·
  1. Autonomous equations and the phase line
  2. Equilibria and the sign of fff
  3. The theorem: one-dimensional solutions never turn around
  4. Stable, unstable, semi-stable
  5. The linear stability test
  6. Logistic growth
  7. Harvesting and the critical rate
  8. What a solver does to this theorem
  9. Stability analysis in fluid dynamics
  10. Exercise: equilibria, classification and the Euler cycle
  11. Problems
  12. Key Takeaways

Run Euler on a population model with a carrying capacity, and for a large enough step the population overshoots the capacity and bounces between two values for ever. The true solution never does that, and nothing in the code tells you which picture is the equation and which is the solver.

3.7 Existence & Uniqueness ended with a promise: a solution of y′=y(1−y)y' = y(1-y) that starts between the constants 00 and 11 is trapped there, and we learned it without solving anything. This lesson turns that trick into a theory for every equation whose right side does not mention tt: cooling, terminal velocity, growth, and the first step of every fluid-stability argument.

▶ Video Phase lines, equilibria and their stability, the logistic equation, harvesting, and Euler's method turning a monotone solution into an oscillationcaptions only
Transcript
  1. Run Euler on a population model with a carrying capacity. For a big enough step it overshoots, and bounces between two values forever. The true solution never does. This lesson shows why, without solving anything.
  2. Autonomous means y prime depends on y alone. Every column of the direction field is identical, and a solution slid along the t axis is another solution. Collapse the columns onto one line: the phase line.
  3. Take y prime equals y, times y minus one, times two minus y. Graph f and mark its roots: zero, one and two, the equilibria. Where f is positive the arrow points up, where negative, down. Solutions follow the arrows. Zero and two attract: stable. One repels: unstable, a watershed.
  4. Why can't a solution turn around? Between two equilibria it is trapped, since solutions cannot cross. A turn needs slope zero, so f would be zero there, an equilibrium, and uniqueness makes it constant. So it is monotone, and converges. In one dimension, an autonomous solution cannot oscillate.
  5. Near an equilibrium, f looks like its tangent line. The gap obeys epsilon prime equals f prime times epsilon. Negative slope: stable. Positive: unstable. The time constant is one over the size of the slope.
  6. But if f prime is zero, the tangent line says nothing. These three all have zero slope at zero, yet they behave differently. The sign of f decides.
  7. Y prime equals y cubed, from one half, blows up at t equals two. Minus y cubed creeps down to 0.2041 at t equals ten. Stable, but slowly, with no time constant.
  8. Logistic growth, with r and K equal to one. Zero has slope plus one: unstable. K has slope minus one: stable. From 0.1 the solution is an S-curve, inflecting at K over two, where f peaks. That happens at t equals ln nine, 2.197.
  9. Now harvest at a constant rate h. The graph of f slides down. At 0.2 the equilibria are 0.7236, stable, and 0.2764, unstable. At 0.25 they merge, semi-stable. At 0.3 there are none.
  10. With none, nothing holds the population up. From one at h equals 0.3, it reaches zero at t equals 10.288. From one half, at 5.144. The critical rate is r K over four, the peak of the growth curve. That collision is a bifurcation, next lesson.
  11. Now the solver. Euler on the logistic equation, from 0.1. At h r of 0.5 it follows the true curve. At 1.5 it overshoots one slightly; at 1.9 it peaks at 1.0807 before settling. All converge, but the true solution never exceeds one.
  12. At h r equals 2.2 it never settles: it alternates 1.1628 and 0.7462, overshooting K. At 2.5 it cycles through four values. The cobweb shows why: Euler on the logistic equation is exactly the logistic map, mu equals one plus h r. Its fixed point is stable only below h r equals two, the lesson 3.2 limit, h below two over the size of f prime.
  13. Same limit, real physics. For the eighty kilogram skydiver, f prime at terminal speed is minus 0.34666 per second, so Euler needs h below 5.769 seconds. At 5.5 it settles at 56.597. At 5.9 it alternates between 67.19 and 43.50, forever.
  14. Smaller steps fix it. RK4 pushes the limit to 2.785, but past it, at 2.8 it settles at 0.8708, and at 3.0 at 0.7608. Neither is an equilibrium of the ODE; that is at one. A steady output can still be wrong.
  15. Fluids use the same idea. Perturb a steady flow and linearise: disturbances grow like e to the sigma t, and the sign of the real part of sigma decides, exactly f prime with a number replaced by an operator. It predicts when a smooth flow becomes unstable to small disturbances. But pipe flow is linearly stable at every tested Reynolds number, yet turns turbulent near two thousand, because finite disturbances trigger transition. That is from memory.

Autonomous equations and the phase line

An ODE is autonomous when the right side has no explicit tt:

y′=f(y).y' = f(y).

Cooling, quadratic drag, y′=y2y' = y^2 and the logistic equation are all of this form. Two consequences follow.

The direction field repeats in tt. The slope at height yy is f(y)f(y) whatever the time, so every vertical column of the field of 3.2 Direction Fields & Euler's Method Revisited is the same column. Solutions are translates of each other: if y(t)y(t) solves the equation, so does y(t−c)y(t-c), because ddty(t−c)=y′(t−c)=f(y(t−c))\frac{d}{dt}y(t-c) = y'(t-c) = f(y(t-c)).

The story lives on a line. Draw the yy-axis vertically and put an arrow on each stretch of it, up where f>0f \gt 0 and down where f<0f \lt 0: this is the phase line. Time has disappeared, yet the picture predicts every solution curve.

Equilibria and the sign of ff

A number y∗y^* with f(y∗)=0f(y^*) = 0 gives the constant solution y≡y∗y \equiv y^*, an equilibrium. These are the solutions lost by dividing by ff in lesson 3.3. Between neighbouring equilibria ff is continuous and never zero, so it has one sign there and the arrow points one way.

Example 1: y′=y(1−y)y' = y(1-y). Equilibria 00 and 11. f<0f \lt 0 for y<0y \lt 0 (down), f>0f \gt 0 on (0,1)(0,1) (up), f<0f \lt 0 for y>1y \gt 1 (down). Solutions leave 00 and are drawn to 11.

Example 2: three equilibria, y′=y(y−1)(2−y)y' = y(y-1)(2-y). The equilibria are 0,1,20, 1, 2, and the sign of ff on each stretch is:

intervaly<0y \lt 0(0,1)(0,1)(1,2)(1,2)y>2y \gt 2
sign of ff++−-++−-
motionupdownupdown

So 00 and 22 attract and 11 repels. The repeller is a watershed: start below 11 and you settle at 00, above it and you settle at 22.

Example 3: y′=y2y' = y^2. One equilibrium, 00, and f≥0f \ge 0 on both sides, so both arrows point up. Below 00 a solution climbs toward 00 (y=−1/(t+1)y = -1/(t+1) from −1-1); above 00 it leaves and reaches infinity at t=1/y0t = 1/y_0 (lesson 3.3).

The sim shows the graph of ff, the phase line and the solution curves on one shared yy axis.

◆ Interactive Phase line, graph of f, and solutions; pick an equation, drag the start, move the harvest and Euler sliders

Loading simulation…

  • Cubic. Curves flatten onto the dashed lines at 00 and 22 and peel away from 11.
  • Harvested logistic. Raise hh: the equilibria approach, merge at h=0.25h = 0.25, and vanish.
  • y′=y2y' = y^2, y′=−y3y' = -y^3. The readout says “inconclusive” when f′(y∗)=0f'(y^*) = 0.

The theorem: one-dimensional solutions never turn around

Here is the backbone of the lesson. Let ff be continuously differentiable (Lipschitz suffices), so that 3.7 Existence & Uniqueness applies.

Theorem. Let y(t)y(t) be a solution of y′=f(y)y' = f(y). (1) It is constant or strictly monotone. (2) It never reaches an equilibrium in finite time. (3) If it stays in a bounded interval inside the domain of ff, it exists for all tt and converges to an equilibrium as t→∞t \to \infty, and likewise as t→−∞t \to -\infty.

Proof of (2). Suppose y(t1)=y∗y(t_1) = y^* with f(y∗)=0f(y^*) = 0. The constant y∗y^* solves the same IVP, so uniqueness gives y≡y∗y \equiv y^*.

Proof of (1). If y′(t1)=0y'(t_1) = 0 then f(y(t1))=0f(y(t_1)) = 0 and y(t1)y(t_1) is an equilibrium, so yy is constant by (2). Otherwise y′=f(y(t))y' = f(y(t)) is continuous and never zero, hence of one sign by the intermediate value theorem, and yy is strictly monotone.

Proof of (3). A bounded solution cannot grow without bound, so by lesson 3.7 it cannot end. Monotone and bounded, it has a limit LL. If f(L)≠0f(L) \ne 0, then ∣y′∣≥12∣f(L)∣|y'| \ge \tfrac12|f(L)| for large tt and yy would grow linearly, contradicting boundedness. So f(L)=0f(L) = 0. □\square

So a solution that starts between two neighbouring equilibria is trapped by non-crossing (3.7) and converges to one of them: the lesson-3.7 trick with its conclusion.

The consequence to remember: in one dimension an autonomous solution is monotone, so it cannot oscillate. No overshoot, no ringing, no periodic orbit; it cannot even cross the equilibrium it approaches. Oscillation needs two or more dimensions (a pendulum needs position and velocity), which is where Modules 5 and 7 go.

The hypotheses matter. Without uniqueness, y′=∣y∣y' = \sqrt{|y|} has solutions (t−c)2/4(t-c)^2/4 that leave the equilibrium 00 at a finite time, violating (2). Without autonomy, y′=cos⁡ty' = \cos t has the oscillating solution sin⁡t\sin t. And (3) needs boundedness: an untrapped solution may run to ±∞\pm\infty forever (y′=1y' = 1) or in finite time (y′=y2y' = y^2).

Stable, unstable, semi-stable

These words describe what nearby solutions do near an isolated equilibrium y∗y^*.

  • Stable: every solution that starts close enough to y∗y^* tends to y∗y^* as t→∞t \to \infty (this is asymptotic stability, which is what most ODE courses mean).
  • Unstable: solutions that start close to y∗y^* on either side move away from it.
  • Semi-stable: attracting from one side, repelling on the other.

By the theorem, signs alone classify every isolated equilibrium. If ff goes from ++ to −- as yy increases through y∗y^*, it is stable; from −- to ++, unstable; if it has the same sign on both sides, semi-stable. For y′=y2y' = y^2 the signs are (+,+)(+,+): semi-stable at 00. In the cubic, 00 and 22 are stable and 11 is unstable.

The linear stability test

Writing y=y∗+εy = y^* + \varepsilon with ε\varepsilon small, and using f(y∗)=0f(y^*) = 0,

ε′=f(y∗+ε)=f′(y∗) ε+o(ε),soε≈ε0 eλt,λ=f′(y∗).\varepsilon' = f(y^* + \varepsilon) = f'(y^*)\,\varepsilon + o(\varepsilon), \qquad\text{so}\qquad \varepsilon \approx \varepsilon_0\, e^{\lambda t},\quad \lambda = f'(y^*).

The linear ODE decides the matter when λ≠0\lambda \ne 0. The precise reason is the phase-line rule: f′(y∗)<0f'(y^*) \lt 0 means f(y)/(y−y∗)f(y)/(y - y^*) is negative near y∗y^*, so ff goes from ++ to −- and the theorem makes y∗y^* stable. Likewise:

f′(y∗)f'(y^*)verdictgap to y∗y^*
<0\lt 0stableshrinks like e−t/τe^{-t/\tau}
>0\gt 0unstablegrows like et/τe^{t/\tau}
=0= 0inconclusivethe tangent line says nothing

with time constant τ=1/∣f′(y∗)∣\tau = 1/|f'(y^*)|, the time for the gap to change by a factor ee.

The skydiver again. In lesson 3.5 the 80 kg skydiver had v′=g(1−v2/vt2)v' = g(1 - v^2/v_t^2) with vt=56.6v_t = 56.6 m/s. At the stable equilibrium

f′(vt)=−2g vtvt2=−2gvt=−0.347 s−1,τ=10.347=2.88 s,f'(v_t) = -\frac{2g\,v_t}{v_t^2} = -\frac{2g}{v_t} = -0.347\ \text{s}^{-1}, \qquad \tau = \frac{1}{0.347} = 2.88\ \text{s},

the figures quoted there.

When f′(y∗)=0f'(y^*) = 0 the test is silent. The three equations y′=y3y' = y^3, y′=−y3y' = -y^3 and y′=y2y' = y^2 all have f′(0)=0f'(0) = 0 and three different behaviours:

  • y′=y3y' = y^3: y=y0/1−2y02ty = y_0/\sqrt{1 - 2y_0^2 t} blows up at t=1/(2y02)t = 1/(2y_0^2) (at t=2t = 2 for y0=0.5y_0 = 0.5). Unstable.
  • y′=−y3y' = -y^3: y=y0/1+2y02ty = y_0/\sqrt{1 + 2y_0^2 t} tends to 00 (from 0.50.5, y(10)=0.2041y(10) = 0.2041). Stable, yet only like t−1/2t^{-1/2}, with no time constant.
  • y′=y2y' = y^2: semi-stable.

There the sign of ff on each side decides. In one dimension the sign rule is the master test and f′(y∗)f'(y^*) is the quick one.

✅ Concept check

For $y' = (y-1)^2(y-3)$, which description of the equilibria is correct?

✅ Concept check

Near a stable equilibrium, $f'(y^*) = -0.8$ per unit time. What is the largest Euler step $h$ for which the Euler iteration stays stable near $y^*$?

Logistic growth

Populations grow in proportion to size until crowding stops them: y′=r y (1−y/K)y' = r\,y\,(1 - y/K), r,K>0r, K \gt 0, y≥0y \ge 0, with carrying capacity KK. The equilibria are 00 and KK, with f′(y)=r(1−2y/K)f'(y) = r(1 - 2y/K):

f′(0)=r    (unstable),f′(K)=−r    (stable).f'(0) = r \;\;(\text{unstable}),\qquad f'(K) = -r \;\;(\text{stable}).

The time scale at both is 1/r1/r. The theorem predicts every shape: from 0<y0<K0 \lt y_0 \lt K the solution rises monotonically to KK, from y0>Ky_0 \gt K it falls monotonically to KK.

Solving it (lesson 3.3). Separate with the partial fractions 1y(1−y/K)=1y+1K−y\frac{1}{y(1 - y/K)} = \frac1y + \frac{1}{K - y}:

ln⁡∣yK−y∣=rt+C  ⟹  yK−y=y0K−y0 ert  ⟹  y=K1+(Ky0−1)e−rt.\ln\left|\frac{y}{K-y}\right| = rt + C \;\Longrightarrow\; \frac{y}{K - y} = \frac{y_0}{K - y_0}\,e^{rt} \;\Longrightarrow\; y = \frac{K}{1 + \left(\dfrac{K}{y_0} - 1\right)e^{-rt}} .

Verify. At t=0t = 0 it gives y0y_0. With D=1+c e−rtD = 1 + c\,e^{-rt}, c=K/y0−1c = K/y_0 - 1: y′=Krc e−rt/D2y' = Krc\,e^{-rt}/D^2 and ry(1−y/K)=rK(D−1)/D2r y (1 - y/K) = rK(D - 1)/D^2, the same. (A central-difference check on three parameter sets gives a residual below 10−910^{-9}.) For y0>0y_0 \gt 0 the formula holds where D≠0D \ne 0: all tt if y0<Ky_0 \lt K; if y0>Ky_0 \gt K, c<0c \lt 0 and it holds for t>1rln⁡(1−K/y0)t \gt \frac1r\ln(1 - K/y_0), a negative time (−0.693-0.693 for r=K=1r = K = 1, y0=2y_0 = 2).

The inflection. Differentiating, y′′=f′(y) y′=r(1−2y/K) y′y'' = f'(y)\,y' = r(1 - 2y/K)\,y', and y′≠0y' \ne 0 on 0<y<K0 \lt y \lt K, so y′′=0y'' = 0 exactly at y=K/2y = K/2. There the growth rate f(K/2)=rK/4f(K/2) = rK/4 is its maximum: before that point growth accelerates, after it the solution decelerates, and the S-curve results. For r=K=1r = K = 1, y0=0.1y_0 = 0.1 it is reached at t=ln⁡9=2.197t = \ln 9 = 2.197.

Harvesting and the critical rate

Remove individuals at a constant rate h≥0h \ge 0:

y′=r y(1−yK)−h.y' = r\,y\left(1 - \frac{y}{K}\right) - h .

Equilibria solve y2−Ky+hK/r=0y^2 - Ky + hK/r = 0:

y∗=K2(1±1−4hrK),f′(y∗)=r(1−2y∗K)=∓ r1−4hrK.y^* = \frac K2\left(1 \pm \sqrt{1 - \frac{4h}{rK}}\right), \qquad f'(y^*) = r\left(1 - \frac{2y^*}{K}\right) = \mp\, r\sqrt{1 - \frac{4h}{rK}} .

For r=K=1r = K = 1 (upper root stable, lower unstable):

hhequilibriaf′f'meaning
0.200.200.72360.7236 and 0.27640.2764−0.4472-0.4472 and +0.4472+0.4472stable upper, unstable lower
0.250.250.50.5 (one)00merged, semi-stable
0.300.30nonef<0f \lt 0 everywhere

(Check at h=0.2h = 0.2: 0.7236 (1−0.7236)=0.20000.7236\,(1 - 0.7236) = 0.2000. The unstable equilibrium is a threshold: from y0=0.2y_0 = 0.2 RK4 hits y=0y = 0 at t=2.152t = 2.152.)

The critical rate is hc=rK/4h_c = rK/4, the maximum of the logistic growth curve from the previous section. No harvest above the best regrowth rate can be sustained. At h=hch = h_c we have f=−rK(y−K2)2≤0f = -\frac rK\left(y - \frac K2\right)^2 \le 0, signs (−,−)(-,-): semi-stable, with f′=0f' = 0 and the linear test silent again.

Above hch_c the parabola ff lies below the axis: no equilibrium, nothing to trap the solution, and every population decreases. For r=K=1r = K = 1, h=0.3h = 0.3, put u=y−12u = y - \tfrac12 (so u0=12u_0 = \tfrac12 from y0=1y_0 = 1): then u′=−(u2+a2)u' = -(u^2 + a^2) with a2=0.05a^2 = 0.05, so u=atan⁡(arctan⁡(u0/a)−at)u = a\tan\big(\arctan(u_0/a) - at\big). From y0=1y_0 = 1 the population hits y=0y = 0 at

t=arctan⁡(u0/a)+arctan⁡(1/(2a))a=10.288,t = \frac{\arctan(u_0/a) + \arctan\big(1/(2a)\big)}{a} = 10.288,

which matches RK4; from y0=0.5y_0 = 0.5 it is 5.1445.144, exactly half. The formal solution would continue to −∞-\infty at t=12.17t = 12.17, but the model ends at y=0y = 0. Extinction in finite time is something no unharvested logistic solution does.

Recovery slows before the cliff: the stable equilibrium has τ=1/(r1−4h/rK)\tau = 1/(r\sqrt{1 - 4h/rK}), which for r=K=1r = K = 1 is 2.242.24 at h=0.2h = 0.2 and 5.005.00 at h=0.24h = 0.24.

At hch_c the two equilibria collide and annihilate: a bifurcation, a qualitative change of the phase line as a parameter moves, and the subject of 3.9 Bifurcations.

✅ Concept check

Euler with step $1.9$ on $y' = y(1 - y)$ from $y(0) = 0.3$ gives $0.3,\ 0.699,\ 1.0988,\ 0.8926,\ 1.0748,\ 0.9221,\dots$ and eventually settles at $1$. What do you conclude?

What a solver does to this theorem

Scaling yy by KK and tt by 1/r1/r leaves y′=y(1−y)y' = y(1-y) and one parameter, the product hrhr. Euler gives yn+1=yn+a yn(1−yn)y_{n+1} = y_n + a\,y_n(1 - y_n) with a=hra = hr. Start at y0=0.1y_0 = 0.1 (r=K=1r = K = 1) and run it:

a=hra = hrwhat Euler does
0.50.5climbs monotonically to 11
1.51.5reaches 11, but peaks at 1.03851.0385 on the way
1.91.9reaches 11 (200 steps: within 10−1010^{-10}), peak 1.08071.0807
2.22.2alternates 1.1628, 0.74621.1628,\ 0.7462 for ever
2.52.5cycles through 1.2250, 0.5359, 1.1577, 0.70121.2250,\ 0.5359,\ 1.1577,\ 0.7012
3.13.1goes negative at step 5 and diverges

The true solution rises monotonically from 0.10.1 toward 11, so no value above 11 and no decrease is possible. The oscillation is the method’s, not the equation’s.

Why: Euler is the logistic map. Put xn=a yn(1+a)Kx_n = \dfrac{a\,y_n}{(1+a)K} (with K=1K = 1 here). Then

xn+1=a1+a(yn(1+a)−ayn2)=(1+a) xn−(1+a) xn2=μ xn(1−xn),μ=1+a.x_{n+1} = \frac{a}{1+a}\Big(y_n(1+a) - a y_n^2\Big) = (1+a)\,x_n - (1+a)\,x_n^2 = \mu\,x_n(1 - x_n), \qquad \mu = 1 + a .

Euler on the logistic ODE is exactly the iteration x↦μx(1−x)x \mapsto \mu x(1-x) (run side by side for 300 steps, the two agree to 10−1510^{-15} for a=0.5,1.9,2.2,2.5a = 0.5, 1.9, 2.2, 2.5). The equilibrium y=Ky = K is the fixed point x∗=1−1/μx^* = 1 - 1/\mu, and its stability factor is μ(1−2x∗)=2−μ=1−a\mu(1 - 2x^*) = 2 - \mu = 1 - a. It is stable iff ∣1−a∣<1|1-a| \lt 1, that is

1<μ<3  ⟺  hr<2  ⟺  h<2∣f′(K)∣.1 \lt \mu \lt 3 \iff hr \lt 2 \iff h \lt \frac{2}{|f'(K)|}.

This is the limit h<2/λh \lt 2/\lambda of lesson 3.2 with λ=∣f′(K)∣=r\lambda = |f'(K)| = r. At μ=3\mu = 3 the fixed point loses stability and a 2-cycle is born; it goes unstable at μ=1+6=3.449\mu = 1 + \sqrt 6 = 3.449 (a=2.449a = 2.449), giving the 4-cycle seen at a=2.5a = 2.5 (direct runs give period 2 at a=2.44a = 2.44, period 4 at a=2.46a = 2.46). The cascade of doublings accumulates near μ≈3.57\mu \approx 3.57: the Lyapunov exponent of the map, computed numerically, changes sign there: −0.018-0.018 at μ=3.565\mu = 3.565, +0.011+0.011 at 3.573.57, +0.184+0.184 at 3.63.6. Beyond it the iterates are generally irregular. For μ>4\mu \gt 4 (a>3a \gt 3) typical starts are pushed out of [0,1][0,1] and then diverge, as in the a=3.1a = 3.1 row.

Why a solver can do what a solution cannot. A curve cannot jump over an equilibrium; an Euler step is a finite hop. With g(y)=y+ay(1−y)g(y) = y + ay(1-y), g(y)−1=(1−y)(ay−1)g(y) - 1 = (1-y)(ay - 1), so a step from y<1y \lt 1 lands above 11 whenever y>1/ay \gt 1/a.

What is a fix, and what is not.

  • Smaller hh. For a≤1a \le 1, gg increases on [0,1][0,1] and maps it into itself, so the iterates increase monotonically to 11, matching the theorem. The condition for qualitative faithfulness is hr≤1hr \le 1, not the stability condition hr<2hr \lt 2: the runs at 1.51.5 and 1.91.9 converge and are still wrong in shape.
  • A higher-order method. RK4 applied to this equation converges to KK at a=2.5a = 2.5, so its limit is further out: ∣1+z+z2/2+z3/6+z4/24∣<1|1 + z + z^2/2 + z^3/6 + z^4/24| \lt 1 holds for −2.785<z<0-2.785 \lt z \lt 0 with z=−hrz = -hr. Beyond it, at a=2.8a = 2.8 and 3.03.0, RK4 settles on 0.87080.8708 and 0.76080.7608, which are not equilibria of the ODE: the limit moves, it does not vanish.
  • Implicit Euler. Solving yn+1=yn+hf(yn+1)y_{n+1} = y_n + h f(y_{n+1}) for the root between yny_n and KK (it exists, since y−yn−hf(y)y - y_n - hf(y) changes sign there) gives monotone iterates for every hh (checked at hr=1,2.5,10,100hr = 1, 2.5, 10, 100): “trapped between equilibria” in discrete form, and the idea of 4.5 Stability Regions & Stiffness.
◆ Interactive Logistic option: drag the Euler step h r across 2 and compare the red iterates with the exact solution

Loading simulation…

Try hr=0.5hr = 0.5, 1.51.5, 2.22.2, 2.52.5 and compare with the table.

The skydiver again. The same threshold is h<2/0.3467=5.77h \lt 2/0.3467 = 5.77 s. Euler on v′=g(1−v2/vt2)v' = g(1 - v^2/v_t^2) from v(0)=0v(0) = 0 settles on vt=56.60v_t = 56.60 at h=5.5h = 5.5 s; at h=5.9h = 5.9 s it alternates between 43.5043.50 and 67.1967.19 m/s for ever.

🐍 PythonPlayground: Euler, RK4 and implicit Euler on the logistic equation, and the skydiver's step limit
⌘/Ctrl + Enter

Stability analysis in fluid dynamics

The limits, honestly. The sign of a growth rate speaks only about small perturbations, and a linearly stable state can still lose to a large enough disturbance. The standard example (from memory, not derived here): Hagen-Poiseuille flow in a circular pipe has no linear instability at any Reynolds number that has been tested, yet in experiments it becomes turbulent at Reynolds numbers of order two thousand, because finite-size disturbances trigger the transition. So stability analysis predicts when the smooth state becomes unstable to small disturbances; it does not by itself predict turbulence, which 11.9 Turbulence: An Introduction takes up.

Exercise: equilibria, classification and the Euler cycle

Implement the three tools of the lesson: find the equilibria of a given ff, classify them by the sign of ff (which works when f′=0f' = 0), and detect the period of the long-run Euler iterates.

🧪 Coding exerciseExercise: find equilibria, classify them by sign, detect an Euler cycle
⌘/Ctrl + Enter

Problems

✏️ Pen & paperPhase line, rate and blow-up●○○

For y′=(y−1)(y−3)y' = (y-1)(y-3): (a) classify the equilibria with f′f'; (b) find lim⁡t→∞y\lim_{t\to\infty} y for y(0)=2.5y(0) = 2.5; (c) for y(0)=3.5y(0) = 3.5 find the blow-up time, using y−3y−1=Ae2t\frac{y-3}{y-1} = A e^{2t}.

Show worked solution

(a) f′=2y−4f' = 2y - 4: f′(1)=−2f'(1) = -2 stable, f′(3)=2f'(3) = 2 unstable.

(b) On (1,3)(1,3), f<0f \lt 0: yy decreases, trapped above 11. The limit is 11.

(c) 1(y−1)(y−3)=12(1y−3−1y−1)\frac{1}{(y-1)(y-3)} = \frac12\big(\frac{1}{y-3} - \frac{1}{y-1}\big) gives ln⁡∣y−3y−1∣=2t+C\ln\left|\frac{y-3}{y-1}\right| = 2t + C. At y0=3.5y_0 = 3.5, A=0.5/2.5=0.2A = 0.5/2.5 = 0.2. Blow-up means y−3y−1→1\frac{y-3}{y-1} \to 1, so 0.2 e2t=10.2\,e^{2t} = 1 and t=12ln⁡5=0.8047t = \tfrac12\ln 5 = 0.8047 (RK4 agrees).

✏️ Pen & paperA solution that turns around?●●○

(a) Can y′=f(y)y' = f(y), with ff continuously differentiable, have a solution with a local maximum at t=1t = 1? (b) y=sin⁡ty = \sin t satisfies y′=1−y2y' = \sqrt{1 - y^2} for ∣t∣≤π/2|t| \le \pi/2. Why is this no contradiction, and what happens at t=π/2t = \pi/2?

Show worked solution

(a) No. At a local maximum y′(1)=0y'(1) = 0, so y(1)y(1) is an equilibrium and uniqueness makes yy constant.

(b) 1−y2\sqrt{1-y^2} is not Lipschitz at y=±1y = \pm 1 (its derivative is unbounded). The solution reaches the equilibrium y=1y = 1 at the finite time π/2\pi/2, which part (2) forbids when uniqueness holds; indeed y≡1y \equiv 1 is a second solution through (π/2,1)(\pi/2, 1). Past π/2\pi/2, sin⁡t\sin t solves the other equation y′=−1−y2y' = -\sqrt{1-y^2}.

✏️ Pen & paperHarvesting in proportion to the stock●●○

Effort EE removes fish at rate E yE\,y: y′=r y(1−y/K)−E yy' = r\,y(1 - y/K) - E\,y. (a) Find the equilibria and their stability for E<rE \lt r and E>rE \gt r. (b) For r=K=1r = K = 1 find the long-run catch E y∗E\,y^* and its maximum. (c) For r=K=1r = K = 1, E=1.2E = 1.2, is extinction reached in finite time?

Show worked solution

(a) f=y((r−E)−ry/K)f = y\big((r - E) - ry/K\big): equilibria 00 and y∗=K(1−E/r)y^* = K(1 - E/r), with f′(0)=r−Ef'(0) = r - E and f′(y∗)=−(r−E)f'(y^*) = -(r - E). For E<rE \lt r, 00 is unstable and y∗y^* stable. For E>rE \gt r, y∗<0y^* \lt 0 is not a population and 00 is stable.

(b) Ey∗=E(1−E)E y^* = E(1-E) has maximum 14\tfrac14 at E=12E = \tfrac12, the same rK/4rK/4 as for constant harvest, at the stable equilibrium y∗=12y^* = \tfrac12.

(c) No. y≡0y \equiv 0 stays an equilibrium, so by part (2) no solution reaches it in finite time. For y′=−y(0.2+y)y' = -y(0.2 + y) from 0.50.5, y(20)=0.00265>0y(20) = 0.00265 \gt 0.

✏️ Pen & paperEuler's equilibria and its step limit●●●

Euler is yn+1=g(yn)y_{n+1} = g(y_n), g(y)=y+hf(y)g(y) = y + h f(y). (a) Show the fixed points of gg are exactly the equilibria of the ODE. (b) Give the stability factor of a fixed point. (c) For y′=y−y3y' = y - y^3 find the equilibria, f′f', and the Euler step limit at each. (d) At h=1.1h = 1.1 from y0=0.5y_0 = 0.5 the iterates settle into the cycle 0.81455, 1.116060.81455,\ 1.11606. Verify one step each way.

Show worked solution

(a) g(y)=yg(y) = y iff hf(y)=0hf(y) = 0 iff f(y)=0f(y) = 0. Euler keeps equilibria and changes their stability.

(b) g′(y∗)=1+hf′(y∗)g'(y^*) = 1 + hf'(y^*); the gap is multiplied by it each step, so stability needs ∣1+hf′(y∗)∣<1|1 + hf'(y^*)| \lt 1.

(c) f=y(1−y)(1+y)f = y(1-y)(1+y): equilibria 0,±10, \pm 1 with f′=1−3y2f' = 1 - 3y^2, so f′(0)=1f'(0) = 1, f′(±1)=−2f'(\pm 1) = -2. At 00 the factor is 1+h>11 + h \gt 1 for every h>0h \gt 0. At ±1\pm 1 it is 1−2h1 - 2h: stable iff h<1h \lt 1.

(d) 0.81455+1.1(0.81455−0.54046)=1.116060.81455 + 1.1(0.81455 - 0.54046) = 1.11606 and 1.11606+1.1(1.11606−1.39016)=0.814551.11606 + 1.1(1.11606 - 1.39016) = 0.81455. At h=0.9h = 0.9 the same start settles at 11.

Key Takeaways

  • An autonomous equation y′=f(y)y' = f(y) has the same direction field at every tt; its behaviour is read off the phase line.
  • Equilibria are the roots of ff; between them ff has one sign, so solutions move monotonically.
  • Theorem. For Lipschitz ff, a 1D autonomous solution is constant or strictly monotone, never reaches an equilibrium in finite time, and if bounded exists for all time and converges to an equilibrium. It cannot oscillate.
  • Stability is read from the sign of ff on each side. f′(y∗)<0f'(y^*) \lt 0 or >0\gt 0 gives stable or unstable with time constant 1/∣f′(y∗)∣1/|f'(y^*)|; f′(y∗)=0f'(y^*) = 0 is inconclusive (y3y^3, −y3-y^3, y2y^2).
  • Logistic: y=K/(1+(K/y0−1)e−rt)y = K/(1 + (K/y_0 - 1)e^{-rt}), equilibria 00 (unstable) and KK (stable), inflection at K/2K/2.
  • Harvesting: equilibria K2(1±1−4h/rK)\frac K2\big(1 \pm \sqrt{1 - 4h/rK}\big) merge at hc=rK/4h_c = rK/4 (a bifurcation); above it the population dies out in finite time.
  • Euler on the logistic equation is exactly the map x↦μx(1−x)x \mapsto \mu x(1-x), μ=1+hr\mu = 1 + hr: stable iff hr<2hr \lt 2, faithful (monotone) iff hr≤1hr \le 1. Wiggles in a 1D solution are artifacts.
  • Linear stability of a PDE replaces f′(y∗)f'(y^*) by eigenvalues; it says when a smooth flow is unstable to small disturbances, not that turbulence follows.
Finished this lesson?

Mark it complete to track your progress (saved in this browser). Reading every section does this for you.

Where this leads

These later lessons build directly on this one: