Autonomous equations and stability
When the rate of change of a quantity depends only on its current value, what does every solution do in the long run, and which constant states survive a small disturbance?
Where you are 6 in First-order equations
Why this exists
Slope fields and Logistic growth read equilibria and their stability from pictures. When the rate depends on the current value alone, the pictures become theorems: every solution is monotone, every bounded one settles at an equilibrium, the slope of the rate at an equilibrium decides its stability, and a parameter can make equilibria collide and vanish. All of it rests on uniqueness (Existence and uniqueness).
Definition
In plain words: an equation is autonomous when the rate of change depends on where the solution is, not on when.
Let I be an open interval and f : I \to \mathbb{R} continuously differentiable. The equation
is autonomous; here y is a function of t. A solution is a differentiable \varphi on an open interval, with values in I and \varphi'(t) = f(\varphi(t)). Since f and f' are continuous, the initial value problem y' = f(y), y(t_0) = y_0 has a solution, and two solutions on one open interval that agree at one point agree on all of it (Direction fields and solution curves, Results 3 and 4). By Result 3 below, every solution is a restriction of one on a largest open interval, called “the solution with y(t_0) = y_0”.
An equilibrium (also called a critical point) is a c \in I with f(c) = 0; the constant y = c is an equilibrium solution. It is isolated if some interval (c - \delta, c + \delta) contains no other zero of f. The phase line is the line of values of y with the equilibria marked and, on each interval between them, an arrow toward larger y where f > 0 and toward smaller y where f < 0.
Stability, as Slope fields describes it, made precise: an isolated equilibrium c is
- stable if there is \delta > 0 such that every solution with |y(t_0) - c| < \delta is defined for all t \ge t_0 and tends to c as t \to \infty;
- unstable if there is \delta > 0 such that every solution with 0 < |y(t_0) - c| < \delta takes a value outside (c - \delta, c + \delta) at some t > t_0;
- semi-stable if there is \delta > 0 such that solutions starting in one of (c - \delta, c) and (c, c + \delta) behave as in the stable case and those starting in the other as in the unstable case.
Many books call the first property asymptotically stable, keeping “stable” for the weaker property that solutions starting near c stay near c. For an isolated equilibrium the two coincide: a solution tending to c is monotone (Result 1), so stays between its start and c; Exercise 12 gives the converse.
For a family y' = f(y, h) with a parameter h, a bifurcation value is an h^* such that every interval around h^* contains values of h whose phase lines differ in the number of equilibria or in their stability. The bifurcation diagram plots the equilibria against h.
Intuition
A solution can never turn around: turning needs f(y) = 0 at the turning value, and the equilibrium solution there would meet it. Nor can it pass an equilibrium. Monotone and trapped, it converges, to a value where the motion stops.
Near an equilibrium c, f(y) \approx f'(c)(y - c), so u = y - c nearly obeys u' = f'(c)\,u, whose solutions u_0e^{f'(c)t} (Exponential growth and decay models) shrink when f'(c) < 0 and grow when f'(c) > 0. When f'(c) = 0 the next term of f decides (Example 5).
Results
1. Monotone or constant. Every solution of y' = f(y) is either an equilibrium solution or strictly monotone, and a strictly monotone solution takes no equilibrium value.
2. Limits are equilibria. Let f be only continuous on I. If \varphi solves y' = f(y) on (\alpha, \infty) and \varphi(t) \to L \in I as t \to \infty, then f(L) = 0; likewise as t \to -\infty. Consequently, under the standing hypotheses, a solution on (\alpha, \infty) with values in a closed bounded interval [p, q] \subseteq I tends to an equilibrium in [p, q].
3. The phase line. Let f > 0 on an interval (a, b) \subseteq I whose ends are each a zero of f or an end of I (possibly \pm\infty). Let y_0 \in (a, b), t_0 \in \mathbb{R}, and
each a positive number or \infty (Improper integrals). The solution with y(t_0) = y_0 is defined exactly on (t_0 - T_-, t_0 + T_+). It is increasing there, its values fill (a, b), and it tends to b at the right end and to a at the left end. If b is a zero of f, then T_+ = \infty; if a is, T_- = \infty. If instead f < 0 on (a, b), the same holds with “decreasing”, the two limits exchanged, \displaystyle T_+ = \int_a^{y_0} \frac{ds}{|f(s)|} and \displaystyle T_- = \int_{y_0}^{b} \frac{ds}{|f(s)|}.
4. Classification by signs. Let c be an equilibrium with no other zero of f in (c - \delta, c + \delta) \subseteq I. Then f has one sign on each of (c - \delta, c) and (c, c + \delta), and c is stable if f > 0 on the left and f < 0 on the right, unstable if f < 0 on the left and f > 0 on the right, and semi-stable if the two signs agree.
5. The derivative test. Let f(c) = 0.
- (a) If f'(c) < 0, then c is stable. For each k with 0 < k < -f'(c) there is \delta > 0 such that every solution with |y(t_0) - c| \le \delta satisfies |y(t) - c| \le |y(t_0) - c|\,e^{-k(t - t_0)} for all t \ge t_0.
- (b) If f'(c) > 0, then c is unstable. For each k with 0 < k < f'(c) there is \delta > 0 such that |y(t) - c| \ge |y(t_0) - c|\,e^{k(t - t_0)} as long as y stays in [c - \delta, c + \delta].
- (c) If f'(c) = 0, there is no conclusion: for y' = -y^3, y' = y^3 and y' = y^2 the equilibrium 0 is stable, unstable and semi-stable.
Why it is true
Result 1. Let \varphi solve on an open interval J. If \varphi'(t_1) = 0 for some t_1 \in J, then c = \varphi(t_1) has f(c) = 0, and the constant c and \varphi are solutions on J that agree at t_1, so by no crossing \varphi = c. Otherwise \varphi' = f \circ \varphi is continuous and never 0 on J, so by the intermediate value theorem (Extreme and intermediate value theorems, proved) it has one sign, and \varphi is strictly monotone. Such a \varphi takes no equilibrium value c, since there \varphi' = f(c) = 0. \square
Result 2. By continuity, \varphi'(t) = f(\varphi(t)) \to f(L). If f(L) > 0, then \displaystyle \varphi'(t) > \frac{1}{2}f(L) for t \ge T, and the mean value theorem gives \displaystyle \varphi(t) \ge \varphi(T) + \frac{1}{2}f(L)(t - T) \to \infty, a contradiction. The other cases are the same. For the consequence, \varphi is monotone (Result 1) and bounded, so it has a limit L \in [p, q] (The monotone convergence theorem), and f(L) = 0. \square
Result 3. Apply Result 2 of Separable equations in general with g = 1 and h = f, so that G(t) = t - t_0 and \displaystyle H(y) = \int_{y_0}^{y} \frac{ds}{f(s)}. The zero-free interval of y_0 is (a, b), on which H is increasing, with H(y) \to T_+ as y \to b and H(y) \to -T_- as y \to a. So H maps (a, b) onto (-T_-, T_+), the interval K^* of that theorem is (t_0 - T_-, t_0 + T_+), and \varphi(t) = H^{-1}(t - t_0) is increasing on it, onto (a, b), with the stated limits. Any solution \psi through (t_0, y_0) has values in (a, b): they form an interval in I containing y_0, and by Result 1 they include no end that is a zero of f. So \psi is a restriction of \varphi, by part (b) of that theorem.
If b is a zero of f, let L > 0 be the maximum of |f'| on [y_0, b]. For s \in [y_0, b) the mean value theorem gives 0 < f(s) = f(s) - f(b) \le L(b - s), so
The same argument works at a. For f < 0, apply all this to z = -y, which solves z' = -f(-z), with right side positive on (-b, -a). \square
Result 4. On (c, c + \delta), f is continuous with no zero, so it has one sign; likewise on the left. If f < 0 on (c, c + \delta), then c is the left end of a largest interval (c, b) with f < 0, and by Result 3 a solution with y(t_0) \in (c, c + \delta) decreases, is defined for all t \ge t_0 (T_+ = \infty), and tends to c. If f > 0 there, put \displaystyle \delta' = \frac{\delta}{2}: a solution with y(t_0) \in (c, c + \delta') increases, and its later values fill (y(t_0), b) with b \ge c + \delta, so it reaches c + \delta'. The left side is the mirror image, and the constant \delta' serves both sides. \square
Result 5. (a) Since f' is continuous and f'(c) < -k, there is \delta > 0 with [c - \delta, c + \delta] \subseteq I and f' \le -k on it. For s there, the mean value theorem gives f(s) = f'(\eta)(s - c) with \eta between c and s, so f has the sign of c - s: c is isolated, and stable by Result 4. By Result 3 a solution with |y(t_0) - c| \le \delta is defined for t \ge t_0 and stays between y(t_0) and c. Put u = y - c. Then
with \eta between c and y(t). Hence e^{2k(t - t_0)}u^2 has derivative e^{2k(t - t_0)}\big((u^2)' + 2ku^2\big) \le 0, so e^{2k(t - t_0)}u(t)^2 \le u(t_0)^2; take square roots.
(b) Now f' \ge k on [c - \delta, c + \delta], f has the sign of s - c there, and c is unstable by Result 4. While y stays in the interval, (u^2)' \ge 2ku^2, and the same computation reverses the bound.
(c) -y^3 is positive for y < 0 and negative for y > 0, y^3 the reverse, and y^2 is positive on both sides; apply Result 4. \square
Worked examples
A phase line with a silent derivative
For y' = y^2(y + 2)(y - 3), find and classify the equilibria, and describe the solutions with y(0) = -3, -1, 1 and 4.
SolutionWrite f(y) = y^4 - y^3 - 6y^2, so f'(y) = 4y^3 - 3y^2 - 12y. The equilibria are -2, 0 and 3, with f'(-2) = -20, f'(0) = 0 and f'(3) = 45. By the derivative test, -2 is stable and 3 is unstable; at 0 the test is silent. Away from 0 the factor y^2 is positive, so f has the sign of (y + 2)(y - 3): positive below -2, negative on (-2, 0) and (0, 3), positive above 3 (Figure 1). Near 0, solutions above approach 0 and those below leave it: 0 is semi-stable.
By Result 3, from -3 the solution increases to -2; from -1 it decreases to -2; from 1 it decreases to 0, and tends to 3 as t \to -\infty. Each is defined for all t \ge 0, since its forward end is a zero. From 4 the solution increases, and for s \ge 4, \displaystyle s - 3 \ge \frac{s}{4} and s + 2 \ge s, so \displaystyle f(s) \ge \frac{s^4}{4} and
So it becomes unbounded before \displaystyle t = \frac{1}{48}; the integral itself is about 0.0094.
Elena’s abalone pens
Elena raises abalone in sea pens. In thousands of animals, the stock y grows logistically (Logistic growth) at rate 0.4 per month with carrying capacity 10, and she harvests a steady h thousand per month, so
Take h = 0.64. Classify the equilibria, describe the long run from every starting stock, and find when a stock of 1.5 thousand runs out.
SolutionHere \displaystyle f(y) = -\frac{(y - 2)(y - 8)}{25} and \displaystyle f'(y) = \frac{10 - 2y}{25}, so \displaystyle f'(2) = \frac{6}{25} and \displaystyle f'(8) = -\frac{6}{25}: 2 thousand is unstable and 8 thousand stable. Since f > 0 on (2, 8) and f < 0 outside, every stock above 2 thousand tends to 8 thousand (Result 3). A stock below 2 thousand decreases, and since f(0) = -0.64 it reaches 0 (Figure 2).
The solution satisfies \displaystyle t = H(y) = \int_{1.5}^{y} \frac{ds}{f(s)}, so the time to reach 0 is the integral of \displaystyle \frac{1}{|f|} over [0, 1.5]. With \displaystyle \frac{1}{(s - 2)(s - 8)} = \frac{1}{6}\left(\frac{1}{s - 8} - \frac{1}{s - 2}\right),
The pens are empty after about 4.91 months.
The largest harvest the pens can bear
For Elena’s family \displaystyle y' = \frac{y(10 - y)}{25} - h with h \ge 0, find the equilibria and their stability for every h, the bifurcation value, and the solution at that value with y(0) = 7.
SolutionThe equilibria solve y^2 - 10y + 25h = 0:
Since \displaystyle f'(y) = \frac{10 - 2y}{25}, \displaystyle f'(y_\pm) = \mp\frac{2}{5}\sqrt{1 - h}. For 0 \le h < 1, y_+ is stable and y_- unstable, and as h rises they move together, meeting at 5 when h = 1 (Figure 3). Completing the square,
For h = 1 the only equilibrium is 5, and f \le 0, so 5 is semi-stable: stocks above it fall toward it, stocks below fall away. For h > 1, f \le 1 - h < 0: there is no equilibrium, and every stock is gone within \displaystyle \frac{y(0)}{h - 1} months. The bifurcation value is h = 1, the largest growth rate of the unharvested stock, \displaystyle \frac{kK}{4} = \frac{0.4 \cdot 10}{4} = 1 in the notation of Logistic growth.
At h = 1, u = y - 5 solves \displaystyle u' = -\frac{u^2}{25} with u(0) = 2. Separating, \displaystyle \frac{1}{u} = \frac{t}{25} + \frac{1}{2}, so
The stock approaches 5 like \displaystyle \frac{25}{t}, not exponentially: after 50 months it is still 0.4 thousand above.
A rate without a formula
Show that every solution of y' = 2 - y - y^3 tends to 1, with |y(t) - 1| \le |y(0) - 1|\,e^{-t} for all t \ge 0, and find a time after which the solution with y(0) = 3 is within 0.01 of 1.
SolutionHere f'(y) = -1 - 3y^2 \le -1, so f is strictly decreasing and 1 is its only zero, with f > 0 below and f < 0 above. By Result 3 every solution is defined for all t \ge 0 and tends to 1. Separation gives no formula for y in terms of t.
The proof of Result 5(a) needs only f' \le -k where the solution lives, and here f' \le -1 everywhere. With u = y - 1, (u^2)' = 2f'(\eta)\,u^2 \le -2u^2, so u(t)^2 \le u(0)^2e^{-2t} for every starting value. From y(0) = 3, |y(t) - 1| \le 2e^{-t} \le 0.01 once e^t \ge 200, that is for t \ge \ln 200 \approx 5.30. The bound is safe but slow: near 1 the slope is f'(1) = -4, and a numerical solution is within 0.01 of 1 from about t = 1.07 on.
A short proof: when the first derivative vanishes
Let f be twice continuously differentiable on I, with f(c) = f'(c) = 0 and f''(c) \ne 0. Prove that c is semi-stable, approached from above when f''(c) < 0 and from below when f''(c) > 0.
SolutionSuppose f''(c) > 0. By continuity there is \delta > 0 with [c - \delta, c + \delta] \subseteq I and f'' > 0 on it. Then f' is strictly increasing there, and f'(c) = 0 gives f' < 0 left of c and f' > 0 right of c. So f decreases on [c - \delta, c] and increases on [c, c + \delta], and f(c) = 0 gives f > 0 on both (c - \delta, c) and (c, c + \delta). By Result 4, c is semi-stable: solutions below rise toward c, and solutions above rise away. If f''(c) < 0, the same argument with every inequality reversed gives f < 0 on both sides: solutions above fall toward c. \square
This confirms Example 1 (f''(0) = -12) and Example 3 at h = 1 (\displaystyle f'' = -\frac{2}{25}).
Boundaries
- Continuity alone is not enough. Without a Lipschitz condition a solution can reach an equilibrium in finite time, as the tank \displaystyle y' = -\frac{1}{10}\sqrt{y} of Existence and uniqueness does; then Results 1 and 3 fail.
- Isolation matters. For \displaystyle f(y) = y^3\sin\frac{1}{y}, equilibria crowd toward 0, and 0 is none of the three types (Exercise 13).
- One dimension. Monotonicity uses the order of the line; solutions of autonomous systems in the plane can circle (Phase portraits). With time in the equation, y = \sin t solves y' = \cos t and turns around.
- The derivative test is local. How far the starts that tend to c reach comes from the phase line: in Example 2, every stock above 2 thousand.
Where this goes next
Nonlinear systems and linearization replaces f'(c) by the Jacobian matrix at an equilibrium of a planar system. The discrete analogue, iterating \mathbf{x}_{k+1} = A\mathbf{x}_k, is Matrix powers, dynamical systems and steady states. Springs, damping and resonance has solutions that oscillate toward equilibrium, which Result 1 rules out here.
Exercises
Phase lines and the derivative test
- For y' = (4 - y^2)(y - 1), find and classify the equilibria, and find \lim_{t \to \infty} y(t) for y(0) = -3, 0, 1.5 and 3.
- Find every equilibrium of y' = 1 + 2\cos y, classify each, and find \lim_{t \to \infty} y(t) when y(0) = 0.
- For y' = y^2 + 2y, y(0) = 1, find T_+ and T_- of Result 3, the interval of the solution, and its limit as t \to -\infty. Check against the formula y = \dfrac{2e^{2t}}{3 - e^{2t}}.
- Show that 0 is the only equilibrium of y' = e^y - 1 - y and classify it. Find \lim_{t \to \infty} y(t) when y(0) = -1, and show that the solution with y(0) = 1 becomes unbounded before t = 2.
- For the equation of Exercise 1, find the largest \delta with f' \le -10 on [-2 - \delta, -2 + \delta], and state the bound that the proof of Result 5(a) then gives.
Harvesting
- Elena harvests at h = 0.91. Find and classify the equilibria, and find when a stock of 3 thousand runs out.
- For 0 \le h < 1, the linear equation u' = f'(y_+)\,u approximates the return to y_+ in Example 3. Find f'(y_+) and the time its solutions take to shrink by a factor of 10, for h = 0.75, 0.91 and 0.99.
- Jin, who runs the next pens, harvests nothing and adds 0.44 thousand young abalone per month, so \displaystyle y' = \frac{y(10 - y)}{25} + 0.44. Find and classify the equilibria, and the long run of a stock that starts at 0.
Bifurcations
- For y' = ay - y^2, find the equilibria and their stability for a < 0, a = 0 and a > 0, and the bifurcation value.
- Find the bifurcation values of y' = y^3 - 3y + h, and classify the equilibria for h = 0 and for h = 2.
Proofs
- Let \varphi and \psi be solutions on \mathbb{R} with \psi(t_1) = \varphi(t_0). Prove that \psi(t) = \varphi(t + t_0 - t_1) for all t.
- Let c be an isolated equilibrium such that for every \varepsilon > 0 there is \eta > 0 for which every solution with |y(t_0) - c| < \eta is defined for all t \ge t_0 and stays in (c - \varepsilon, c + \varepsilon). Prove that c is stable.
- Let \displaystyle f(y) = y^3\sin\frac{1}{y} for y \ne 0 and f(0) = 0. Prove that f is continuously differentiable, that 0 is not isolated but has the property of Exercise 12, and that solutions starting arbitrarily near 0 do not tend to 0.
- Let f be three times continuously differentiable, with f(c) = f'(c) = f''(c) = 0 and f'''(c) < 0. Prove that c is stable.
- Let \varphi be a solution on all of \mathbb{R} with values in a closed bounded interval [p, q] \subseteq I. Prove that \varphi is an equilibrium solution, or else it tends to equilibria a \ne b at the two ends of time with no equilibrium strictly between a and b.
Answers
- f'(y) = -3y^2 + 2y + 4, with f'(-2) = -12, f'(1) = 3, f'(2) = -4: -2 and 2 are stable, 1 is unstable. Since f(-3) = 20, f(0) = -4, f(1.5) = 0.875 and f(3) = -10, the limits are -2, -2, 2 and 2.
- \displaystyle \cos y = -\frac{1}{2} gives \displaystyle y = \pm\frac{2\pi}{3} + 2\pi n, n an integer. With f'(y) = -2\sin y, \displaystyle f' = -\sqrt{3} at \displaystyle \frac{2\pi}{3} + 2\pi n (stable) and \displaystyle f' = \sqrt{3} at \displaystyle -\frac{2\pi}{3} + 2\pi n (unstable). Since f(0) = 3 > 0 and no zero lies in \displaystyle \left(0, \frac{2\pi}{3}\right), the limit is \displaystyle \frac{2\pi}{3}.
- f(y) = y(y + 2) > 0 on (0, \infty), and 0 is a zero, so T_- = \infty. Since \displaystyle \frac{1}{s(s + 2)} = \frac{1}{2}\left(\frac{1}{s} - \frac{1}{s + 2}\right), \displaystyle T_+ = \int_1^\infty \frac{ds}{s(s + 2)} = \frac{\ln 3}{2} \approx 0.549. The solution lives on \displaystyle \left(-\infty, \frac{\ln 3}{2}\right) and tends to 0 as t \to -\infty. The formula agrees: y(0) = 1, both sides of the equation equal \displaystyle \frac{12e^{2t}}{(3 - e^{2t})^2}, and e^{2t} = 3 at the end.
- f'(y) = e^y - 1 vanishes only at 0 and f'' = e^y > 0, so f has its minimum f(0) = 0 there and is positive elsewhere. By Example 5, with f''(0) = 1, 0 is semi-stable, approached from below; from -1 the solution increases to 0. For s \ge 0, \displaystyle e^s - 1 - s \ge \frac{s^2}{2} (the terms of the series for e^s are nonnegative), so \displaystyle T_+ \le \int_1^\infty \frac{2\,ds}{s^2} = 2. (In fact T_+ \approx 0.78.)
- f' \le -10 means 3y^2 - 2y - 14 \ge 0, that is \displaystyle y \le \frac{1 - \sqrt{43}}{3} or \displaystyle y \ge \frac{1 + \sqrt{43}}{3}. The interval lies in the first set exactly when \displaystyle -2 + \delta \le \frac{1 - \sqrt{43}}{3}, so \displaystyle \delta = \frac{7 - \sqrt{43}}{3} \approx 0.148. Then |y(0) + 2| \le \delta gives |y(t) + 2| \le |y(0) + 2|\,e^{-10t} for t \ge 0.
- \displaystyle f = -\frac{(y - 3.5)(y - 6.5)}{25}, with f'(3.5) = 0.12 (unstable) and f'(6.5) = -0.12 (stable). From 3 the stock falls, and the time is \displaystyle \int_0^3 \frac{25\,ds}{(s - 3.5)(s - 6.5)}, which is \displaystyle \frac{25}{3}\ln\frac{49}{13} \approx 11.06 months.
- \displaystyle f'(y_+) = -\frac{2}{5}\sqrt{1 - h}, which is -0.2, -0.12 and -0.04 per month. Solutions u_0e^{f'(y_+)t} shrink by 10 after \displaystyle \frac{\ln 10}{|f'(y_+)|}, about 11.51, 19.19 and 57.56 months: recovery slows as h nears the bifurcation value.
- \displaystyle f = -\frac{(y - 11)(y + 1)}{25}: -1 is unstable and 11 stable (\displaystyle f' = \pm\frac{12}{25}). Since f(0) = 0.44 > 0, the stock rises to 11 thousand, above the carrying capacity.
- Equilibria 0 and a, with f'(0) = a and f'(a) = -a. For a < 0, 0 is stable and a unstable; for a > 0 the reverse. For a = 0, y' = -y^2 has the single equilibrium 0, semi-stable, approached from above. The bifurcation value is a = 0, where the equilibria pass through each other and exchange stability.
- Equilibria satisfy h = 3y - y^3, which decreases on (-\infty, -1], increases on [-1, 1] to the value 2, and decreases on [1, \infty), with value -2 at -1. So there are three equilibria for |h| < 2 and one for |h| > 2: the bifurcation values are h = \pm 2. With f' = 3y^2 - 3: for h = 0, 0 is stable (f' = -3) and \displaystyle \pm\sqrt{3} unstable (f' = 6). For h = 2, f = (y - 1)^2(y + 2): -2 is unstable (f' = 9), and 1 is semi-stable, approached from below, by Example 5 with f''(1) = 6.
- \chi(t) = \varphi(t + t_0 - t_1) is defined on \mathbb{R}, and by the chain rule \chi'(t) = \varphi'(t + t_0 - t_1) = f(\chi(t)), so it is a solution with \chi(t_1) = \varphi(t_0) = \psi(t_1). By no crossing, \chi = \psi. \square
- Choose \varepsilon > 0 with [c - \varepsilon, c + \varepsilon] \subseteq I containing no zero of f but c, and let \eta be as given. A solution with |y(t_0) - c| < \eta is defined on [t_0, \infty) with values in [c - \varepsilon, c + \varepsilon], so by Result 2 it tends to an equilibrium there, which can only be c. So c is stable with \delta = \eta. \square
- For y \ne 0, \displaystyle f'(y) = 3y^2\sin\frac{1}{y} - y\cos\frac{1}{y}, which tends to 0, and \displaystyle f'(0) = \lim_{y \to 0} y^2\sin\frac{1}{y} = 0, so f' is continuous. The zeros \displaystyle \pm\frac{1}{m\pi} tend to 0, so 0 is not isolated. Given \varepsilon, take \displaystyle \eta = \frac{1}{m\pi} < \varepsilon. A solution with |y(t_0)| < \eta is an equilibrium or starts between two consecutive zeros inside (-\eta, \eta), and stays between them (Result 3). But the equilibria \displaystyle \frac{1}{m\pi}, as near 0 as we like, do not tend to 0. \square
- By continuity f''' < 0 on some [c - \delta, c + \delta] \subseteq I, so f'' is strictly decreasing there, positive left of c and negative right of it. So f' increases up to c and decreases after it, with f'(c) = 0: f' < 0 on both sides. Then f is strictly decreasing on [c - \delta, c + \delta], and f(c) = 0 gives f > 0 left of c and f < 0 right of it. By Result 4, c is stable. \square
- By Result 1, \varphi is an equilibrium solution or strictly monotone; say increasing. Being bounded, it has limits a as t \to -\infty and b as t \to \infty (The monotone convergence theorem), equilibria by Result 2, with a < \varphi(0) < b. Its values fill (a, b), and by Result 1 none is an equilibrium. \square