math.wiki
Calculus › Differential equations › First-order equations › Separable equations in general
Sign in

Separable equations in general

Separating the variables turns dy/dx = g(x)h(y) into two integrals. Does that find every solution, what happens where h vanishes, and which other equations can be brought to this form?

Where you are 6 in First-order equations

Why this exists

Separable differential equations separates \dfrac{dy}{dx} = g(x)\,h(y), integrates, adds back the constant solutions lost by dividing, and picks the branch through the starting point. Two questions were left open. Does the formula catch every solution of the initial value problem, without the continuous derivative of h that the uniqueness theorem stated there assumes? And which equations that are not separable become separable after a change of variable?

This lesson proves that separation finds every solution as long as it stays away from the zeros of h, assuming only that h is continuous; gives an integral test that decides whether a solution can reach a zero of h; and reduces y' = F(y/x) and y' = F(ax + by + k) to separable equations. Throughout, y is a function of x, or of t when the variable is time.

Definition

In plain words: an equation is separable when its slope is a factor that sees only x times a factor that sees only y.

Let I and J be open intervals, g : I \to \mathbb{R} and h : J \to \mathbb{R} continuous. The equation

\frac{dy}{dx} = g(x)\,h(y)

is separable. A solution is a differentiable \varphi on an open interval K \subseteq I, with values in J and \varphi'(x) = g(x)\,h(\varphi(x)), as in Direction fields and solution curves. An initial value problem adds y(x_0) = y_0 with x_0 \in I, y_0 \in J. A function f on I \times J is separable if f(x, y) = g(x)\,h(y) for some functions g and h.

If h(c) = 0, the constant y = c is an equilibrium solution. If h(y_0) \ne 0, the zero-free interval J_0 of y_0 is the largest open interval inside J that contains y_0 and no zero of h; each finite endpoint of J_0 is a zero of h or an endpoint of J. Put

\begin{aligned} H(y) &= \int_{y_0}^{y} \frac{ds}{h(s)} \quad (y \in J_0), \\ G(x) &= \int_{x_0}^{x} g(t)\,dt \quad (x \in I). \end{aligned}

At a zero c of h with no zero in (c, c + \delta], the integral \displaystyle \displaystyle\int_c^{c + \delta} \frac{ds}{|h(s)|} means the limit as \varepsilon \to 0^+ of the integral over [c + \varepsilon, c + \delta]; it is finite or \infty.

A function f defined for x \ne 0 is homogeneous of degree zero if f(\lambda x, \lambda y) = f(x, y) for every \lambda > 0. Taking \lambda = \dfrac{1}{x} shows f(x, y) = F\!\left(\dfrac{y}{x}\right) on x > 0, with F(v) = f(1, v), and y' = F\!\left(\dfrac{y}{x}\right) is called a homogeneous equation. (The word means something unrelated for linear equations in Linear equations and integrating factors.)

Intuition

Take g = 1 first. Then H(y) is the x-distance a solution needs to climb from height y_0 to height y: where |h| is large the curve is steep and the trip is short. Separation says exactly this, H(\varphi(x)) = x - x_0. A general g only changes the clock, from x - x_0 to G(x).

Near a zero c of h the slope dies away, and the question is whether the trip to c is finite. If |h(s)| \le L|s - c|, the trip costs at least \displaystyle \displaystyle\int \frac{ds}{L|s - c|}, a logarithm, which is infinite: the solution creeps toward c like e^{-Lx} and never arrives. If h vanishes more gently, like |s - c|^{3/4}, the trip is finite, the solution arrives, and the equilibrium and the arriving curve are two solutions through one point (Figure 1).

Results

1. Test for separability. Let f be defined on I \times J. Then f is separable if and only if

f(x, y)\,f(a, b) = f(x, b)\,f(a, y)

for all x, a \in I and y, b \in J. When f(a, b) \ne 0, one factorization is g(x) = f(x, b), h(y) = \dfrac{f(a, y)}{f(a, b)}.

2. Separation finds every solution away from the zeros of h. Let g be continuous on I, h continuous on J, h(y_0) \ne 0, and J_0, H, G as above. Then H is strictly monotone on J_0 and H(J_0) is an open interval containing 0. Let K^* be the largest open interval containing x_0 on which G(x) \in H(J_0). Then

  • (a) \varphi^*(x) = H^{-1}(G(x)) solves the initial value problem on K^*;
  • (b) every solution of it on an interval K with values in J_0 has K \subseteq K^* and equals \varphi^* on K.

No Lipschitz condition and no derivative of h is assumed.

3. Reaching an equilibrium. Let g be continuous on I, h continuous on J, h(c) = 0, and h \ne 0 on (c, c + \delta] \subseteq J.

  • (a) If \displaystyle \displaystyle\int_c^{c + \delta} \frac{ds}{|h(s)|} = \infty, no solution takes both the value c and a value in (c, c + \delta).
  • (b) If that integral is finite, h > 0 on (c, c + \delta] and g > 0 on [x_0, x_0 + \varepsilon) for some \varepsilon > 0, then the initial value problem y(x_0) = c has a solution equal to c for x \le x_0 and greater than c for x > x_0. It is not unique.
  • (c) If h is Lipschitz on J, that is |h(y_1) - h(y_2)| \le L|y_1 - y_2|, then every solution either is an equilibrium or has all its values in one zero-free interval, where Result 2 describes it.

The statements for (c - \delta, c) and for the other signs of g and h follow by replacing y by 2c - y or x by 2x_0 - x.

4. Substitutions. Let F be continuous on an open interval V.

  • (a) On an interval K \subseteq (0, \infty), y solves y' = F\!\left(\dfrac{y}{x}\right) with \dfrac{y}{x} \in V if and only if v = \dfrac{y}{x} solves the separable equation v' = \dfrac{1}{x}\big(F(v) - v\big). Each v_0 with F(v_0) = v_0 gives the line solution y = v_0 x.
  • (b) If b \ne 0, y solves y' = F(ax + by + k) with ax + by + k \in V if and only if v = ax + by + k solves v' = a + b\,F(v), whose right side depends on v alone.
Why it is true

Result 1. If f = g(x)\,h(y), both sides of the identity equal g(x)h(y)g(a)h(b). Conversely, if f is identically 0, take g = h = 0; otherwise choose (a, b) with f(a, b) \ne 0 and divide the identity by it. \square

Result 2. On J_0, H' = \dfrac{1}{h} is continuous and never 0, so by the intermediate value theorem it keeps one sign and H is strictly monotone. A continuous strictly monotone function maps an open interval onto an open interval, and H(y_0) = 0. The inverse H^{-1} : H(J_0) \to J_0 is differentiable with (H^{-1})'(u) = h(H^{-1}(u)) (Derivatives of inverse functions). The set of x with G(x) \in H(J_0) is open, since G is continuous, and contains x_0, so K^* exists.

(a) By the chain rule and the fundamental theorem, \varphi^{*\prime}(x) = h(\varphi^*(x))\,g(x) on K^*, and \varphi^*(x_0) = H^{-1}(0) = y_0.

(b) For a solution \varphi on K with values in J_0,

\frac{d}{dx}H(\varphi(x)) = \frac{\varphi'(x)}{h(\varphi(x))} = g(x) = G'(x),

so H(\varphi) - G is constant on K, and it is 0 at x_0. Hence G(x) = H(\varphi(x)) \in H(J_0) for x \in K, so K \subseteq K^* and \varphi = H^{-1} \circ G = \varphi^*. \square

Result 3(a). Suppose \varphi solves on K with \varphi(x_1) \in (c, c + \delta) and \varphi(x_2) = c, say x_1 < x_2 (the other order is the same argument reversed). Let x^* be the least x > x_1 with \varphi(x) = c. On [x_1, x^*), \varphi > c, and \varphi(x) \to c as x \to x^*, so some x_3 < x^* has c < \varphi < c + \delta on [x_3, x^*). The computation in (b) gives, for x_3 \le x < x^*,

\int_{\varphi(x_3)}^{\varphi(x)} \frac{ds}{h(s)} = \int_{x_3}^{x} g(t)\,dt.

As x \to x^{*-} the right side tends to the finite \int_{x_3}^{x^*} g. Since h keeps one sign on (c, c + \delta], the left side has absolute value \displaystyle \int_{\varphi(x)}^{\varphi(x_3)} \frac{ds}{|h(s)|}, which tends to \infty as \varphi(x) \to c^+. Contradiction. \square

Result 3(b). Let \displaystyle P(y) = \int_c^y \frac{ds}{h(s)} on [c, c + \delta]: finite, continuous, strictly increasing, P(c) = 0, \displaystyle P' = \frac{1}{h} on (c, c + \delta]. G is strictly increasing on [x_0, x_0 + \varepsilon) with G(x_0) = 0; choose \eta \in (0, \varepsilon) with G(x_0 + \eta) < P(c + \delta). Set \varphi = c for x \le x_0 and \varphi = P^{-1}(G(x)) for x_0 < x < x_0 + \eta. For x > x_0, \varphi'(x) = h(\varphi)\,g as in Result 2; for x < x_0 both sides are 0. At x_0, \varphi is continuous with left derivative 0, and by the mean value theorem its right difference quotient is h(\varphi(\xi))\,g(\xi) for some \xi \in (x_0, x), which tends to h(c)\,g(x_0) = 0. \square

Result 3(c). If every value of \varphi is a zero of h, then \varphi' = g\,h(\varphi) = 0 and \varphi is an equilibrium. Otherwise \varphi(x_1) lies in a zero-free interval J_1. If \varphi left J_1, the point x_2 nearest x_1 with \varphi(x_2) \notin J_1 would have \varphi(x_2) = c, an endpoint of J_1 and a zero of h, with values of \varphi in J_1 arbitrarily close to c. But |h(s)| = |h(s) - h(c)| \le L|s - c|, so \displaystyle \int \frac{ds}{|h(s)|} \ge \int \frac{ds}{L|s - c|} = \infty next to c, and (a) is contradicted. \square

Result 4(a). For x \ne 0, y is differentiable exactly when \displaystyle v = \frac{y}{x} is, and y = xv gives y' = v + xv'. So y' = F(v) if and only if xv' = F(v) - v. If F(v_0) = v_0, the constant v_0 solves the second equation. Part (b) is the same computation with \displaystyle y = \frac{v - ax - k}{b} (Exercise 13). \square

Worked examples

Example 1

Recovering the factors

Decide whether y' = x^2y^2 - x^2 + y^2 - 1 is separable, and solve it with y(0) = 0.

Solution

Apply Result 1 with (a, b) = (0, 0), where f(0, 0) = -1 \ne 0. The candidate factors are f(x, 0) = -x^2 - 1 and \dfrac{f(0, y)}{f(0, 0)} = 1 - y^2, with product (x^2 + 1)(y^2 - 1) = x^2y^2 - x^2 + y^2 - 1. So the equation is separable: y' = (x^2 + 1)(y^2 - 1).

The zeros of h(y) = y^2 - 1 are \pm 1, so J_0 = (-1, 1), and by partial fractions

\begin{aligned} H(y) &= \int_0^y \frac{ds}{s^2 - 1} = \frac{1}{2}\ln\frac{1 - y}{1 + y}, \\ G(x) &= \frac{x^3}{3} + x. \end{aligned}

H maps (-1, 1) onto \mathbb{R}, so K^* = \mathbb{R}, and H(y) = G(x) gives \dfrac{1 - y}{1 + y} = e^{2G}, that is

y = -\tanh\left(\frac{x^3}{3} + x\right), \quad x \in \mathbb{R}.

Check: y' = -\left(1 - \tanh^2\right)(x^2 + 1), which is (x^2 + 1)(y^2 - 1). Result 2(b) makes this the only solution with values in (-1, 1), and since h is Lipschitz on bounded intervals, Result 3(c) keeps every solution through (0, 0) inside (-1, 1). So it is the only solution.

Example 2

The maximal interval from the theorem

Solve y' = 2(1 + y^2)\cos x, y(0) = 0, and find the maximal interval.

Solution

Here h(y) = 1 + y^2 has no zero, so J_0 = \mathbb{R}, H(y) = \arctan y and H(J_0) = \left(-\dfrac{\pi}{2}, \dfrac{\pi}{2}\right), while G(x) = 2\sin x. So K^* is the largest interval around 0 with |\sin x| < \dfrac{\pi}{4}. Since \sin increases on \left[-\dfrac{\pi}{2}, \dfrac{\pi}{2}\right] and \dfrac{\pi}{4} < 1,

\begin{aligned} K^* &= \left(-\arcsin\frac{\pi}{4}, \arcsin\frac{\pi}{4}\right) \\ &\approx (-0.9033, 0.9033), \end{aligned}

and the solution is y = \tan(2\sin x). Check: y' = \sec^2(2\sin x) \cdot 2\cos x = 2(1 + y^2)\cos x. Every solution restricts this one (Result 2(b)), and y \to \pm\infty at the ends.

Example 3

An equilibrium that is reached

Show that y' = 4x|y|^{3/4}, y(1) = 0, has solutions other than y = 0.

Solution

Here h(y) = |y|^{3/4} is continuous, zero only at 0, and g(x) = 4x > 0 near 1. The integral

\int_0^1 \frac{ds}{s^{3/4}} = \Big[4s^{1/4}\Big]_0^1 = 4

is finite, so Result 3(b) applies. Its construction has P(y) = 4y^{1/4} and G(x) = 2x^2 - 2, and P(\varphi) = G gives \varphi = \left(\dfrac{x^2 - 1}{2}\right)^4 for x > 1. Leaving at any a \ge 1 works the same way:

\varphi_a(x) = \begin{cases} 0, & x \le a, \\ \left(\dfrac{x^2 - a^2}{2}\right)^4, & x > a. \end{cases}

For x > a, \varphi_a' = 4x\left(\dfrac{x^2 - a^2}{2}\right)^3, which is 4x|\varphi_a|^{3/4} since x^2 - a^2 > 0. At x = a both one-sided derivatives are 0, as in the proof of Result 3(b). So y = 0 and every \varphi_a with a \ge 1 solve the problem on \mathbb{R} (Figure 1). Result 3(c) cannot apply: \dfrac{h(y)}{|y|} = |y|^{-1/4} is unbounded near 0, so h is not Lipschitz there.

1 2 1 2 3 x y
Figure 1Solutions of y' = 4x|y|^{3/4} with y(1) = 0 (Example 3): the equilibrium y = 0, and the curves that leave it at x = a for a = 1, 1.25, 1.5, 1.75.
Example 4

Andre and Layla’s bioreactor

Andre and Layla model the yeast biomass B, in grams, in a bioreactor t hours after seeding by

\frac{dB}{dt} = \frac{1}{2}B\ln\frac{400}{B}, \qquad B(0) = 25.

Solve it, find when B = 200, and show that B stays between 0 and 400 although the right side is not Lipschitz near B = 0.

Solution

Here B is a function of t. Extend \displaystyle h(B) = \frac{1}{2}B\ln\frac{400}{|B|} by h(0) = 0, which is continuous since B\ln B \to 0; only B \ge 0 has a meaning. The zeros are 0 and \pm 400, so J_0 = (0, 400). With \displaystyle u = \ln\frac{400}{s}, \displaystyle du = -\frac{ds}{s},

\begin{aligned} H(B) &= \int_{25}^{B} \frac{2\,ds}{s\ln(400/s)} \\ &= -2\ln\frac{\ln(400/B)}{\ln 16}. \end{aligned}

As B \to 0^+, H \to -\infty; as B \to 400^-, H \to \infty. So H(J_0) = \mathbb{R}, K^* = \mathbb{R}, and H(B) = t gives \displaystyle \ln\frac{400}{B} = 4\ln 2 \cdot e^{-t/2}:

B(t) = 400 \cdot 2^{-4e^{-t/2}}.

Check: \dfrac{B'}{B} = 2\ln 2 \cdot e^{-t/2}, which is \dfrac{1}{2}\ln\dfrac{400}{B}. The biomass is 200 when 4e^{-t/2} = 1, at t = 4\ln 2 \approx 2.77 hours, and B \to 400 as t \to \infty.

H unbounded at both ends of J_0 means the integral of \displaystyle \frac{1}{|h|} diverges at 0 and at 400, so by Result 3(a) no solution starting in (0, 400) reaches either, although \displaystyle \frac{h(B)}{B} = \frac{1}{2}\ln\frac{400}{B} \to \infty rules out a Lipschitz bound at 0. The same divergence for B < 0 shows that B = 0 is the only solution with B(t_0) = 0.

Example 5

A homogeneous equation and a spiral

Solve y' = \dfrac{2x + y}{x - 2y}, y(1) = 0, and find the maximal interval.

Solution

The right side is defined off the line \displaystyle y = \frac{x}{2} and is unchanged when x and y are both multiplied by \lambda, so it is homogeneous with F(v) = \dfrac{2 + v}{1 - 2v}; the start has v = 0 in \displaystyle V = \left(-\infty, \frac{1}{2}\right). Since F(v) - v = \dfrac{2(1 + v^2)}{1 - 2v} has no zero on V, Results 4(a) and 2 give

\begin{aligned} \int_0^v \frac{1 - 2s}{1 + s^2}\,ds &= \int_1^x \frac{2}{t}\,dt, \\ \arctan v - \ln(1 + v^2) &= 2\ln x. \end{aligned}

With \displaystyle v = \frac{y}{x} this is \displaystyle \arctan\frac{y}{x} = \ln(x^2 + y^2). On x > 0, \displaystyle \arctan\frac{y}{x} is the polar angle \theta and x^2 + y^2 = r^2, so \theta = 2\ln r: the logarithmic spiral r = e^{\theta/2}.

The restriction x > 0 came from the substitution, not the equation. On the spiral x = e^{\theta/2}\cos\theta, y = e^{\theta/2}\sin\theta,

\begin{aligned} \frac{dx}{d\theta} &= \tfrac{1}{2}e^{\theta/2}(\cos\theta - 2\sin\theta), \\ \frac{dy}{d\theta} &= \tfrac{1}{2}e^{\theta/2}(\sin\theta + 2\cos\theta), \end{aligned}

with ratio \dfrac{2x + y}{x - 2y}, so the spiral is an integral curve of the field (Direction fields and solution curves, Result 1). For \theta_+ - \pi < \theta < \theta_+, with \displaystyle \theta_+ = \arctan\frac{1}{2}, \cos\theta - 2\sin\theta > 0, so x increases with \theta and the arc is the graph of a solution through (1, 0); at both ends it meets \displaystyle y = \frac{x}{2}, outside the domain. Off that line the right side has a continuous partial derivative in y, so by no crossing (Direction fields and solution curves, Result 4) this arc is the solution (Figure 2). Its maximal interval runs from \displaystyle x_- = -\frac{2}{\sqrt{5}}e^{(\theta_+ - \pi)/2} \approx -0.2344 to \displaystyle x_+ = \frac{2}{\sqrt{5}}e^{\theta_+/2} \approx 1.1278, crossing the y-axis at \left(0, -e^{-\pi/4}\right).

−1 1 −1 1 x y
Figure 2The field of y' = \dfrac{2x + y}{x - 2y} and the spiral r = e^{\theta/2} (Example 5). The solid arc, between the two points where the spiral meets the dashed line y = \tfrac{x}{2}, is the solution through (1, 0), defined for -0.2344 < x < 1.1278.
Example 6

Leaving one equilibrium line for the next

Solve y' = \cos(x - y), y(0) = -\dfrac{\pi}{2}.

Solution

Result 4(b) with a = 1, b = -1, k = 0: v = x - y solves v' = 1 - \cos v, with \displaystyle v(0) = \frac{\pi}{2}. The zeros v = 2\pi n give the line solutions y = x - 2\pi n. Since |\sin v| \le 1, h(v) = 1 - \cos v is Lipschitz with L = 1, so by Result 3(c) v stays in (0, 2\pi): the curve lies strictly between y = x and y = x - 2\pi. Separating, with \displaystyle 1 - \cos v = 2\sin^2\frac{v}{2},

\int \frac{dv}{2\sin^2(v/2)} = -\cot\frac{v}{2} = x + C,

and \displaystyle \cot\frac{\pi}{4} = 1 gives C = -1, so \displaystyle \cot\frac{v}{2} = 1 - x. On (0, \pi) the solution of \cot w = u is \displaystyle w = \frac{\pi}{2} - \arctan u, hence

y = x - \pi + 2\arctan(1 - x), \quad x \in \mathbb{R}.

Check: y' = 1 - \dfrac{2}{1 + (1 - x)^2}, and \cos(x - y) = -\cos\big(2\arctan(1 - x)\big) is \dfrac{(1 - x)^2 - 1}{(1 - x)^2 + 1}, the same. As x \to -\infty, y - x \to 0; as x \to \infty, y - x \to -2\pi. The solution leaves one equilibrium line and approaches the next without touching either (Figure 3).

−4 4 −8 −4 4 x y
Figure 3The field of y' = \cos(x - y), the equilibrium lines y = x and y = x - 2\pi (dashed), and the solution y = x - \pi + 2\arctan(1 - x) through \left(0, -\tfrac{\pi}{2}\right) (Example 6).
Example 7

A short proof: dilations

Prove that if \varphi solves y' = F\!\left(\dfrac{y}{x}\right) on an interval K \subseteq (0, \infty), then for each \lambda > 0 the function \psi(x) = \lambda\varphi\!\left(\dfrac{x}{\lambda}\right) solves it on \lambda K = \{\lambda x : x \in K\}.

Solution

For x \in \lambda K, \displaystyle \frac{x}{\lambda} \in K, and by the chain rule

\begin{aligned} \psi'(x) &= \varphi'\!\left(\frac{x}{\lambda}\right) = F\!\left(\frac{\varphi(x/\lambda)}{x/\lambda}\right) \\ &= F\!\left(\frac{\psi(x)}{x}\right). \qquad \square \end{aligned}

In Example 5, the scaled spirals r = \lambda e^{\theta/2} are the solution curves through other points.

Boundaries

  • Result 2 stops at a zero of h. Whether a solution can reach one is Result 3’s question, and past such a point uniqueness may fail (Example 3, Exercise 16).
  • Result 4(a) works on x > 0 or on x < 0. A solution may still cross the y-axis (Example 5); that is read from the curve, not from v.
  • Most equations reduce to neither form, for example y' = x^2 + y^2 of Direction fields and solution curves.

Where this goes next

y' = x + y is not separable, yet Result 4(b) solves it (v = x + y gives v' = 1 + v); Linear equations and integrating factors solves every y' + p(x)\,y = q(x) with an integrating factor. Existence and uniqueness proves uniqueness for y' = f(x, y) with f Lipschitz in y, more than Result 2 needs. Autonomous equations and stability take g = 1 and read long-run behavior from the zeros of h. Exact equations see the implicit solution H(y) - G(x) = C as a level curve of a function of two variables.

Exercises

Recognizing separable equations

  1. Use Result 1 to decide which are separable, and factor those that are: (a) f(x, y) = xy - 2x - 3y + 6; (b) f(x, y) = 1 + xy; (c) f(x, y) = \sin(x + y) + \sin(x - y).
  2. Show that y' = 2^{x + y} + 2^{x - y} is separable, solve it with y(0) = 0, and find the maximal interval.

The separation theorem

  1. Solve y' = (1 + y^2)e^x, y(0) = 0, and find the maximal interval from Result 2.
  2. Solve y' = e^y\sin x, y(0) = 0, and find the maximal interval from Result 2.
  3. Solve y' = x(4 - y^2), y(0) = 0. Show the solution is defined for every x and find its limits as x \to \pm\infty.
  4. Without finding a formula, prove that the solution of y' = (1 + y^4)\cos x, y(0) = 0, is defined on all of \mathbb{R} and is bounded. You may use \displaystyle \displaystyle\int_0^\infty \frac{ds}{1 + s^4} = \frac{\pi}{2\sqrt{2}}.

Equilibria

  1. For y' = 5|y|^{4/5}, show that y = x^5 and y = 0 both solve y(0) = 0 on \mathbb{R}, find a third solution, and say which part of Result 3 allows this.
  2. For each h, decide by Result 3 whether a solution of y' = h(y) with values in \displaystyle \left(0, \frac{1}{2}\right) can also take the value 0: (a) h(y) = |y|^{3/2}; (b) h(y) = y\ln^2|y| for y \ne 0, h(0) = 0.
  3. Solve y' = y\ln^2 y, y(0) = e^{-1}, on the interval where 0 < y < 1. With h(0) = 0 as in Exercise 8, show that the function equal to 0 for x \le -1 and to your solution for x > -1 solves the equation on \mathbb{R}.

Substitutions

  1. Solve xy' = y + 2xe^{-y/x}, y(1) = 0, and find the maximal interval.
  2. Solve y' = \dfrac{x^2 + y^2}{xy}, y(1) = 3, and find the maximal interval.
  3. Solve y' = e^{x + y} - 1, y(0) = 0, and find the maximal interval.

Proofs

  1. Prove Result 4(b).
  2. Let h be continuous, differentiable at c, with h(c) = 0 and no zero in (c, c + \delta]. Prove that \displaystyle \displaystyle\int_c^{c + \delta} \frac{ds}{|h(s)|} = \infty.
  3. Let g be continuous on \mathbb{R} and h continuous and never 0 on \mathbb{R}, with \displaystyle \int_0^{\infty} \frac{ds}{|h(s)|} and \displaystyle \int_{-\infty}^{0} \frac{ds}{|h(s)|} both infinite. Prove that every initial value problem for y' = g(x)\,h(y) has a solution on \mathbb{R} of which every solution is a restriction.
  4. Refute: “If g and h are continuous and h(y_0) \ne 0, the initial value problem has exactly one maximal solution.” Use y' = -4|y|^{3/4}, y(0) = 1.
  5. Let F be continuous on V, with F(v) - v Lipschitz on V and F(v_0) = v_0. Prove that a solution of y' = F\!\left(\dfrac{y}{x}\right) on an interval K \subseteq (0, \infty), with \dfrac{y}{x} \in V, that meets the line y = v_0x at one point is that line on all of K.
Answers
  1. (a) Separable: with (a, b) = (0, 0), \dfrac{(6 - 2x)(6 - 3y)}{6} = (x - 3)(y - 2) = f. (b) Not separable: f(1, 1)f(0, 0) = 2 but f(1, 0)f(0, 1) = 1. (c) Separable: with \displaystyle (a, b) = \left(\frac{\pi}{2}, 0\right), f(x, 0) = 2\sin x, \displaystyle f\left(\frac{\pi}{2}, y\right) = 2\cos y, and f = 2\sin x\cos y by the sum formulas.
  2. f = 2^x\left(2^y + 2^{-y}\right). With u = 2^y, \displaystyle \int \frac{dy}{2^y + 2^{-y}} = \frac{\arctan 2^y}{\ln 2}, so \displaystyle \arctan 2^y = 2^x + \frac{\pi}{4} - 1 and \displaystyle y = \log_2\tan\left(2^x + \frac{\pi}{4} - 1\right). H(J_0) requires \displaystyle 0 < 2^x + \frac{\pi}{4} - 1 < \frac{\pi}{2}, so x runs from \displaystyle \log_2\left(1 - \frac{\pi}{4}\right) to \displaystyle \log_2\left(1 + \frac{\pi}{4}\right): the interval (-2.2203, 0.8362) to four places.
  3. y = \tan(e^x - 1) for \displaystyle x < \ln\left(1 + \frac{\pi}{2}\right) \approx 0.9442. Here \displaystyle G = e^x - 1 > -1 > -\frac{\pi}{2} always, and \displaystyle G < \frac{\pi}{2} exactly when \displaystyle x < \ln\left(1 + \frac{\pi}{2}\right).
  4. y = -\ln\cos x on \displaystyle \left(-\frac{\pi}{2}, \frac{\pi}{2}\right): H(y) = 1 - e^{-y} maps \mathbb{R} onto (-\infty, 1), and G = 1 - \cos x < 1 exactly when \cos x > 0.
  5. y = 2\tanh(x^2), tending to 2 as x \to \pm\infty. \displaystyle H(y) = \frac{1}{4}\ln\frac{2 + y}{2 - y} maps (-2, 2) onto \mathbb{R}, so K^* = \mathbb{R}, and \displaystyle H(y) = \frac{x^2}{2} gives \displaystyle \frac{2 + y}{2 - y} = e^{2x^2}.
  6. J_0 = \mathbb{R} and \displaystyle H(J_0) = \left(-\frac{\pi}{2\sqrt{2}}, \frac{\pi}{2\sqrt{2}}\right), with \displaystyle \frac{\pi}{2\sqrt{2}} \approx 1.1107 > 1 \ge |\sin x| = |G(x)|. So K^* = \mathbb{R}, and since H^{-1} is increasing, H^{-1}(-1) \le y \le H^{-1}(1).
  7. (x^5)' = 5x^4 = 5|x^5|^{4/5}. A third: 0 for x \le 1 and (x - 1)^5 for x > 1. \displaystyle \int_0^1 \frac{ds}{5s^{4/5}} = 1 is finite, so Result 3(b) applies, on both sides of 0.
  8. (a) No: \displaystyle \int_\varepsilon^{1/2} s^{-3/2}\,ds = \frac{2}{\sqrt{\varepsilon}} - 2\sqrt{2} tends to \infty, so Result 3(a) applies. (b) Yes: \displaystyle -\frac{1}{\ln s} is an antiderivative, so the integral over \displaystyle \left(0, \frac{1}{2}\right] is \displaystyle \frac{1}{\ln 2}, finite, and Result 3(b) gives a solution leaving 0.
  9. y = e^{-1/(x + 1)} on (-1, \infty): \displaystyle \int \frac{dy}{y\ln^2 y} = -\frac{1}{\ln y} = x + C with C = 1. For x \le -1, 0 solves. At -1 the left derivative is 0 and the right one is \displaystyle \lim_{x \to -1^+} \frac{e^{-1/(x + 1)}}{x + 1} = 0 = h(0).
  10. y = x\ln(1 + 2\ln x) on \left(e^{-1/2}, \infty\right): F(v) = v + 2e^{-v}, xv' = 2e^{-v}, e^v = 1 + 2\ln x, which must be positive.
  11. \displaystyle y = x\sqrt{9 + 2\ln x} on \left(e^{-9/2}, \infty\right): \displaystyle F(v) - v = \frac{1}{v}, \displaystyle vv' = \frac{1}{x}, v^2 = 9 + 2\ln x; at x = e^{-9/2}, y = 0, where the equation is undefined.
  12. y = -x - \ln(1 - x) on (-\infty, 1): v = x + y gives v' = e^v, -e^{-v} = x - 1.
  13. Since b \ne 0, v = ax + by + k is differentiable exactly when y is, and v' = a + by'. If y' = F(v) then v' = a + bF(v). Conversely, if v' = a + bF(v), then \displaystyle y = \frac{v - ax - k}{b} has \displaystyle y' = \frac{v' - a}{b} = F(ax + by + k). \square
  14. Since \displaystyle \frac{h(s) - h(c)}{s - c} \to h'(c), some \eta \in (0, \delta] has |h(s)| \le L(s - c) on (c, c + \eta] with L = |h'(c)| + 1. Then \displaystyle \int_{c + \varepsilon}^{c + \eta} \frac{ds}{|h(s)|} \ge \frac{1}{L}\ln\frac{\eta}{\varepsilon}, which tends to \infty. \square
  15. Here J_0 = \mathbb{R}, and h has one sign, say positive. Then \displaystyle H(y) = \int_{y_0}^y \frac{ds}{h(s)} \to \pm\infty as y \to \pm\infty, since each tail integral diverges and the part between 0 and y_0 is finite. So H(J_0) = \mathbb{R}, K^* = \mathbb{R}, and \varphi^* solves on \mathbb{R}. Every solution has values in J_0 = \mathbb{R}, so it is a restriction of \varphi^* by Result 2(b). \square
  16. On y > 0, \displaystyle \int_1^y \frac{ds}{-4s^{3/4}} = 1 - y^{1/4} = x gives y = (1 - x)^4 for x < 1, reaching 0 at x = 1; \displaystyle \int_0^1 \frac{ds}{4s^{3/4}} = 1 is finite. Two different solutions on \mathbb{R} continue it: 0 for x > 1, and -(x - 1)^4 for x > 1, whose derivative is -4(x - 1)^3 = -4|y|^{3/4}. Both join with derivative 0 at x = 1, so the claim is false.
  17. By Result 4(a), \displaystyle v = \frac{y}{x} solves \displaystyle v' = \frac{1}{x}\big(F(v) - v\big), with \displaystyle \frac{1}{x} continuous on (0, \infty) and h = F(v) - v Lipschitz on V. At the meeting point v = v_0, a zero of h, so v is not confined to a zero-free interval; by Result 3(c) it is the equilibrium v = v_0, and y = v_0x on K. \square
math.wiki · CC-BY-SA-4.0