math.wiki
Calculus › Differential equations › First-order equations › Direction fields and solution curves
Sign in

Direction fields and solution curves

The equation y’ = f(x, y) names a slope at every point of the plane. Which curves fit those slopes, and what can be proved about them before any formula for a solution is found?

Where you are 6 in First-order equations

Why this exists

Slope fields drew short segments of slope f(x, y) and sketched curves along them by eye. A sketch suggests; it does not prove. The equation y' = 4x - y^2 has no solution formula in the functions of calculus, yet a short argument proves that every solution lying between \displaystyle 2\sqrt{x} - 1 and \displaystyle 2\sqrt{x} at some x \ge 1 stays there while it exists (Example 2), and another that the solution of y' = x^2 + y^2 with y(0) = 1 cannot be continued to x = 1 (Example 4). This lesson says exactly which curves fit a field, introduces fences, which solutions cross in one direction only, and states the no-crossing theorem the rest of the course relies on.

Definition

In plain words: a direction field attaches a direction to every point, and a solution curve runs, at each of its points, in the direction attached there.

Throughout, y is a function of x. Let D be an open set in the plane and f : D \to \mathbb{R}. A solution of the first-order equation

y' = f(x, y)

on an open interval I is a differentiable \varphi : I \to \mathbb{R} such that for every x \in I the point (x, \varphi(x)) lies in D and \varphi'(x) = f(x, \varphi(x)). Its graph is a solution curve.

The direction field assigns to each (x, y) \in D the line through it with slope f(x, y); a picture draws a short piece of it at grid points. The Calculus lessons call it the slope field.

An integral curve is a curve t \mapsto (x(t), y(t)), t in an open interval, lying in D, with continuously differentiable coordinates and a velocity that is never zero and is at every t a multiple of (1, f(x(t), y(t))): its tangent line at each point is the line of the field.

An initial value problem adds a condition y(x_0) = y_0 with (x_0, y_0) \in D; its solutions are solutions on open intervals containing x_0 with \varphi(x_0) = y_0. A solution is maximal if it is not the restriction of a solution on a larger interval.

Fences. Let a < b \le \infty. A lower fence on [a, b) is a function \alpha, differentiable on [a, b) (one-sided at a), with graph in D and \alpha'(x) < f(x, \alpha(x)) for a < x < b. An upper fence \beta has \beta'(x) > f(x, \beta(x)) for a < x < b. Finally, \dfrac{\partial f}{\partial y} is the derivative of f in y with x held fixed (Partial derivatives).

Intuition

A solution curve obeys the field at every point. Every slope f(x, y) is finite, so it never turns vertical and never turns back: it is a graph. A closed curve such as an ellipse can run along a field, but its points with vertical tangent lie outside D, and it splits into separate solution curves (Example 1). Euler’s method (Euler's method) obeys the field only at the corners of a broken line, and between corners can do what no solution can (Example 3).

Along a lower fence the field is steeper than the fence, so a solution curve touching it leaves upward: solutions pass a lower fence upward and never downward. An upper fence is the mirror image, and a lower fence below an upper fence makes a funnel that solutions enter but never leave (Figure 2). When f is smooth enough, two solution curves never meet at all, so the order of solutions at one x is their order at every x.

Results

1. Solution curves are integral curves. Let f be continuous on D. Every solution is continuously differentiable, and x \mapsto (x, \varphi(x)) is an integral curve. Conversely, every integral curve is, traced by x, the curve of a solution on an open interval.

2. Fence theorem. Let \varphi be a solution on an open interval J containing a, and \alpha a lower fence on [a, b). If either

  • (i) \alpha(a) < \varphi(a), or
  • (ii) \alpha(a) = \varphi(a) and \alpha'(a) < f(a, \varphi(a)),

then \alpha(x) < \varphi(x) for every x \in J with a < x < b. For an upper fence, with every inequality reversed, \varphi(x) < \beta(x) there. No continuity of f is assumed.

3. Existence and uniqueness (proved in Existence and uniqueness). If f and \dfrac{\partial f}{\partial y} are continuous on D and (x_0, y_0) \in D, the initial value problem y' = f(x, y), y(x_0) = y_0 has a solution, and any two of its solutions agree on some open interval containing x_0.

4. No crossing. Under the hypotheses of Result 3, if \varphi_1 and \varphi_2 are solutions on the same open interval I and \varphi_1(x_0) = \varphi_2(x_0) for one x_0 \in I, then \varphi_1 = \varphi_2 on I.

5. Order is kept. Under the same hypotheses, if \varphi_1(x_0) < \varphi_2(x_0) for one x_0 \in I, then \varphi_1(x) < \varphi_2(x) for every x \in I.

Why it is true

Result 1. Proof. A solution is continuous, so \varphi'(x) = f(x, \varphi(x)) is continuous, and the velocity of (x, \varphi(x)) is (1, f(x, \varphi(x))), never zero. Conversely, let an integral curve have velocity \lambda(t)\,(1, f(x(t), y(t))). Then \lambda = x', which is continuous and never 0, since otherwise the velocity would vanish. By the intermediate value theorem x' keeps one sign, so x is strictly monotone and maps the parameter interval onto an open interval I, with a differentiable inverse \tau (Derivatives of inverse functions). For \varphi = y \circ \tau the chain rule gives

\varphi'(x) = \frac{y'(\tau(x))}{x'(\tau(x))} = \frac{\lambda f}{\lambda} = f(x, \varphi(x)),

with \lambda and f taken at t = \tau(x). \square

Result 2. Proof. Put \psi = \varphi - \alpha. First, \psi > 0 on some (a, a + \delta): in case (i) by continuity; in case (ii) because \psi(a) = 0 and \psi'(a) > 0, so \dfrac{\psi(x)}{x - a} > 0 for x just right of a.

Suppose \psi(x) \le 0 for some x \in J with a < x < b, and let x_1 be the infimum of such x. Then x_1 \ge a + \delta, \psi > 0 on (a, x_1), and by continuity \psi(x_1) = 0, so \varphi(x_1) = \alpha(x_1) and the fence inequality gives

\psi'(x_1) = f(x_1, \alpha(x_1)) - \alpha'(x_1) > 0.

But for a < x < x_1 the quotient \dfrac{\psi(x) - \psi(x_1)}{x - x_1} = \dfrac{\psi(x)}{x - x_1} is negative, so its limit \psi'(x_1) is at most 0, a contradiction. The upper fence statement follows by applying this to -\varphi (Exercise 10). \square

The strict inequality did the work. A second solution meets the fence condition only with equality, and comparing two solutions needs Result 3.

Result 4. Proof. Let s be the supremum of the x \ge x_0 in I with \varphi_1 = \varphi_2 on [x_0, x]. If s were in I, continuity would give \varphi_1(s) = \varphi_2(s); both would solve the initial value problem at (s, \varphi_1(s)) \in D, and Result 3 would make them agree on [x_0, s + \varepsilon] for some \varepsilon > 0, against the choice of s. The same argument with an infimum handles x \le x_0. \square

Result 5. Proof. If \varphi_1(x_1) \ge \varphi_2(x_1) for some x_1 \in I, then g = \varphi_2 - \varphi_1 has g(x_0) > 0 \ge g(x_1), so g(x_2) = 0 at some x_2 between them by the intermediate value theorem, and Result 4 gives \varphi_1 = \varphi_2 on I, against g(x_0) > 0. \square

Worked examples

Example 1

Ellipses that are two solutions each

For y' = -\dfrac{4x}{y} on D = \{(x, y) : y \ne 0\}, show that every solution curve lies on an ellipse 4x^2 + y^2 = r^2, and find the solution with y(0) = 2 and its maximal interval.

Solution

D is the plane without the x-axis. If \varphi is a solution, then

\begin{aligned} \frac{d}{dx}\left(4x^2 + \varphi^2\right) &= 8x + 2\varphi\varphi' \\ &= 8x - 8x = 0, \end{aligned}

so 4x^2 + \varphi(x)^2 is constant. Conversely, implicit differentiation (Implicit differentiation) of 4x^2 + y^2 = r^2 gives y' = -\dfrac{4x}{y} wherever y \ne 0, so off the x-axis each ellipse runs along the field (Figure 1).

−1 1 −3 −2 −1 1 2 3 x y
Figure 1The field of y' = -\dfrac{4x}{y} and the ellipses 4x^2 + y^2 = r^2 for r = 1, 2, 3 (Example 1). The field has no segment on the x-axis, so each ellipse is two solution curves, an upper and a lower half; the dot is the initial point (0, 2).

Let \psi solve the problem with \psi(0) = 2 on an interval J. Then 4x^2 = 4 - \psi(x)^2 < 4 for each x \in J, since \psi(x) \ne 0, so J \subseteq (-1, 1); and \psi is continuous, never 0 and positive at 0, so positive, which forces \displaystyle \psi(x) = 2\sqrt{1 - x^2}. This function does solve the problem on (-1, 1):

\begin{aligned} \varphi'(x) &= \frac{-2x}{\sqrt{1 - x^2}}, \\ -\frac{4x}{\varphi(x)} &= \frac{-4x}{2\sqrt{1 - x^2}}. \end{aligned}

So the solution is unique and (-1, 1) is its maximal interval. As x \to 1^-, \varphi \to 0 and \varphi' \to -\infty: the curve runs into the edge of D where the ellipse’s tangent turns vertical. The lower half is the solution with y(0) = -2; the ellipse’s two points on the x-axis belong to no solution.

Example 2

A funnel for an equation with no formula

Amara and Kofi are printing the direction field of y' = 4x - y^2 as a poster for their makerspace and want to label it with something proved. Show that \displaystyle \beta(x) = 2\sqrt{x} is an upper fence and \displaystyle \alpha(x) = 2\sqrt{x} - 1 a lower fence on [a, \infty) for every a \ge 1, and describe the long run of a solution with \displaystyle 2\sqrt{a} - 1 \le y(a) \le 2\sqrt{a}.

Solution

The isocline of slope 0 is y^2 = 4x, whose upper half is \beta. Along it \displaystyle f(x, \beta(x)) = 0 < \dfrac{1}{\sqrt{x}} = \beta'(x) for every x > 0. For \alpha,

\begin{aligned} f(x, \alpha(x)) &= 4x - \left(4x - 4\sqrt{x} + 1\right) \\ &= 4\sqrt{x} - 1, \end{aligned}

and \displaystyle \alpha'(x) = \dfrac{1}{\sqrt{x}}. Multiplying by \displaystyle \sqrt{x} > 0, the inequality \alpha' < f(x, \alpha) becomes \displaystyle 1 < 4x - \sqrt{x}, and for x \ge 1, \displaystyle 4x - \sqrt{x} \ge 3x \ge 3. Both fence inequalities hold at x = a as well.

So for a solution \varphi on J \ni a starting in the funnel, Result 2 applies to each fence, through case (i), or case (ii) if \varphi starts on that fence:

\begin{aligned} &2\sqrt{x} - 1 < \varphi(x) < 2\sqrt{x} \\ &\text{for every } x > a \text{ in } J. \end{aligned}

If J contains [a, \infty), dividing by \displaystyle 2\sqrt{x} shows \displaystyle \dfrac{\varphi(x)}{2\sqrt{x}} \to 1: every such solution grows like \displaystyle 2\sqrt{x}. (It does exist on [a, \infty): a solution that stays bounded over a bounded interval can be continued, a fact stated here without proof that rests on the existence theorem of Existence and uniqueness.)

1 2 3 4 1 2 3 4 x y
Figure 2The funnel of Example 2 for y' = 4x - y^2: the dashed curves are the upper fence \displaystyle y = 2\sqrt{x} and the lower fence \displaystyle y = 2\sqrt{x} - 1, drawn for x \ge 1. Solutions through (1, 1) and (1, 2) stay inside; the one through (1.5, 4) crosses the upper fence downward, the one direction it allows, and is then trapped.

The solution through (1.5, 4) starts with slope 6 - 16 = -10, falls through the upper fence, which an upper fence allows, and is trapped from then on.

Example 3

Kofi’s polygon jumps the fence

Kofi checks the poster with Euler’s method, h = 0.5, from (1, 1.5) to x = 3 on y' = 4x - y^2. Compare each point with the funnel of Example 2.

Solution

Each step is y_{n+1} = y_n + 0.5\,f(x_n, y_n); the funnel at x runs from \displaystyle 2\sqrt{x} - 1 to \displaystyle 2\sqrt{x}.

x_ny_nslopefunnel
11.51.751 to 2
1.52.3750.35941.4495 to 2.4495
22.55471.47361.8284 to 2.8284
2.53.2915-0.83382.1623 to 3.1623
32.87462.4641 to 3.4641

At x = 2.5 the polygon is above the upper fence, where by Example 2 no solution from the funnel can be; the solution itself, traced by an accurate numerical solver, has y(2.5) \approx 3.052 (Figure 3).

1.5 2 2.5 3 1 2 3 x y
Figure 3Euler’s method with h = 0.5 from (1, 1.5) (Example 3): the thin broken line with dots. At x = 2.5 it lies above the upper fence \displaystyle y = 2\sqrt{x}, where no solution from inside the funnel can go; the thick curve is the solution itself.

From (2, 2.5547) the polygon keeps the slope 1.4736 for the whole step, while the field’s slope drops to 0 on the fence and is negative above it. The proof of Result 2 used \varphi'(x_1) = f(x_1, \varphi(x_1)) at the meeting point, and the polygon’s slope there was computed at an earlier point. Smaller steps bring the polygon closer to the solution (Euler's method), but a fence is a statement about solutions, not polygons.

Example 4

A solution that becomes unbounded early

Let \varphi solve y' = x^2 + y^2, y(0) = 1, on an open interval J containing 0. Show that \varphi(x) \ge \dfrac{1}{1 - x} for x \in J with 0 \le x < 1, and that J does not contain 1.

Solution

For c > 1 put \alpha_c(x) = \dfrac{1}{c - x} on [0, c), a solution of y' = y^2. For 0 < x < c,

\alpha_c'(x) = \alpha_c(x)^2 < x^2 + \alpha_c(x)^2,

so \alpha_c is a lower fence, and \alpha_c(0) = \dfrac{1}{c} < 1 = \varphi(0). Case (i) of Result 2 gives \varphi(x) > \dfrac{1}{c - x} for 0 < x < c in J; letting c \to 1^+ gives \varphi(x) \ge \dfrac{1}{1 - x}. If J contained 1, then \varphi would be bounded on [0, 1], while \dfrac{1}{1 - x} \to \infty as x \to 1^-. So the maximal interval ends at or before 1 (Figure 4); an accurate numerical solver puts the end near x \approx 0.970.

0.5 1 2 4 6 8 x y
Figure 4The solution of y' = x^2 + y^2, y(0) = 1 (solid) and the curve y = \dfrac{1}{1 - x} (dashed), which it stays above (Example 4). The solution becomes unbounded before the dashed vertical line x = 1.

The curve \dfrac{1}{1 - x} itself would not do as the fence: it starts at \varphi(0) = 1 with slope 1 = f(0, 1), so neither case of Result 2 holds. The family with c > 1 starts strictly below and avoids the tie.

Example 5

Yuna’s proof: staying above a line

Yuna studies y' = 1 - x^2 + y^2. Show that y = x is a solution, and prove that the solution \varphi with \varphi(0) = \tfrac{1}{2}, on any open interval J containing 0, satisfies \varphi(x) > x for every x \in J and \varphi(x) > x + \tfrac{1}{2} for every x > 0 in J.

Solution

For y = x: y' = 1 = 1 - x^2 + x^2. Since f and \dfrac{\partial f}{\partial y} = 2y are continuous on the plane, Result 5 applies to the line and \varphi on J, and \varphi(0) = \tfrac{1}{2} > 0 gives \varphi(x) > x on J. The fence theorem could not: the line meets the fence condition with equality.

For x \ge 0 in J, \varphi(x) > x \ge 0 gives \varphi(x)^2 > x^2, so \varphi'(x) > 1. For x > 0 the mean value theorem (The mean value theorem and its consequences) gives \xi \in (0, x) with \varphi(x) - \varphi(0) = \varphi'(\xi)\,x > x, so \varphi(x) > x + \tfrac{1}{2}. \square Exercise 13 shows this solution cannot reach x = 2.

Boundaries

  • A fence must be strict. For \displaystyle y' = -3\sqrt[3]{y^2} the constant \alpha = 0 has \alpha' = 0 = f(x, 0), and \varphi(x) = (1 - x)^3 is a solution, since \displaystyle \varphi' = -3(1 - x)^2 = -3\sqrt[3]{(1 - x)^6}. It starts above \alpha and crosses it at (1, 0), where \displaystyle \dfrac{\partial f}{\partial y} = -\dfrac{2}{\sqrt[3]{y}} does not exist: two solutions, 0 and (1 - x)^3, pass through that point.
  • Fences trap; they do not keep a solution alive. That a solution exists across a whole funnel is the separate continuation fact of Example 2.
  • Results 4 and 5 need \dfrac{\partial f}{\partial y} continuous on all of D and compare solutions only where both exist.
  • A polygon is not a solution (Example 3), and an integral curve stops at the edge of D (Example 1).

Where this goes next

Separable equations in general and Linear equations and integrating factors find formulas for two large classes of equations, each readable against its field. Existence and uniqueness proves Result 3 by Picard’s iteration, completing the proofs of Results 4 and 5. Autonomous equations and stability treats fields that repeat along horizontal lines, where Result 5 makes every bounded solution monotone. Phase portraits draw direction fields for systems of two equations, where solution curves again never cross.

Exercises

Solutions and integral curves

  1. For y' = \dfrac{x}{y} on D = \{(x, y) : y \ne 0\}, show that every solution curve lies on a curve y^2 - x^2 = C. Find the solution with y(2) = 1 and its maximal interval, and the solution with y(0) = -3 and its maximal interval.
  2. For y' = \dfrac{2y}{x} on D = \{(x, y) : x > 0\}, show that y = Cx^2 is a solution on (0, \infty) for every constant C, and use Result 4 to show that every solution through (2, 2) agrees with y = \dfrac{x^2}{2} wherever it is defined.
  3. Take three Euler steps of h = 0.5 from (0, 2) for y' = -\dfrac{4x}{y}, the equation of Example 1. Where does the polygon end, and why can no solution through (0, 2) be followed that far?

Fences and funnels

  1. For y' = x - 3y, show that \beta(x) = \dfrac{x}{3} is an upper fence and \alpha(x) = \dfrac{x - 1}{3} a lower fence on [0, \infty). Verify that y = \dfrac{x}{3} - \dfrac{1}{9} + Ce^{-3x} is a solution, find C for y(0) = -\dfrac{1}{6}, and check that this solution stays between the fences for x > 0.
  2. For y' = y^2 - x: (a) show that \displaystyle -\sqrt{x} is a lower fence on [a, \infty) for every a > 0; (b) find the x > 0 at which \displaystyle \beta(x) = 1 - \sqrt{x} satisfies \beta'(x) > f(x, \beta(x)); (c) for the solution with y(1) = -\tfrac{1}{2}, conclude that \displaystyle -\sqrt{x} < y(x) < 1 - \sqrt{x} for x > 1 in its interval, and find the limit of \displaystyle \dfrac{y(x)}{\sqrt{x}} if it exists on [1, \infty).
  3. For y' = 4x - y^2 of Examples 2 and 3, take two Euler steps of h = 0.5 from (1, 2). Compare the points with the funnel, and say what the solution through (1, 2) does instead.

Uniqueness and order

  1. For y' = (y - x)^2 + 1, show that y = x is a solution. Without solving, explain why the solution with y(0) = -1 stays below the line y = x. Then verify that y = x - \dfrac{1}{x + 1} is this solution, and give its maximal interval.
  2. For \displaystyle y' = 2\sqrt{|y|} on the whole plane, show that y = 0 and the function equal to 0 for x \le 0 and to x^2 for x > 0 both solve the initial value problem with y(0) = 0 on \mathbb{R}. Find a third solution, and say which hypothesis of Result 3 fails.
  3. Find every line y = mx + k that is a solution curve of y' = y^2 - 4x^2 + 2. Then show that the solution with y(0) = -1 satisfies y(x) < 2x on its whole interval.

Proofs

  1. Prove the upper fence half of Result 2 from the lower fence half, by applying it to -\varphi and a new equation.
  2. Let f be defined on the whole plane with f(x, y) \ge 2 everywhere, and let \varphi solve y' = f(x, y), y(0) = 0, on an open interval containing [0, \infty). Prove that \varphi(x) \ge 2x for x \ge 0, so that \varphi(x) \to \infty as x \to \infty.
  3. Let \varphi solve y' = 1 + x + y^2, y(0) = 0, on an open interval J containing 0. Prove that J contains no point x \ge \dfrac{\pi}{2}.
  4. For the solution \varphi of Example 5, put u = \varphi - x. Show that u' = u^2 + 2xu, and prove that the interval of \varphi contains no point x \ge 2.
  5. Let f and \dfrac{\partial f}{\partial y} be continuous on the whole plane, and let c_1 < c_2 be constants with f(x, c_1) = f(x, c_2) = 0 for every x. Prove that a solution with c_1 < \varphi(x_0) < c_2 at one point satisfies c_1 < \varphi(x) < c_2 on its whole interval.
  6. Let f and \dfrac{\partial f}{\partial y} be continuous on the whole plane with f(x, -y) = -f(x, y). Prove that y = 0 is a solution, that -\varphi is a solution whenever \varphi is, and that a solution positive at one point is positive on its whole interval.
Answers
  1. Along a solution, \dfrac{d}{dx}\left(\varphi^2 - x^2\right) = 2\varphi\varphi' - 2x = 0. Through (2, 1): C = -3 and \displaystyle y = \sqrt{x^2 - 3}, with \displaystyle y' = \dfrac{x}{\sqrt{x^2 - 3}} = \dfrac{x}{y}. On any solution through (2, 1), x^2 = 3 + y^2 > 3 and the interval contains 2, so the maximal interval is \displaystyle (\sqrt{3}, \infty); at \displaystyle \sqrt{3} the curve reaches y = 0, outside D. Through (0, -3): C = 9, \displaystyle y = -\sqrt{x^2 + 9}, never 0, maximal interval \mathbb{R}.
  2. y' = 2Cx = \dfrac{2Cx^2}{x}. Here f = \dfrac{2y}{x} and \dfrac{\partial f}{\partial y} = \dfrac{2}{x} are continuous on D, so a solution through (2, 2) on I \subseteq (0, \infty) and \dfrac{x^2}{2} restricted to I agree at 2, hence on I by Result 4.
  3. Slopes 0, -1, -\dfrac{8}{3} give the points (0.5, 2), (1, 1.5) and \left(1.5, \dfrac{1}{6}\right). Every solution through (0, 2) lives inside (-1, 1) (Example 1). The polygon follows the field only at its corners, and (1, 1.5) is already off the ellipse 4x^2 + y^2 = 4 (there 4x^2 + y^2 = 6.25), so nothing stops it at x = 1.
  4. f(x, \beta) = 0 < \tfrac{1}{3} = \beta' and f(x, \alpha) = 1 > \tfrac{1}{3} = \alpha'. Both y' and x - 3y equal \tfrac{1}{3} - 3Ce^{-3x}. C = -\tfrac{1}{6} + \tfrac{1}{9} = -\dfrac{1}{18}. Then \beta - y = \tfrac{1}{9} + \tfrac{1}{18}e^{-3x} > 0 and y - \alpha = \tfrac{2}{9} - \tfrac{1}{18}e^{-3x} > 0 for x \ge 0.
  5. (a) \displaystyle f(x, -\sqrt{x}) = 0 > -\dfrac{1}{2\sqrt{x}}. (b) \displaystyle f(x, 1 - \sqrt{x}) = 1 - 2\sqrt{x} and \displaystyle \beta' = -\dfrac{1}{2\sqrt{x}}; with \displaystyle s = \sqrt{x} the condition is 4s^2 - 2s - 1 > 0, so \displaystyle s > \dfrac{1 + \sqrt{5}}{4}, that is \displaystyle x > \dfrac{3 + \sqrt{5}}{8} \approx 0.6545. (c) Both fences hold on [1, \infty) and -1 < -\tfrac{1}{2} < 0, so case (i) applies to each; dividing by \displaystyle \sqrt{x}, \displaystyle -1 < \dfrac{y}{\sqrt{x}} < \dfrac{1}{\sqrt{x}} - 1, and the limit is -1.
  6. Slope 0 gives (1.5, 2), inside the funnel; slope 2 gives (2, 3), above it, since \displaystyle 2\sqrt{2} \approx 2.8284. The solution starts on the upper fence with slope 0 < \beta'(1) = 1, so by case (ii) it satisfies \displaystyle 2\sqrt{x} - 1 < y(x) < 2\sqrt{x} for every x > 1 in its interval.
  7. y' = 1 = (x - x)^2 + 1. Since f and \dfrac{\partial f}{\partial y} = 2(y - x) are continuous and -1 < 0, Result 5 keeps the solution below y = x. Check: y' = 1 + \dfrac{1}{(x + 1)^2} = (y - x)^2 + 1 and y(0) = -1. The maximal interval is (-1, \infty), since y \to -\infty as x \to -1^+.
  8. For x > 0, \displaystyle (x^2)' = 2x = 2\sqrt{x^2}; for x < 0 both sides are 0, and both one-sided derivatives at 0 are 0. A third: 0 for x \le 1 and (x - 1)^2 for x > 1. Result 3 needs \dfrac{\partial f}{\partial y} continuous, but at y = 0 the quotient \displaystyle \dfrac{2\sqrt{|k|}}{k} is unbounded, so the partial derivative does not exist.
  9. The equation needs m = (mx + k)^2 - 4x^2 + 2 for all x: m^2 = 4, 2mk = 0, k^2 + 2 = m, so k = 0, m = 2. The only line is y = 2x. With f and \dfrac{\partial f}{\partial y} = 2y continuous and -1 < 0, Result 5 gives y(x) < 2x.
  10. Let \beta be an upper fence and \varphi(a) < \beta(a), or \varphi(a) = \beta(a) with \beta'(a) > f(a, \varphi(a)). Put g(x, y) = -f(x, -y) on the reflection of D, \psi = -\varphi, \alpha = -\beta. Then \psi' = -f(x, -\psi) = g(x, \psi), and \alpha' = -\beta' < -f(x, -\alpha) = g(x, \alpha) for a < x < b, so \alpha is a lower fence for g. At a, \alpha(a) < \psi(a), or \alpha(a) = \psi(a) and \alpha'(a) < g(a, \psi(a)). The lower fence theorem gives -\beta < -\varphi, that is \varphi < \beta.
  11. For x > 0 the mean value theorem gives \xi \in (0, x) with \varphi(x) = \varphi(0) + f(\xi, \varphi(\xi))\,x \ge 2x; at x = 0 both sides are 0. Since 2x \to \infty, so does \varphi(x).
  12. Otherwise J \supseteq \left[0, \dfrac{\pi}{2} + \delta\right] for some 0 < \delta < \dfrac{\pi}{2}. Put \alpha(x) = \tan(x - \delta) on \left[0, \dfrac{\pi}{2} + \delta\right). Then \alpha' = 1 + \alpha^2 < 1 + x + \alpha^2 for x > 0, and \alpha(0) = -\tan\delta < 0, so case (i) gives \varphi(x) > \tan(x - \delta) there. The right side is unbounded as x \to \left(\dfrac{\pi}{2} + \delta\right)^-, but \varphi is bounded on \left[0, \dfrac{\pi}{2} + \delta\right]: a contradiction.
  13. u' = \varphi' - 1 = \varphi^2 - x^2 = u(u + 2x) = u^2 + 2xu, with u(0) = \tfrac{1}{2} and u > 0 by Example 5. If the interval contained a point x \ge 2, it would contain [0, c] for some c > 2. Then \alpha(x) = \dfrac{1}{c - x} has \alpha' = \alpha^2 < \alpha^2 + 2x\alpha for x > 0 and \alpha(0) = \dfrac{1}{c} < \dfrac{1}{2}, so Result 2 for y' = y^2 + 2xy gives u > \dfrac{1}{c - x} on (0, c), unbounded near c, while u is bounded on [0, c]: a contradiction.
  14. The constants c_1 and c_2 are solutions on \mathbb{R}, since 0 = f(x, c_i). On the interval of \varphi, Result 5 applied to c_1 and \varphi, then to \varphi and c_2, gives c_1 < \varphi(x) < c_2 throughout.
  15. f(x, 0) = -f(x, 0) forces f(x, 0) = 0, so 0 is a solution. (-\varphi)' = -f(x, \varphi) = f(x, -\varphi), so -\varphi is a solution. If \varphi(x_0) > 0, Result 5 with the zero solution gives \varphi > 0 on the whole interval.
math.wiki · CC-BY-SA-4.0