math.wiki
Calculus › Differential equations › First-order equations › Linear equations and integrating factors
Sign in

Linear equations and integrating factors

The equation y’ + p(x)y = q(x) is not separable, yet every equation of this form can be solved. What makes its left side integrable, and what does the resulting formula prove about where solutions exist and how they behave?

Where you are 6 in First-order equations

Why this exists

The equation y' = x + y is not separable: its slope is a sum, not a product. Separable equations in general solved it with the substitution v = x + y, a trick tied to the special form. Many models share its shape with variable coefficients, such as a tank whose volume changes or water warming with the time of day: the unknown and its derivative appear only to the first power, multiplied by functions of the independent variable.

Every such equation, y' + p(x)\,y = q(x), yields to one device: multiply by a function that turns the left side into the derivative of a product. The resulting formula proves that an initial value problem has its solution on the whole interval where p and q are continuous, which no nonlinear equation promises (the solution of y' = x^2 + y^2 in Direction fields and solution curves becomes unbounded before x = 1), and it shows how solutions forget their starting value. Throughout, y is a function of x, or of t when the variable is time.

Definition

In plain words: a first-order equation is linear when y and y' appear only to the first power, multiplied by functions of x alone.

Let I be an open interval and p, q : I \to \mathbb{R} continuous. The equation

y' + p(x)\,y = q(x)

is a first-order linear equation in standard form. It is homogeneous if q = 0 on I and nonhomogeneous otherwise (unrelated to the homogeneous equations of Separable equations in general). An equation a(x)\,y' + b(x)\,y = c(x) with continuous a, b, c is also linear. On an interval where a has no zero, dividing by a gives the standard form; a point where a = 0 is a singular point of the equation.

A solution is a differentiable function \varphi on an open interval K \subseteq I with \varphi'(x) + p(x)\,\varphi(x) = q(x) for every x \in K. An initial value problem adds a condition y(x_0) = y_0 with x_0 \in I.

An integrating factor is a differentiable function \mu on I, never 0, with \mu' = p\,\mu. For every differentiable y, the product rule then gives

(\mu y)' = \mu y' + \mu' y = \mu\,(y' + p\,y).

For x_0 \in I put P(x) = \int_{x_0}^{x} p(s)\,ds. By the fundamental theorem, P' = p, so \mu = e^{P} is an integrating factor. Any antiderivative of p in place of P works as well, which is what \mu(x) = e^{\int p(x)\,dx} means; by Exercise 17, every integrating factor is a constant multiple of e^{P}.

Intuition

Multiplying by \mu changes the unknown, not the problem. With u = \mu y the equation becomes u' = \mu q, whose right side does not involve u: one integration solves it, and dividing by \mu returns y.

The direction field shows the shape of the answer. On each vertical line x = c the slope q(c) - p(c)\,y is an affine function of y, so the difference d of two solutions obeys d' = -p(x)\,d, which no longer mentions q. From x_1 to x_2 the gap between two solutions is multiplied by e^{-(P(x_2) - P(x_1))}, whatever the forcing does. Where p > 0 the gap shrinks: this is how a tank of water forgets the temperature it started at (Figure 2).

Results

1. Existence, uniqueness and the formula. Let p and q be continuous on an open interval I, let x_0 \in I and y_0 \in \mathbb{R}, and let P(x) = \int_{x_0}^{x} p(s)\,ds. The initial value problem y' + p(x)\,y = q(x), y(x_0) = y_0, has a solution on all of I,

\begin{aligned} y(x) &= e^{-P(x)}\,y_0 \\ &\quad + e^{-P(x)}\int_{x_0}^{x} e^{P(t)}\,q(t)\,dt, \end{aligned}

and every solution of it on an open interval K \subseteq I containing x_0 is the restriction of this one to K.

2. Structure of the solutions. Let p, q, q_1, q_2 be continuous on I.

  • (a) Superposition: if y_1 solves y' + p\,y = q_1 and y_2 solves y' + p\,y = q_2 on I, then c_1y_1 + c_2y_2 solves y' + p\,y = c_1q_1 + c_2q_2.
  • (b) The solutions of the homogeneous equation y' + p\,y = 0 on I are exactly y = Ce^{-P(x)}, C \in \mathbb{R}. Each is identically 0 or never 0, and together they form a one-dimensional vector space spanned by y_h = e^{-P} (Span).
  • (c) If y_p is one solution of y' + p\,y = q on I, the solutions on I are exactly y = y_p + Ce^{-P(x)}.

3. Solutions forget their start. Let p and q be continuous on an open interval containing [x_0, \infty), with p(x) \ge k > 0 for x \ge x_0. For any two solutions on that interval, the difference d = y_1 - y_2 satisfies, for x \ge x_0,

|d(x)| \le |d(x_0)|\,e^{-k(x - x_0)}.

4. Bernoulli equations. Let p and q be continuous on I and n a real number other than 0 and 1. On an open interval K \subseteq I, a positive function y solves

y' + p(x)\,y = q(x)\,y^n

if and only if v = y^{1 - n} is a positive solution of the linear equation

v' + (1 - n)\,p(x)\,v = (1 - n)\,q(x).

When n > 0, y = 0 is also a solution.

Why it is true

Result 1. Let \mu = e^{P}, so \mu > 0 and \mu' = p\,\mu. If y solves the problem on K, then (\mu y)' = \mu\,(y' + p\,y) = \mu q on K, so the function

\mu(x)\,y(x) - \int_{x_0}^{x} \mu(t)\,q(t)\,dt

has derivative 0 on the interval K and is constant there by the mean value theorem. At x_0 it equals y_0, since P(x_0) = 0. Dividing by \mu(x) gives the formula on K. Conversely, \mu q is continuous on all of I, so the formula defines a differentiable function on I with \mu y = y_0 + \int_{x_0}^{x} \mu q. Hence (\mu y)' = \mu q, that is \mu\,(y' + p\,y) = \mu q, and dividing by \mu > 0 gives the equation; also y(x_0) = y_0. \square

Result 2. (a) Differentiation is linear, so

\begin{aligned} &(c_1y_1 + c_2y_2)' + p\,(c_1y_1 + c_2y_2) \\ &\quad = c_1(y_1' + p\,y_1) + c_2(y_2' + p\,y_2), \end{aligned}

which is c_1q_1 + c_2q_2. (b) With q = 0, Result 1 gives y = y(x_0)\,e^{-P} for every solution, and every Ce^{-P} solves. Since e^{-P} never vanishes, Ce^{-P} is identically 0 when C = 0 and never 0 otherwise. These functions are exactly the scalar multiples of the nonzero function e^{-P}, so they form its span. (c) By (a) with c_1 = 1, c_2 = -1, y - y_p solves the homogeneous equation, so y - y_p = Ce^{-P} by (b); conversely y_p + Ce^{-P} solves by (a). \square

Result 3. By Result 2(a), d solves d' + p\,d = 0, so d(x) = d(x_0)\,e^{-P(x)} by Result 2(b). For x \ge x_0, P(x) = \int_{x_0}^{x} p(s)\,ds \ge k(x - x_0). \square

Result 4. On (0, \infty) the powers s^{1 - n} and s^{1/(1 - n)} are differentiable and inverse to each other, so a positive y is differentiable exactly when v = y^{1 - n} is, and v' = (1 - n)\,y^{-n}y' by the chain rule. If y' = q\,y^n - p\,y, then

v' = (1 - n)\left(q - p\,y^{1 - n}\right) = (1 - n)\,q - (1 - n)\,p\,v.

Conversely, if v solves the linear equation, then, using y^n v = y,

y' = \frac{y^n v'}{1 - n} = y^n(q - p\,v) = q\,y^n - p\,y. \qquad \square

Worked examples

Example 1

The equation the substitution solved

Solve y' = x + y, y(0) = 0, with an integrating factor.

Solution

In standard form, y' - y = x, so p = -1, q = x and I = \mathbb{R}. Here P(x) = -x, and multiplying by \mu = e^{-x} makes the left side a derivative:

\left(e^{-x}y\right)' = xe^{-x}.

Integration by parts gives \int xe^{-x}\,dx = -(x + 1)e^{-x} + C, so e^{-x}y = -(x + 1)e^{-x} + C and y = -x - 1 + Ce^{x}. The condition y(0) = 0 gives C = 1:

y = e^{x} - x - 1, \quad x \in \mathbb{R}.

Check: y' = e^{x} - 1, and x + y = e^{x} - 1. By Result 1 this is the only solution, and it lives on all of \mathbb{R}. In the form of Result 2(c), y_p = -x - 1 and y_h = e^{x}.

Example 2

A singular point

Solve xy' + 2y = 4x^2, y(1) = 2, give the interval Result 1 guarantees, and find every solution defined on all of \mathbb{R}.

Solution

The coefficient of y' vanishes at the singular point x = 0. The initial point x_0 = 1 lies in (0, \infty), where the standard form is

y' + \frac{2}{x}\,y = 4x.

Here \displaystyle p = \frac{2}{x} and q = 4x are continuous on I = (0, \infty). An antiderivative of p is 2\ln x, so \mu = x^2, and (x^2y)' = 4x^3 gives x^2y = x^4 + C, that is y = x^2 + \dfrac{C}{x^2}. From y(1) = 1 + C = 2, C = 1:

y = x^2 + \frac{1}{x^2}, \quad 0 < x < \infty.

This is also the maximal interval, since y \to \infty as x \to 0^+. Check: xy' = 2x^2 - \dfrac{2}{x^2} and 2y = 2x^2 + \dfrac{2}{x^2}, which add to 4x^2.

For a solution on \mathbb{R}, the same computation on (-\infty, 0) gives y = x^2 + \dfrac{C_-}{x^2} there and y = x^2 + \dfrac{C_+}{x^2} on (0, \infty). Continuity at 0 forces C_- = C_+ = 0, and y = x^2 does solve at x = 0, where both sides are 0. So exactly one solution is defined on all of \mathbb{R}, the parabola through the origin (Figure 1). An initial value problem y(0) = y_0 with y_0 \ne 0 has no solution at all, because at x = 0 the equation reads 2y(0) = 0.

−2 −1 1 2 −1 1 2 3 4 5 x y
Figure 1The field of xy' + 2y = 4x^2 and the solutions y = x^2 + \dfrac{C}{x^2} for C = 1, \tfrac{1}{4}, -\tfrac{1}{4}, -1 on each side of the singular line x = 0 (Example 2). Only the parabola y = x^2 (thick, C = 0) crosses that line; the dot at (1, 2) starts the solution of the example.
Example 3

A solution with no elementary formula

Solve y' + 2xy = 1, y(0) = 0, show that the solution is odd, and find \lim_{x \to \infty} y(x) and \lim_{x \to \infty} x\,y(x).

Solution

Here P(x) = x^2, and Result 1 gives the solution on all of \mathbb{R}:

y(x) = e^{-x^2}\int_0^x e^{t^2}\,dt.

The function e^{t^2} has no elementary antiderivative (a theorem of Liouville, not proved here), so this definite integral is the answer; it can be evaluated to any accuracy, for example y(1) \approx 0.5381. It is odd: the substitution t = -s gives \int_0^{-x} e^{t^2}\,dt = -\int_0^{x} e^{s^2}\,ds, while e^{-x^2} is even.

As x \to \infty, the integral and e^{x^2} both tend to \infty, so L’Hôpital’s rule for the form \displaystyle \frac{\infty}{\infty} applies to their quotient. The ratio of derivatives is \dfrac{e^{x^2}}{2xe^{x^2}} = \dfrac{1}{2x} \to 0, so y(x) \to 0. For the second limit write x\,y(x) = \dfrac{\int_0^x e^{t^2}\,dt}{x^{-1}e^{x^2}}. The denominator has derivative e^{x^2}\left(2 - x^{-2}\right), and the ratio of derivatives \dfrac{1}{2 - x^{-2}} tends to \dfrac{1}{2}. So x\,y(x) \to \dfrac{1}{2}.

Example 4

Mei’s holding tank

At an oyster farm, Mei turns a holding tank of 120 liters of fresh water into seawater for a shipment, while draining it down for transport. Seawater with 35 grams of salt per liter flows in at 4 liters per minute, and the well-mixed water drains at 6 liters per minute. Find the salt x(t), in grams, after t minutes, and the concentration when 60 liters remain.

Solution

Here x is a function of t. The volume is V = 120 - 2t liters, so the model holds for 0 \le t < 60. Salt enters at 4 \cdot 35 = 140 grams per minute and leaves at 6 \cdot \dfrac{x}{V} grams per minute, so

x' + \frac{3}{60 - t}\,x = 140, \qquad x(0) = 0.

The coefficient is continuous on (-\infty, 60), so by Result 1 the solution lasts as long as the tank holds water. An antiderivative of \displaystyle \frac{3}{60 - t} is -3\ln(60 - t), so \mu = (60 - t)^{-3} and

\left((60 - t)^{-3}x\right)' = 140\,(60 - t)^{-3}.

Integrating, (60 - t)^{-3}x = 70\,(60 - t)^{-2} + C, so x = 70\,(60 - t) + C\,(60 - t)^3. From x(0) = 0, C = -\dfrac{4200}{216{,}000} = -\dfrac{7}{360}. Dividing by V = 2(60 - t), the concentration is

\frac{x}{V} = 35 - \frac{7}{720}\,(60 - t)^2 \text{ grams per liter}.

The tank holds 60 liters at t = 30, when the concentration is 35 - \dfrac{7 \cdot 900}{720} = 26.25 grams per liter and x = 1575 grams. As the tank empties, the concentration tends to 35, that of the seawater.

Example 5

Malik’s nursery tank

Malik tracks the water temperature T, in degrees Celsius, of the oyster farm’s outdoor nursery tank, t hours after midnight. It obeys Newton’s law of cooling toward the air temperature A(t) = 24 + 3\cos\dfrac{\pi t}{12}, with rate constant k = \dfrac{\pi}{12} per hour:

T' = k\,\big(A(t) - T\big), \qquad T(0) = 20.

Solve it, describe the oscillation that remains, and find when T is within 0.1 degree of it.

Solution

In standard form T' + kT = 24k + 3k\cos kt, since the air’s frequency \displaystyle \frac{\pi}{12} equals k here. With \mu = e^{kt}, \left(e^{kt}T\right)' = 24k\,e^{kt} + 3k\,e^{kt}\cos kt. Integrating by parts twice shows that e^{as}\cos bs has the antiderivative \dfrac{e^{as}(a\cos bs + b\sin bs)}{a^2 + b^2}, and with a = b = k,

\begin{aligned} e^{kt}T &= 24e^{kt} + C \\ &\quad + \tfrac{3}{2}e^{kt}(\cos kt + \sin kt). \end{aligned}

So \displaystyle T = 24 + \frac{3}{2}(\cos kt + \sin kt) + Ce^{-kt}, and T(0) = 25.5 + C = 20 gives C = -5.5. As in Combining a sine and a cosine into one wave, \displaystyle \cos kt + \sin kt = \sqrt{2}\cos\left(kt - \frac{\pi}{4}\right), so

\begin{aligned} T &= 24 + \frac{3}{\sqrt{2}}\cos\left(\frac{\pi t}{12} - \frac{\pi}{4}\right) \\ &\quad - 5.5\,e^{-\pi t/12}. \end{aligned}

The first two terms are T_p, the one solution of period 24 hours. It swings with amplitude \displaystyle \frac{3}{\sqrt{2}} \approx 2.12 degrees, less than the air’s 3, and peaks \displaystyle \frac{\pi/4}{\pi/12} = 3 hours after the air does (Amplitude, period and phase shift). The last term is the transient: by Result 3, every solution approaches T_p at this rate, whatever its start (Figure 2). Here |T - T_p| = 5.5e^{-\pi t/12} < 0.1 exactly when \displaystyle t > \frac{12}{\pi}\ln 55 \approx 15.31 hours, a little after 3 p.m.

0 12 24 36 48 20 22 24 26 28 t T
Figure 2The nursery tank of Example 5: the air temperature A(t) (dashed), the periodic solution T_p (thin) and the solutions with T(0) = 20 and T(0) = 28 (thick). The gap to T_p shrinks by the factor e^{-\pi t/12} (Result 3).
Example 6

A Bernoulli equation that blows up

Solve y' + 2y = e^{x}y^2, y(0) = 2, and find the maximal interval.

Solution

This is Result 4 with n = 2, p = 2, q = e^{x}. The right side f(x, y) = e^{x}y^2 - 2y has a continuous partial derivative in y, and y = 0 is a solution, so by no crossing (Direction fields and solution curves, Result 4) the solution through (0, 2) stays positive, and Result 4 applies on its whole interval. With v = y^{-1}: v' - 2v = -e^{x}, \displaystyle v(0) = \frac{1}{2}. Multiplying by e^{-2x},

\left(e^{-2x}v\right)' = -e^{-x}, \qquad e^{-2x}v = e^{-x} + C,

so v = e^{x} + Ce^{2x}, and \displaystyle v(0) = 1 + C = \frac{1}{2} gives \displaystyle C = -\frac{1}{2}. The problem for v has its solution on all of \mathbb{R}, but \displaystyle y = \frac{1}{v} needs v > 0, that is e^{x} < 2:

y = \frac{2}{2e^{x} - e^{2x}}, \quad -\infty < x < \ln 2.

As x \to (\ln 2)^-, v \to 0^+ and y \to \infty, so the interval is maximal. Check: with D = 2e^{x} - e^{2x}, y' = -\dfrac{2D'}{D^2} and y' + 2y = \dfrac{4e^{x}}{D^2} = e^{x}y^2. The nonlinearity has not gone away; it has moved into the step \displaystyle y = \frac{1}{v}.

Example 7

A short proof: forcing that dies out

Let p and q be continuous on an open interval containing [x_0, \infty), with p(x) \ge k > 0 for x \ge x_0 and q(x) \to 0 as x \to \infty. Prove that every solution of y' + p(x)\,y = q(x) on that interval tends to 0 as x \to \infty.

Solution

Let \varepsilon > 0 and choose X \ge x_0 with |q(t)| \le \varepsilon k for t \ge X. Result 1 with initial point X writes the solution, for x \ge X, as

\begin{aligned} y(x) &= e^{-\int_X^x p}\,y(X) \\ &\quad + \int_X^x e^{-\int_t^x p}\,q(t)\,dt. \end{aligned}

Since \int_t^x p \ge k(x - t) for X \le t \le x, the second term has absolute value at most \varepsilon k\int_X^x e^{-k(x - t)}\,dt, which equals \varepsilon\left(1 - e^{-k(x - X)}\right) and is less than \varepsilon. The first is at most |y(X)|\,e^{-k(x - X)}, below \varepsilon for all large x. So |y(x)| < 2\varepsilon for all large x, and \varepsilon was arbitrary. \square

Boundaries

  • Result 1 stops where p or q stops being continuous. A solution may extend past such a point or not (Example 2), and at a singular point uniqueness can fail: every function equal to c_1x^2 for x \ge 0 and c_2x^2 for x < 0 solves xy' - 2y = 0 with y(0) = 0 on \mathbb{R}.
  • Linearity in y is essential. A single term such as y^2 brings back solutions that become unbounded where the coefficients are smooth (Example 6).
  • A closed form needs antiderivatives that may not exist (Example 3). The formula with definite integrals is still an exact answer.

Where this goes next

Existence and uniqueness gives a general y' = f(x, y) a unique solution near the starting point only; Result 1 gives linear equations one on the whole interval. Exact equations seek integrating factors for other equations. Constant-coefficient equations carry Result 2 to second order, with a two-dimensional space of homogeneous solutions. Variation of parameters writes y = u(x)\,e^{-P(x)}, solves for u, and works at every order. Autonomous equations and stability read nonlinear y' = f(y) without a formula.

Exercises

Solving linear equations

  1. Solve y' + 3y = e^{-x}, y(0) = 0.
  2. Solve y' + (\tan x)\,y = \cos x, y(0) = 1, and give the interval Result 1 guarantees.
  3. Solve xy' - y = x^3e^{x}, y(1) = 2, on (0, \infty).
  4. Solve (1 + x^2)\,y' + 2xy = 3x^2, y(0) = 2.
  5. Solve y' + 2y = xe^{-2x}, y(0) = 3.
  6. Solve y' + \dfrac{y}{x} = \dfrac{\sin x}{x}, y(\pi) = 0, on (0, \infty).
  7. Write the solution of y' - 2xy = 1, y(0) = 0, as a definite integral, compute y(1) to four decimal places, and find \lim_{x \to \infty} e^{-x^2}y(x). You may use \displaystyle \int_0^\infty e^{-t^2}\,dt = \frac{\sqrt{\pi}}{2}.

Structure of the solutions

  1. Verify that x^2 solves y' + \dfrac{y}{x} = 3x on (0, \infty), and use Result 2(c) to write every solution and the one with y(1) = 0.
  2. The constant 2 solves y' + 2y = 4, and e^{x} solves y' + 2y = 3e^{x}. Use superposition to solve y' + 2y = 4 + 9e^{x}, y(0) = 0.
  3. The functions e^{x} and e^{x} + e^{-x^2} both solve one equation y' + p(x)\,y = q(x) on \mathbb{R}. Find p and q.

Long-run behavior and models

  1. Solve y' + 3y = 10\sin x. Show that exactly one solution is periodic, and find its amplitude.
  2. A 200-liter tank at the oyster farm is full of fresh water. Water flows in and out at 10 liters per minute, and the incoming water, drawn from a reservoir that is itself being salted, carries 30\left(1 - e^{-t/10}\right) grams of salt per liter at time t minutes. Find the salt x(t) in the well-mixed tank, x(20) to the nearest gram, and the limit of x(t) as t \to \infty.
  3. Two solutions of y' + (2 + \sin x)\,y = q(x), with q continuous on \mathbb{R}, differ by 5 at x = 0. Find their difference exactly, and show that it is less than 0.01 for every x \ge \ln 500.
  4. Show that exactly one solution of y' - y = \cos x is bounded on [0, \infty), and find its value at 0.

Bernoulli equations

  1. Solve y' + y = xy^3, y(0) = 1, and find the left end of the maximal interval to four decimal places.
  2. Solve the logistic equation P' = \dfrac{1}{2}P\left(1 - \dfrac{P}{80}\right), P(0) = 10, as a Bernoulli equation (Logistic growth solves it by separation).

Proofs

  1. Let \mu be differentiable and never 0 on I, with (\mu y)' = \mu\,(y' + p\,y) for every differentiable y on I. Prove that \mu = Ce^{P} for a constant C \ne 0.
  2. Let p and q be continuous on \mathbb{R} and periodic with period T > 0, with \int_0^T p(s)\,ds \ne 0. Prove that y' + p\,y = q has exactly one solution of period T.
  3. Let y_1 and y_2 solve y' + p\,y = q on I, with y_1(x_0) < y_2(x_0). Prove that y_1 < y_2 on all of I.
  4. Refute: “If p is continuous and positive on [0, \infty), every solution of y' + p(x)\,y = 0 tends to 0.” Use p(x) = \dfrac{1}{(1 + x)^2}.
  5. Let a, b, c be continuous on I and y_1 a solution of the Riccati equation y' = a(x) + b(x)\,y + c(x)\,y^2. Prove that on an interval where y \ne y_1, y solves the Riccati equation if and only if v = \dfrac{1}{y - y_1} solves the linear equation v' + (b + 2cy_1)\,v = -c.
  6. Let p and q be continuous on an open interval containing [x_0, \infty), with p(x) \to k > 0 and q(x) \to m as x \to \infty. Prove that every solution of y' + p\,y = q tends to \dfrac{m}{k}. (Use Example 7.)
Answers
  1. \displaystyle y = \frac{1}{2}\left(e^{-x} - e^{-3x}\right): \left(e^{3x}y\right)' = e^{2x}, so \displaystyle e^{3x}y = \frac{1}{2}e^{2x} + C with \displaystyle C = -\frac{1}{2}.
  2. y = (x + 1)\cos x on \displaystyle \left(-\frac{\pi}{2}, \frac{\pi}{2}\right), the interval around 0 where \tan x is continuous. Here \mu = \sec x, (y\sec x)' = 1 and y\sec x = x + 1.
  3. y = x(x - 1)e^{x} + 2x. In standard form \displaystyle y' - \frac{y}{x} = x^2e^{x}, \displaystyle \mu = \frac{1}{x}, \displaystyle \left(\frac{y}{x}\right)' = xe^{x}, and \displaystyle \frac{y}{x} = (x - 1)e^{x} + 2.
  4. y = \dfrac{x^3 + 2}{1 + x^2} on \mathbb{R}. The left side is already \left((1 + x^2)\,y\right)', so (1 + x^2)\,y = x^3 + 2.
  5. \displaystyle y = \left(\frac{x^2}{2} + 3\right)e^{-2x}: \left(e^{2x}y\right)' = x, so \displaystyle e^{2x}y = \frac{x^2}{2} + 3.
  6. y = -\dfrac{1 + \cos x}{x}: \mu = x, (xy)' = \sin x, xy = -\cos x + C, and 0 = 1 + C.
  7. y = e^{x^2}\int_0^x e^{-t^2}\,dt (here P = -x^2), y(1) \approx 2.0301, and \displaystyle e^{-x^2}y(x) = \int_0^x e^{-t^2}\,dt \to \frac{\sqrt{\pi}}{2}.
  8. \displaystyle (x^2)' + \frac{x^2}{x} = 3x. From x_0 = 1, \displaystyle e^{-P} = \frac{1}{x}, so the solutions are y = x^2 + \dfrac{C}{x}, and y(1) = 0 gives y = x^2 - \dfrac{1}{x}.
  9. y = 2 + 3e^{x} - 5e^{-2x}. By Result 2(a), 2 + 3e^{x} solves with forcing 4 + 9e^{x}; add Ce^{-2x} and use y(0) = 5 + C = 0.
  10. p = 2x and q = (1 + 2x)e^{x}. The difference e^{-x^2} solves the homogeneous equation, so p = -\dfrac{(e^{-x^2})'}{e^{-x^2}} = 2x, and q = y_1' + p\,y_1 with y_1 = e^{x}.
  11. y = 3\sin x - \cos x + Ce^{-3x}, since \displaystyle \frac{1}{10}e^{3x}(3\sin x - \cos x) is an antiderivative of e^{3x}\sin x. For C \ne 0 the solution is unbounded as x \to -\infty, so not periodic; for C = 0 it has period 2\pi and amplitude \displaystyle \sqrt{10}.
  12. x = 6000 + 6000e^{-t/10} - 12{,}000e^{-t/20} solves \displaystyle x' + \frac{x}{20} = 300\left(1 - e^{-t/10}\right), x(0) = 0 (with \mu = e^{t/20}). So x(20) \approx 2397 grams, and x \to 6000 grams, the volume times 30 grams per liter.
  13. d(x) = 5e^{-(2x + 1 - \cos x)} by Result 2(b), since \int_0^x (2 + \sin s)\,ds = 2x + 1 - \cos x. For x > 0 this exceeds x, as x + 1 - \cos x > 0, so |d| < 5e^{-x} \le 0.01 when x \ge \ln 500.
  14. \displaystyle y = \frac{1}{2}(\sin x - \cos x) + Ce^{x}, from \left(e^{-x}y\right)' = e^{-x}\cos x. It is bounded on [0, \infty) exactly when C = 0, and that solution has \displaystyle y(0) = -\frac{1}{2}.
  15. \displaystyle y = \sqrt{\dfrac{2}{2x + 1 + e^{2x}}} on (-0.6392, \infty). With n = 3, v = y^{-2} solves v' - 2v = -2x, v(0) = 1, so \displaystyle v = x + \frac{1}{2} + \frac{1}{2}e^{2x}. The solution stays positive by no crossing with y = 0, and the left end is the zero of 2x + 1 + e^{2x}, where y \to \infty.
  16. P = \dfrac{80}{1 + 7e^{-t/2}} for all real t. With n = 2, \displaystyle v = \frac{1}{P} solves \displaystyle v' + \frac{v}{2} = \frac{1}{160}, so \displaystyle v = \frac{1}{80} + Ce^{-t/2}, and \displaystyle v(0) = \frac{1}{10} gives \displaystyle C = \frac{7}{80}.
  17. Taking y = 1 gives \mu' = p\,\mu. Then \left(\mu e^{-P}\right)' = (\mu' - p\,\mu)\,e^{-P} = 0 on the interval I, so \mu e^{-P} is a constant C by the mean value theorem, and C \ne 0 since \mu is never 0. \square
  18. Let y be the solution on \mathbb{R} with y(0) = y_0 (Result 1). Then z(x) = y(x + T) solves the same equation, since z'(x) = q(x + T) - p(x + T)\,z(x) and p, q have period T. If y(T) = y_0, then z(0) = y(0) and z = y by uniqueness, so y has period T; conversely a solution of period T has y(T) = y(0). By Result 1, y(T) = e^{-P(T)}(y_0 + c) with c = \int_0^T e^{P(t)}q(t)\,dt, so the condition reads y_0\left(1 - e^{-P(T)}\right) = e^{-P(T)}c. As P(T) \ne 0, exactly one y_0 works. \square
  19. y_2 - y_1 solves the homogeneous equation, so by Result 2(b) it equals \big(y_2(x_0) - y_1(x_0)\big)\,e^{-P(x)}, which is positive on all of I. \square
  20. Here \displaystyle P(x) = \int_0^x \frac{ds}{(1 + s)^2} = \frac{x}{1 + x}, so the solution with y(0) = 1 is y = e^{-x/(1 + x)}, which tends to e^{-1} \ne 0. The claim is false: p is positive, but \int_0^\infty p = 1 is finite.
  21. Put u = y - y_1, never 0, and \displaystyle v = \frac{1}{u}. Subtracting the equation for y_1, y solves the Riccati equation if and only if u' = b\,u + c\,(2y_1u + u^2), that is u' = (b + 2cy_1)\,u + c\,u^2. If so, \displaystyle v' = -\frac{u'}{u^2} = -(b + 2cy_1)\,v - c. Conversely, if v' + (b + 2cy_1)\,v = -c, then \displaystyle u' = -\frac{v'}{v^2} works out to (b + 2cy_1)\,u + c\,u^2, and y = y_1 + u solves. \square
  22. Let \displaystyle z = y - \frac{m}{k}. Then \displaystyle z' + p\,z = q - \frac{m}{k}\,p, a continuous forcing r with \displaystyle r(x) \to m - \frac{m}{k}k = 0. Since p \to k, there is X \ge x_0 with \displaystyle p \ge \frac{k}{2} > 0 on [X, \infty). Example 7, applied from X with the constant \displaystyle \frac{k}{2} and forcing r, gives z \to 0, so \displaystyle y \to \frac{m}{k}. \square
math.wiki · CC-BY-SA-4.0