Variation of parameters

Undetermined coefficients needs constant coefficients and a forcing whose derivatives keep its shape. Once two independent solutions of the homogeneous equation are known, how is a particular solution found for every continuous forcing, such as csc 3x or 1/(1 + x^2)?

Where you are 5 in Second-order linear equations

Why this exists

Undetermined coefficients works only when the coefficients are constant and the forcing is an exponential polynomial. For \csc 3x, \displaystyle \frac{1}{1 + x^2} or \ln x no trial shape exists, and variable coefficients defeat it entirely. Yet by Result 1 of Constant-coefficient equations, once the homogeneous equation is solved, one particular solution is all that is missing.

Variation of parameters supplies it for every continuous forcing and every equation y'' + p(x)\,y' + q(x)\,y = f(x) whose homogeneous solutions are known, at the price of two integrals. It also writes the answer as one integral of f against a fixed kernel, which shows how a solution responds to the whole history of the forcing. Throughout, y is a function of x, or of t when the variable is time.

Definition

In plain words: keep the shape c_1y_1 + c_2y_2 of the homogeneous solution, let the constants vary with x, and choose how they vary so that the forcing comes out.

Let p, q, f be continuous on an open interval I and L[y] = y'' + p(x)\,y' + q(x)\,y, so that L[y] = f is in standard form (leading coefficient 1). An equation P(x)\,y'' + Q(x)\,y' + R(x)\,y = F(x) with P never 0 on I is put in standard form by dividing by P; its forcing is then \displaystyle f = \frac{F}{P}. Let y_1, y_2 be a fundamental set of L[y] = 0 on I (Result 4 of Constant-coefficient equations): two solutions whose Wronskian

W(x) = y_1(x)\,y_2'(x) - y_1'(x)\,y_2(x)

has no zero on I. Variation of parameters looks for a solution

y_p = u_1(x)\,y_1(x) + u_2(x)\,y_2(x)

with differentiable functions u_1, u_2 satisfying the two variation equations

\begin{aligned} u_1'\,y_1 + u_2'\,y_2 &= 0, \\ u_1'\,y_1' + u_2'\,y_2' &= f. \end{aligned}

The Green’s function of the initial value problem for L is, for x, s \in I,

G(x, s) = \frac{y_1(s)\,y_2(x) - y_1(x)\,y_2(s)}{W(s)}.

Intuition

Two unknown functions for one equation leave room for one extra condition, and the first variation equation is chosen to keep the work short. It makes y_p' = u_1y_1' + u_2y_2': at each x, the value and slope of y_p agree with those of the homogeneous solution whose constants are frozen at u_1(x), u_2(x). So y_p slides from one homogeneous solution to the next, only its second derivative feels the sliding, and the second variation equation asks that what it feels be f. At each x the two equations are a 2 \times 2 linear system for u_1'(x), u_2'(x) with determinant W(x) \ne 0, so they have exactly one solution.

The Green’s function has a mechanical reading. For fixed s, x \mapsto G(x, s) solves L[y] = 0, is 0 at x = s and has slope 1 there (Result 2): the motion of a system at rest that receives a unit kick of velocity at time s. A forcing f adds velocity f(s)\,ds during each short time ds, and by linearity the responses to these kicks add up to \int_{x_0}^{x} G(x, s)\,f(s)\,ds.

Results

1. Variation of parameters. Let p, q, f be continuous on the open interval I, and let y_1, y_2 be a fundamental set of L[y] = 0 on I with Wronskian W. At each x \in I the variation equations hold exactly when

u_1' = -\frac{y_2\,f}{W}, \qquad u_2' = \frac{y_1\,f}{W}.

These right sides are continuous on I, so antiderivatives u_1, u_2 exist, and for every such choice y_p = u_1y_1 + u_2y_2 solves L[y] = f on I. Every solution of L[y] = f on I is y_p + c_1y_1 + c_2y_2 for exactly one pair of constants.

2. The Green’s function. Under the same hypotheses, let x_0 \in I. Then

y_p(x) = \int_{x_0}^{x} G(x, s)\,f(s)\,ds

is the only solution of L[y] = f on I with y(x_0) = 0 and y'(x_0) = 0, and the solution with y(x_0) = k_0, y'(x_0) = k_1 is y_h + y_p, where y_h solves L[y] = 0 with those initial values. For each s \in I, x \mapsto G(x, s) solves L[y] = 0 with G(s, s) = 0 and \displaystyle \frac{\partial G}{\partial x}(s, s) = 1 (the derivative in x with s held fixed). G does not depend on which fundamental set is used.

3. Constant coefficients. Let a \ne 0, b, c be real, f continuous on an open interval I, and x_0 \in I. Let g solve a g'' + b g' + c g = 0 with g(0) = 0 and \displaystyle g'(0) = \frac{1}{a}. Then the solution of a y'' + b y' + c y = f on I with y(x_0) = 0 and y'(x_0) = 0 is

y(x) = \int_{x_0}^{x} g(x - s)\,f(s)\,ds.
Why it is true

Result 1. At a fixed x the variation equations are a linear system in u_1', u_2' with coefficient matrix \begin{bmatrix} y_1 & y_2 \\ y_1' & y_2' \end{bmatrix} of determinant W(x) \ne 0 (Computing a determinant). By Cramer's rule its only solution is

\begin{aligned} u_1' &= \frac{1}{W}\begin{vmatrix} 0 & y_2 \\ f & y_2' \end{vmatrix} = -\frac{y_2\,f}{W}, \\ u_2' &= \frac{1}{W}\begin{vmatrix} y_1 & 0 \\ y_1' & f \end{vmatrix} = \frac{y_1\,f}{W}. \end{aligned}

These quotients are continuous, since y_j, y_j', f and W are and W has no zero, so by the The fundamental theorem of calculus each has an antiderivative on I. Take any such u_1, u_2. By the product rule and the first variation equation,

\begin{aligned} y_p' &= (u_1'y_1 + u_2'y_2) + u_1y_1' + u_2y_2' \\ &= u_1y_1' + u_2y_2'. \end{aligned}

This is differentiable, as u_j is differentiable and y_j twice differentiable, and by the second variation equation

\begin{aligned} y_p'' &= (u_1'y_1' + u_2'y_2') + u_1y_1'' + u_2y_2'' \\ &= f + u_1y_1'' + u_2y_2''. \end{aligned}

Adding p\,y_p' and q\,y_p,

\begin{aligned} L[y_p] &= f + u_1\big(y_1'' + p\,y_1' + q\,y_1\big) \\ &\quad + u_2\big(y_2'' + p\,y_2' + q\,y_2\big) = f, \end{aligned}

since L[y_1] = L[y_2] = 0. Finally, if L[y] = f on I then L[y - y_p] = 0 (Result 1 of Constant-coefficient equations), and by Result 4 there y - y_p = c_1y_1 + c_2y_2 for exactly one pair. \square

Result 2. Choose the antiderivatives that vanish at x_0:

\begin{aligned} u_1(x) &= -\int_{x_0}^{x}\frac{y_2(s)\,f(s)}{W(s)}\,ds, \\ u_2(x) &= \int_{x_0}^{x}\frac{y_1(s)\,f(s)}{W(s)}\,ds. \end{aligned}

The factors y_1(x), y_2(x) do not depend on s, so they move inside the integrals, and u_1y_1 + u_2y_2 = \int_{x_0}^{x} G(x, s)\,f(s)\,ds. By Result 1 this solves L[y] = f. At x_0 both integrals are 0, so y_p(x_0) = 0 and, by the formula for y_p' above, y_p'(x_0) = 0. No other solution has these initial values, by Result 2 (existence and uniqueness) of Constant-coefficient equations; for the same reason y_h + y_p is the solution with y(x_0) = k_0, y'(x_0) = k_1.

For fixed s, \displaystyle G(x, s) = \frac{y_1(s)}{W(s)}\,y_2(x) - \frac{y_2(s)}{W(s)}\,y_1(x) is a constant combination of y_1, y_2, so it solves L[y] = 0. Its numerator is 0 at x = s, and its derivative in x there is \displaystyle \frac{y_1(s)\,y_2'(s) - y_1'(s)\,y_2(s)}{W(s)} = 1.

Last, let \tilde y_1, \tilde y_2 be another fundamental set. By Result 4 of Constant-coefficient equations, \tilde y_1 = m_{11}y_1 + m_{21}y_2 and \tilde y_2 = m_{12}y_1 + m_{22}y_2 for constants forming a matrix M. The numerator of G is the determinant of the matrix with rows \big(y_1(s), y_2(s)\big) and \big(y_1(x), y_2(x)\big), and W(s) the determinant with rows \big(y_1(s), y_2(s)\big) and \big(y_1'(s), y_2'(s)\big). The new set multiplies both matrices on the right by M, so both determinants are multiplied by \det M, which is not 0 because \tilde W = W\det M has no zero. The quotient G is unchanged. \square

Result 3. Dividing by a gives standard form with constant \displaystyle p = \frac{b}{a}, \displaystyle q = \frac{c}{a} and forcing \displaystyle \frac{f}{a}; by Result 2 of Constant-coefficient equations, g exists on all of \mathbb{R}. Fix s \in I. By the chain rule x \mapsto a\,g(x - s) has derivatives a\,g'(x - s) and a\,g''(x - s), so, the coefficients being constant, it solves the homogeneous equation, with value 0 and slope 1 at x = s. By Result 2 so does x \mapsto G(x, s), hence G(x, s) = a\,g(x - s) by uniqueness. Result 2 with forcing \displaystyle \frac{f}{a} now gives

\begin{aligned} y(x) &= \int_{x_0}^{x} a\,g(x - s)\,\frac{f(s)}{a}\,ds \\ &= \int_{x_0}^{x} g(x - s)\,f(s)\,ds. \end{aligned}

\square

Worked examples

Example 1

A cosecant forcing

Find the general solution of y'' + 9y = \csc 3x on \displaystyle \left(0, \frac{\pi}{3}\right).

Solution

The forcing is continuous there, where \sin 3x > 0, but has no trial shape. The roots \pm 3i give y_1 = \cos 3x and y_2 = \sin 3x, with W = 3\cos^2 3x + 3\sin^2 3x = 3. Result 1 gives

\begin{aligned} u_1' &= -\frac{\sin 3x\,\csc 3x}{3} = -\frac{1}{3}, \\ u_2' &= \frac{\cos 3x\,\csc 3x}{3} = \frac{\cot 3x}{3}. \end{aligned}

So \displaystyle u_1 = -\frac{x}{3}, and \displaystyle u_2 = \frac{1}{9}\ln(\sin 3x) by u-substitution with w = \sin 3x; no absolute value is needed since \sin 3x > 0. By Result 1 the general solution is

\begin{aligned} y &= c_1\cos 3x + c_2\sin 3x \\ &\quad - \tfrac{x}{3}\cos 3x \\ &\quad + \tfrac{1}{9}\sin 3x\,\ln(\sin 3x). \end{aligned}
Example 2

Dropping a homogeneous part

Find a particular solution of \displaystyle y'' - 3y' + 2y = \frac{1}{1 + e^{-x}}.

Solution

The roots 1 and 2 give y_1 = e^{x} and y_2 = e^{2x}, with W = e^{x} \cdot 2e^{2x} - e^{x} \cdot e^{2x} = e^{3x}. Result 1 gives

\begin{aligned} u_1' &= -\frac{e^{2x}}{e^{3x}}\cdot\frac{1}{1 + e^{-x}} = -\frac{1}{e^{x} + 1}, \\ u_2' &= \frac{e^{x}}{e^{3x}}\cdot\frac{1}{1 + e^{-x}} = \frac{e^{-x}}{e^{x} + 1}. \end{aligned}

Multiplying top and bottom of u_1' by e^{-x} gives \displaystyle -\frac{e^{-x}}{1 + e^{-x}}, the derivative of \ln(1 + e^{-x}), so u_1 = \ln(1 + e^{-x}). For u_2 substitute w = e^{-x}, so e^{-x}\,dx = -dw and \displaystyle \frac{1}{e^{x} + 1} = \frac{w}{1 + w}:

\begin{aligned} \int \frac{e^{-x}\,dx}{e^{x} + 1} &= -\int\left(1 - \frac{1}{1 + w}\right)dw \\ &= -w + \ln(1 + w), \end{aligned}

so u_2 = -e^{-x} + \ln(1 + e^{-x}). Then

\begin{aligned} u_1y_1 + u_2y_2 &= (e^{x} + e^{2x})\ln(1 + e^{-x}) \\ &\quad - e^{x}. \end{aligned}

The last term solves the homogeneous equation, so by Result 1 of Constant-coefficient equations it may be dropped:

y_p = (e^{x} + e^{2x})\ln(1 + e^{-x}).
Example 3

Variable coefficients and standard form

The functions x and x^3 solve x^2y'' - 3xy' + 3y = 0 on (0, \infty). Find the general solution of x^2y'' - 3xy' + 3y = 2x^4e^{x} there.

Solution

The substitutions 0 - 3x + 3x = 0 and 6x^3 - 9x^3 + 3x^3 = 0 confirm the two solutions, and W = x \cdot 3x^2 - 1 \cdot x^3 = 2x^3 has no zero on (0, \infty), so they form a fundamental set. Result 1 needs standard form, so divide by x^2, which is not 0 there: the forcing is f = 2x^2e^{x}, not 2x^4e^{x}. Then

\begin{aligned} u_1' &= -\frac{x^3 \cdot 2x^2e^{x}}{2x^3} = -x^2e^{x}, \\ u_2' &= \frac{x \cdot 2x^2e^{x}}{2x^3} = e^{x}. \end{aligned}

Integrating by parts twice (Integration by parts), u_1 = -(x^2 - 2x + 2)\,e^{x}, and u_2 = e^{x}. So

\begin{aligned} y_p &= -x(x^2 - 2x + 2)\,e^{x} + x^3e^{x} \\ &= (2x^2 - 2x)\,e^{x}, \end{aligned}

and the general solution is

y = c_1x + c_2x^3 + (2x^2 - 2x)\,e^{x}.
Example 4

An initial value problem as one integral

Solve \displaystyle y'' + 2y' + y = \frac{e^{-x}}{1 + x^2}, y(0) = 0, y'(0) = 0.

Solution

The forcing is continuous on \mathbb{R}, so Result 3 applies with a = 1 on all of \mathbb{R}. The root -1 is double, so g = (c_1 + c_2x)\,e^{-x} (Complex and repeated roots); g(0) = 0 forces c_1 = 0 and g'(0) = c_2 = 1. So g(x) = x\,e^{-x} and

\begin{aligned} y &= \int_0^x (x - s)\,e^{-(x - s)}\,\frac{e^{-s}}{1 + s^2}\,ds \\ &= e^{-x}\int_0^x \frac{x - s}{1 + s^2}\,ds, \end{aligned}

because e^{-(x - s)}e^{-s} = e^{-x} does not depend on s. With \displaystyle \int \frac{ds}{1 + s^2} = \arctan s and \displaystyle \int \frac{s\,ds}{1 + s^2} = \frac{1}{2}\ln(1 + s^2),

y = e^{-x}\left(x\arctan x - \tfrac{1}{2}\ln(1 + x^2)\right).

Check: write y = e^{-x}h. As -1 is a double root, Result 2 of Undetermined coefficients gives y'' + 2y' + y = e^{-x}h'', and here h' = \arctan x, \displaystyle h'' = \frac{1}{1 + x^2}. Also h(0) = h'(0) = 0, so y(0) = 0 and y'(0) = h'(0) - h(0) = 0.

Example 5

A short proof: bounded forcing, bounded response

Let f be continuous on \mathbb{R} with |f(x)| \le M for x \ge 0. Prove that for each \varepsilon > 0 every solution of y'' + 4y' + 3y = f satisfies \displaystyle |y(x)| < \frac{M}{3} + \varepsilon for all large x, and that \displaystyle \frac{M}{3} cannot be made smaller.

Solution

The roots are -1 and -3, and \displaystyle g(x) = \frac{1}{2}\left(e^{-x} - e^{-3x}\right) has g(0) = 0 and \displaystyle g'(0) = \frac{1}{2}(-1 + 3) = 1. For w \ge 0, e^{-w} \ge e^{-3w}, so g(w) \ge 0. By Results 2 and 3 with x_0 = 0, a solution is y = y_h + \int_0^x g(x - s)\,f(s)\,ds with y_h = c_1e^{-x} + c_2e^{-3x}. For x \ge 0, substituting w = x - s,

\begin{aligned} \left|\int_0^x g(x - s)\,f(s)\,ds\right| &\le M\int_0^x g(w)\,dw \\ &\le M\int_0^\infty g(w)\,dw = \frac{M}{3}, \end{aligned}

since the improper integral is \displaystyle \frac{1}{2}\left(1 - \frac{1}{3}\right) = \frac{1}{3}. As y_h(x) \to 0, |y_h(x)| < \varepsilon for all large x, and then \displaystyle |y(x)| < \frac{M}{3} + \varepsilon. For f = M the constant \displaystyle \frac{M}{3} is a solution, so no smaller bound holds. \square

Example 6

Aisha’s marionette

Aisha works a large marionette from a bridge above the stage. The puppet hangs on strings 2.45 meters long, so for small swings its horizontal position y in meters, a function of the time t in seconds, obeys \displaystyle y'' = -\frac{9.8}{2.45}\big(y - h(t)\big) = -4\big(y - h(t)\big), where h(t) is the position of her control bar. Bar and puppet rest at 0 until t = 0; then she slides the bar 1 meter at constant speed over T seconds: \displaystyle h(t) = \frac{t}{T} for 0 \le t \le T and h(t) = 1 for t \ge T. Find y(t), and the amplitude of the swing left after the move, for T = 1 and T = \pi.

Solution

The equation is y'' + 4y = 4h(t) with h = 0 for t < 0. The forcing is continuous on \mathbb{R}, as the pieces agree at 0 and T, but is not an exponential polynomial on any interval around 0 or T. Result 3 with a = 1 and \displaystyle g(t) = \frac{1}{2}\sin 2t gives

y(t) = 2\int_0^t \sin 2(t - s)\,h(s)\,ds.

For 0 \le t \le T integrate by parts, using the antiderivative \displaystyle \frac{1}{2}\cos 2(t - s) of \sin 2(t - s) in s:

\begin{aligned} \int_0^t s\sin 2(t - s)\,ds &= \Big[\tfrac{s}{2}\cos 2(t - s)\Big]_0^t \\ &\quad - \tfrac{1}{2}\int_0^t \cos 2(t - s)\,ds \\ &= \tfrac{t}{2} - \tfrac{1}{4}\sin 2t, \end{aligned}

so \displaystyle y = \frac{t}{T} - \frac{\sin 2t}{2T}. For t \ge T split the integral at T: the same computation over [0, T] gives \displaystyle \frac{T}{2}\cos 2(t - T) + \frac{1}{4}\big(\sin 2(t - T) - \sin 2t\big), and 2\int_T^t \sin 2(t - s)\,ds = 1 - \cos 2(t - T). Adding, the cosines cancel:

\begin{aligned} y &= 1 + \frac{\sin 2(t - T) - \sin 2t}{2T} \\ &= 1 - \frac{\sin T}{T}\cos(2t - T), \end{aligned}

by \displaystyle \sin A - \sin B = 2\cos\frac{A + B}{2}\sin\frac{A - B}{2}. After the move the puppet swings about the bar with amplitude \displaystyle \frac{|\sin T|}{T} meters (Amplitude, period and phase shift).

For T = 1 that is \sin 1 \approx 0.841 meters. For T = \pi it is 0: the puppet arrives at rest under the bar (Figure 1). The free swing has period \pi, so a move lasting one period leaves no swing, while a sudden move gives y = 1 - \cos 2t, amplitude 1, the limit of \displaystyle \frac{\sin T}{T} as T \to 0.

2 4 6 8 1 2 t y
Figure 1Example 6: the puppet’s position y(t) when the bar moves 1 meter in T = 1 second (dashed) and in T = \pi seconds (solid). The swing left after the move has amplitude \displaystyle \frac{|\sin T|}{T}.

Boundaries

  • The homogeneous solutions must be known. The method uses a fundamental set and does not find one; with variable coefficients that is the hard part. Abel’s identity (Example 5 of Constant-coefficient equations) gives a second solution from a first.
  • Standard form, where the leading coefficient has no zero. In Example 3 the forcing is 2x^2e^{x}, not 2x^4e^{x}, and solutions live on (0, \infty) or (-\infty, 0), not across 0.
  • The forcing must be continuous on the interval. Each interval between zeros of \sin 3x gets its own solutions of Example 1. For a forcing with a jump, the integral of Result 2 still gives continuous y and y', but y'' jumps and the equation fails at the jump; The Laplace transform handles such forcing.
  • The integrals need not be elementary. For \displaystyle y'' + y = \frac{1}{1 + x^2} with zero initial values, Result 3 gives the exact answer \displaystyle \int_0^x \frac{\sin(x - s)}{1 + s^2}\,ds; where no antiderivative can be written in closed form, the integral is evaluated numerically.
  • Linear equations only. Result 1’s proof split L[u_1y_1 + u_2y_2] using linearity; for y'' + y^2 = f nothing splits.
  • Higher order works the same way with n functions u_j, n variation equations and the n \times n Wronskian (Exercise 18).

Where this goes next

Springs, damping and resonance applies Result 3 with \displaystyle g(t) = \frac{\sin\omega_0 t}{\omega_0} to a spring under any force, and finds the growing response t\sin\omega_0 t when the forcing matches the natural frequency. The Laplace transform recognizes the integral of Result 3 as a convolution g * f, which the transform turns into a product. Linear systems of differential equations uses the same idea for \mathbf{x}' = A\mathbf{x} + \mathbf{f}(t), with the constant vector of the homogeneous solution replaced by a vector of functions.

Exercises

Particular solutions

Find a particular solution on the given interval.

  1. y'' + 4y = \sec^2 2x on \displaystyle \left(-\frac{\pi}{4}, \frac{\pi}{4}\right)
  2. \displaystyle y'' - 2y' + y = \frac{e^{x}}{x^2} on (0, \infty)
  3. y'' + y = \sec x\,\tan x on \displaystyle \left(-\frac{\pi}{2}, \frac{\pi}{2}\right)
  4. y'' + 2y' + 5y = e^{-x}\sec 2x on \displaystyle \left(-\frac{\pi}{4}, \frac{\pi}{4}\right)

Variable coefficients

On (0, \infty), check that the two functions solve the homogeneous equation, then solve the forced one.

  1. x^2 and x^2\ln x; find the general solution of x^2y'' - 3xy' + 4y = x^2.
  2. e^{x} and x + 1; find the general solution of xy'' - (x + 1)\,y' + y = x^2.
  3. x and \displaystyle \frac{1}{x}; find a particular solution of x^2y'' + xy' - y = x.

The integral form

  1. Find the function g of Result 3 for y'' + 16y = f, and use it to solve y'' + 16y = x, y(0) = 0, y'(0) = 0.
  2. Solve y'' + y = \sec x, y(0) = 1, y'(0) = 0, on \displaystyle \left(-\frac{\pi}{2}, \frac{\pi}{2}\right).
  3. Find g for 2y'' + 5y' + 2y = f, and use it to solve 2y'' + 5y' + 2y = 6, y(0) = 0, y'(0) = 0.
  4. Find a particular solution of y'' - y = 4e^{x} by variation of parameters and by undetermined coefficients, and explain why the two answers differ.

The marionette

  1. Sarah, who works the second puppet, makes the move of Example 6 with T = \pi. Show that during the move \displaystyle h(t) - y(t) = \frac{\sin 2t}{2\pi}, give the largest distance between bar and puppet, and show that y(\pi) = 1 and y'(\pi) = 0.
  2. Sarah tries \displaystyle T = \frac{\pi}{2}. Find the amplitude of the swing left after the move. Find every T > 0 that leaves no swing, and show that every ramp with T > 0 leaves a smaller swing than a sudden move.

Proofs

  1. Show that replacing u_1, u_2 in Result 1 by other antiderivatives changes y_p by a solution of L[y] = 0, and that as the constants of integration run over all real pairs, u_1y_1 + u_2y_2 runs over every solution of L[y] = f on I.
  2. Let f be continuous on \mathbb{R} with \int_0^\infty |f(x)|\,dx convergent. Prove that every solution of y'' + y = f is bounded on [0, \infty).
  3. Let \omega > 0 and f be continuous on \mathbb{R}. Using only the subtraction formula for sine and the fundamental theorem of calculus, prove that \displaystyle y(x) = \frac{1}{\omega}\int_0^x \sin\omega(x - s)\,f(s)\,ds solves y'' + \omega^2y = f with y(0) = 0 and y'(0) = 0.
  4. Let k > 0 and f be continuous on \mathbb{R} with f(x) \ge 0 for x \ge 0. Prove that the solution of y'' - k^2y = f with y(0) = 0, y'(0) = 0 has y(x) \ge 0 and y'(x) \ge 0 for all x \ge 0.
  5. Let p_0, p_1, p_2, f be continuous on an open interval I, M[y] = y''' + p_2y'' + p_1y' + p_0y, and y_1, y_2, y_3 solutions of M[y] = 0 on I. Prove that if differentiable u_1, u_2, u_3 satisfy \sum_j u_j'y_j = 0, \sum_j u_j'y_j' = 0 and \sum_j u_j'y_j'' = f on I, then u_1y_1 + u_2y_2 + u_3y_3 solves M[y] = f.
Answers
  1. \displaystyle y_p = \frac{1}{4}\big(\sin 2x\,\ln(\sec 2x + \tan 2x) - 1\big). Here W = 2; \displaystyle u_1' = -\frac{\sin 2x}{2\cos^2 2x} gives \displaystyle u_1 = -\frac{1}{4}\sec 2x, and \displaystyle u_2' = \frac{1}{2}\sec 2x gives \displaystyle u_2 = \frac{1}{4}\ln(\sec 2x + \tan 2x). Then \displaystyle u_1\cos 2x = -\frac{1}{4}.
  2. y_p = -e^{x}\ln x. With y_2 = xe^{x}, W = e^{2x}, \displaystyle u_1' = -\frac{1}{x}, \displaystyle u_2' = \frac{1}{x^2}, so u_1 = -\ln x, \displaystyle u_2 = -\frac{1}{x}, and u_1y_1 + u_2y_2 = -e^{x}\ln x - e^{x}; the homogeneous -e^{x} is dropped.
  3. y_p = x\cos x - \sin x\,\ln(\cos x). Here W = 1; u_1' = -\tan^2 x = 1 - \sec^2 x, so u_1 = x - \tan x, and u_2' = \tan x, so u_2 = -\ln(\cos x). The sum has an extra -\sin x, which is dropped.
  4. \displaystyle y_p = e^{-x}\left(\frac{1}{4}\cos 2x\,\ln(\cos 2x) + \frac{x}{2}\sin 2x\right). With y_1 = e^{-x}\cos 2x, y_2 = e^{-x}\sin 2x, W = 2e^{-2x}: \displaystyle u_1' = -\frac{1}{2}\tan 2x, \displaystyle u_1 = \frac{1}{4}\ln(\cos 2x), and \displaystyle u_2' = \frac{1}{2}, \displaystyle u_2 = \frac{x}{2}.
  5. \displaystyle y = c_1x^2 + c_2x^2\ln x + \frac{1}{2}x^2(\ln x)^2. Both solve: 2x^2 - 6x^2 + 4x^2 = 0, and (x^2\ln x)' = 2x\ln x + x, (x^2\ln x)'' = 2\ln x + 3 give 0 too. W = x^3 and the standard-form forcing is 1, so \displaystyle u_1' = -\frac{\ln x}{x}, \displaystyle u_1 = -\frac{1}{2}(\ln x)^2, \displaystyle u_2' = \frac{1}{x}, u_2 = \ln x.
  6. y = c_1e^{x} + c_2(x + 1) - x^2. Both solve: xe^{x} - (x + 1)e^{x} + e^{x} = 0 and -(x + 1) + (x + 1) = 0. The forcing is x and W = -xe^{x}, so u_1' = (x + 1)e^{-x}, u_1 = -(x + 2)e^{-x}, u_2' = -1, u_2 = -x. The sum is -x^2 - 2(x + 1), and -2(x + 1) is dropped.
  7. \displaystyle y_p = \frac{1}{2}x\ln x. Both solve: x - x = 0 and \displaystyle \frac{2}{x} - \frac{1}{x} - \frac{1}{x} = 0. The forcing is \displaystyle \frac{1}{x} and \displaystyle W = -\frac{2}{x}, so \displaystyle u_1' = \frac{1}{2x}, \displaystyle u_1 = \frac{1}{2}\ln x, \displaystyle u_2' = -\frac{x}{2}, \displaystyle u_2 = -\frac{x^2}{4}. The sum has an extra \displaystyle -\frac{x}{4}, which is dropped.
  8. \displaystyle g(x) = \frac{1}{4}\sin 4x and \displaystyle y = \frac{x}{16} - \frac{\sin 4x}{64}. Indeed g'' + 16g = 0, g(0) = 0, g'(0) = 1, and \displaystyle y = \frac{1}{4}\int_0^x s\sin 4(x - s)\,ds, where by parts the integral is \displaystyle \frac{x}{4} - \frac{\sin 4x}{16}.
  9. y = \cos x + \cos x\,\ln(\cos x) + x\sin x. Here W = 1, u_1' = -\tan x, u_2' = 1; the antiderivatives vanishing at 0, u_1 = \ln(\cos x) and u_2 = x, give the solution with zero initial values (Result 2). Add y_h = \cos x, with y_h(0) = 1, y_h'(0) = 0.
  10. \displaystyle g(x) = \frac{1}{3}\left(e^{-x/2} - e^{-2x}\right) and y = 3 - 4e^{-x/2} + e^{-2x}. The roots are \displaystyle -\frac{1}{2} and -2, so g = A\left(e^{-x/2} - e^{-2x}\right) with \displaystyle g'(0) = \frac{3A}{2} = \frac{1}{2}. Then y = 6\int_0^x g(w)\,dw.
  11. Variation of parameters gives y_p = 2xe^{x} - e^{x}; undetermined coefficients gives y_p = 2xe^{x}. With y_2 = e^{-x}, W = -2: u_1' = 2, u_2' = -2e^{2x}, u_1 = 2x, u_2 = -e^{2x}. The trial Axe^{x} gives 2Ae^{x} = 4e^{x}, A = 2. The answers differ by -e^{x}, a homogeneous solution, as Result 1 allows.
  12. The largest distance is \displaystyle \frac{1}{2\pi} \approx 0.159 meters, at \displaystyle t = \frac{\pi}{4} and \displaystyle t = \frac{3\pi}{4}. By Example 6, \displaystyle y = \frac{t}{\pi} - \frac{\sin 2t}{2\pi} while \displaystyle h = \frac{t}{\pi}. Then y(\pi) = 1, and \displaystyle y'(t) = \frac{1 - \cos 2t}{\pi} gives y'(\pi) = 0.
  13. The amplitude is \displaystyle \frac{2}{\pi} \approx 0.637 meters, from \displaystyle \frac{|\sin T|}{T}. There is no swing exactly when \sin T = 0: T = k\pi for a positive integer k. For T > 0, |\sin T| < T: if T > 1 because |\sin T| \le 1, and if T \le 1 because \sin T = \int_0^T \cos u\,du with \cos u < 1 for u > 0. So \displaystyle \frac{|\sin T|}{T} < 1, the amplitude of the sudden move.
  14. Antiderivatives of one function on an interval differ by constants, so the other choices are u_1 + c_1, u_2 + c_2, and y_p becomes y_p + c_1y_1 + c_2y_2, which differs from y_p by a solution of L[y] = 0. Every pair c_1, c_2 occurs, and by the last sentence of Result 1 every solution of L[y] = f on I has this form. \square
  15. Here g(x) = \sin x, so by Results 2 and 3 every solution is y = c_1\cos x + c_2\sin x + \int_0^x \sin(x - s)\,f(s)\,ds. For x \ge 0, |\sin(x - s)| \le 1 gives |y(x)| \le |c_1| + |c_2| + \int_0^\infty |f(s)|\,ds, a bound independent of x. \square
  16. By the subtraction formula, \displaystyle y = \frac{1}{\omega}\big(\sin\omega x\,A(x) - \cos\omega x\,B(x)\big) with A(x) = \int_0^x \cos(\omega s)\,f(s)\,ds and B(x) = \int_0^x \sin(\omega s)\,f(s)\,ds, and by the fundamental theorem A' = \cos(\omega x)\,f, B' = \sin(\omega x)\,f. In y' the two terms carrying f cancel, leaving y' = \cos\omega x\,A + \sin\omega x\,B. Differentiating again, the terms carrying f add to (\cos^2\omega x + \sin^2\omega x)\,f = f, so y'' = -\omega\sin\omega x\,A + \omega\cos\omega x\,B + f, which is -\omega^2y + f. As A(0) = B(0) = 0, y(0) = 0 and y'(0) = 0. \square
  17. Take y_1 = \cosh kx, y_2 = \sinh kx, with W = k. By the subtraction formula for \sinh, \displaystyle G(x, s) = \frac{\sinh k(x - s)}{k}, so \displaystyle y(x) = \int_0^x \frac{\sinh k(x - s)}{k}\,f(s)\,ds by Result 2, and for 0 \le s \le x the integrand is \ge 0. By the proof of Result 1, y' = u_1y_1' + u_2y_2' with \displaystyle u_1 = -\int_0^x \frac{\sinh(ks)\,f(s)}{k}\,ds, \displaystyle u_2 = \int_0^x \frac{\cosh(ks)\,f(s)}{k}\,ds, so y'(x) = \int_0^x \cosh k(x - s)\,f(s)\,ds \ge 0. \square
  18. Let y_p = \sum_j u_jy_j. By the product rule and the first condition, y_p' = \sum_j u_jy_j'; differentiating and using the second, y_p'' = \sum_j u_jy_j''; once more, with the third, y_p''' = f + \sum_j u_jy_j'''. These exist since each u_j is differentiable and each y_j three times differentiable. So M[y_p] = f + \sum_j u_j\,M[y_j] = f. \square
math.wiki · CC-BY-SA-4.0