Undetermined coefficients

When the right side of ay’’ + by’ + cy = f is a polynomial, an exponential, a sine or cosine, or a product of these, a particular solution can be found with the same shape. Which shape exactly, and why does the guess always work?

Where you are 5 in Second-order linear equations

Why this exists

By Result 1 of Constant-coefficient equations, once one solution y_p of a forced equation a y'' + b y' + c y = f is known, every solution is y_p + y_h with y_h a homogeneous solution, and Complex and repeated roots found every y_h. What remains is one y_p.

For polynomials, exponentials, sines and cosines, and their products, every derivative of f has the shape of f. So a guess of that shape with unknown coefficients turns the differential equation into linear equations for numbers. This lesson says exactly which shape to guess, including the factor x^m needed when the forcing already solves the homogeneous equation, and proves that the guess always succeeds, with exactly one answer. Throughout, y is a function of x, or of t when the variable is time.

Definition

In plain words: a forcing term of exponential type is one whose derivatives never leave a finite list of functions, and the method guesses a combination from that list.

Fix real a, b, c with a \ne 0 and write

\begin{aligned} L[y] &= a y'' + b y' + c y, \\ p(r) &= a r^2 + b r + c, \end{aligned}

so p is the characteristic polynomial and p'(r) = 2ar + b. For a complex number s, the multiplicity m of s as a root of p is 0 if p(s) \ne 0, 1 if p(s) = 0 \ne p'(s), and 2 if p(s) = p'(s) = 0. Since p'' = 2a \ne 0 there is no other case. A complex s with nonzero imaginary part has m \le 1, because the other root of p is \bar{s} \ne s.

An exponential polynomial is a finite sum of functions x^k e^{\alpha x}\cos\beta x and x^k e^{\alpha x}\sin\beta x, with k \ge 0 an integer and \alpha, \beta real. Grouping terms by \alpha and by \beta \ge 0 (cosine is even and sine is odd) writes it as a sum of blocks

f(x) = e^{\alpha x}\big(P_1(x)\cos\beta x + P_2(x)\sin\beta x\big),

with real polynomials P_1, P_2, not both 0. The block’s exponent is s = \alpha + \beta i and its degree d is the larger of the degrees of P_1 and P_2. When \beta = 0 the block is e^{\alpha x}P_1(x).

The trial function for that block is

y_p = x^m e^{\alpha x}\big(A(x)\cos\beta x + B(x)\sin\beta x\big),

where m is the multiplicity of s and A, B are polynomials of degree at most d whose 2(d + 1) coefficients are unknown; for \beta = 0 it is x^m e^{\alpha x}A(x), with d + 1 unknowns. These are the undetermined coefficients. The method of undetermined coefficients substitutes the trial into L[y] = f and matches the coefficients of each x^k e^{\alpha x}\cos\beta x and x^k e^{\alpha x}\sin\beta x.

block fstrial when m = 0
P(x) of degree d0A(x)
k\,e^{\alpha x}\alphaA\,e^{\alpha x}
k_1\cos\beta x + k_2\sin\beta x\beta iA\cos\beta x + B\sin\beta x

When m \ge 1 each trial is multiplied by x^m. A cosine alone in f still needs both A and B.

Intuition

Differentiating x^k e^{sx} gives k\,x^{k-1}e^{sx} + s\,x^k e^{sx}: the exponential stays and the power of x does not rise. So L maps the functions Q(x)\,e^{sx} with \deg Q \le n, a space of dimension n + 1 (Bases and dimension, proved), into itself. In a basis that map is a square matrix, and a square matrix with only 0 in its kernel is invertible, so the map is onto (The invertible matrix theorem). So when p(s) \ne 0 every P(x)\,e^{sx} is reached, from a Q of the same degree.

When s is a root of multiplicity m, L kills e^{sx}, and x\,e^{sx} too if m = 2. Then L lowers degrees by m, and the cure is to start the guess m degrees higher, with the factor x^m. Sines and cosines enter through \cos\beta x = \operatorname{Re} e^{i\beta x}: since a, b, c are real, the real part of a complex solution is a real solution.

Results

1. Superposition of forcing. If L[y_j] = f_j for j = 1, \dots, n on an interval, then L[y_1 + \dots + y_n] = f_1 + \dots + f_n there. If L[y_p] = f, the solutions of L[y] = f on an open interval are exactly y_p + y_h with L[y_h] = 0.

2. Exponential shift. For a complex number s and a twice differentiable complex-valued function Q,

\begin{aligned} L\big[Q\,e^{sx}\big] &= \big(a Q'' + p'(s)\,Q' \\ &\qquad + p(s)\,Q\big)\,e^{sx}. \end{aligned}

3. Complex exponential forcing. Let P be a polynomial with complex coefficients, of degree d, and let s be a complex number of multiplicity m as a root of p. There is exactly one polynomial Q of degree at most d with

L\big[x^m Q(x)\,e^{sx}\big] = P(x)\,e^{sx},

and its degree is exactly d. If s and the coefficients of P are real, so are those of Q.

4. Real forcing. Let f be a block with exponent s = \alpha + \beta i, \beta > 0, degree d, and let m be the multiplicity of s (0 or 1). There is exactly one pair of real polynomials A, B of degree at most d for which the trial function solves L[y] = f. For \beta = 0 the same holds with B absent.

5. Every solution. Let f be an exponential polynomial and y_p the sum of the trial solutions of its blocks. Then y_p solves L[y] = f on \mathbb{R}, and every solution on an open interval is the restriction of y_p + c_1y_1 + c_2y_2 for exactly one pair c_1, c_2, where y_1, y_2 are the homogeneous solutions of the table in Complex and repeated roots. Every initial value problem y(x_0) = k_0, y'(x_0) = k_1 has exactly one solution, and it is defined on \mathbb{R}.

Why it is true

Result 1. L is linear (Result 1 of Constant-coefficient equations), so L\left[\sum y_j\right] = \sum f_j, and L[y] = f exactly when L[y - y_p] = 0. \square

Result 2. The product rule holds for complex-valued functions (expand into real and imaginary parts), and (e^{sx})' = s\,e^{sx} by Result 2 of Complex and repeated roots. So

\begin{aligned} (Qe^{sx})' &= (Q' + sQ)\,e^{sx}, \\ (Qe^{sx})'' &= (Q'' + 2sQ' + s^2Q)\,e^{sx}. \end{aligned}

Collecting, the coefficient of Q' in L[Qe^{sx}]\,e^{-sx} is 2as + b = p'(s) and that of Q is as^2 + bs + c = p(s). \square

Result 3. By Result 2, and since e^{sx} is never 0, the task is to solve T(R) = P for R = x^m Q, where T(R) = a R'' + p'(s)\,R' + p(s)\,R. Write R = \sum_j r_j x^j with r_j = 0 for j < m and for j > d + m; the unknowns r_m, \dots, r_{d+m} are the d + 1 coefficients of Q. The coefficient of x^k in T(R) is

\begin{aligned} &p(s)\,r_k + (k + 1)\,p'(s)\,r_{k+1} \\ &\quad + (k + 2)(k + 1)\,a\,r_{k+2}. \end{aligned}

In each case the first term that survives multiplies r_{k+m}, with a nonzero factor: p(s) when m = 0; (k + 1)\,p'(s) when m = 1, since then p(s) = 0; (k + 2)(k + 1)\,a when m = 2, since then p(s) = p'(s) = 0. So the equation for x^k involves r_{k+m} and higher-indexed unknowns only, and for k > d both sides are 0. Taking k = d, d - 1, \dots, 0 in turn, each equation fixes one more of r_{d+m}, \dots, r_m, uniquely: the system is triangular with nonzero diagonal. So exactly one Q exists, and its top coefficient r_{d+m} is the leading coefficient of P divided by a nonzero number, so \deg Q = d. When s and P are real, every number in this back substitution is real. With P = 0 the same equations force every r_j = 0: T(x^m Q) = 0 only for Q = 0. \square

Result 4. First a lemma: if U, V are real polynomials, \beta > 0, and U(x)\cos\beta x + V(x)\sin\beta x = 0 for all real x, then U = V = 0. At \displaystyle x = \frac{k\pi}{\beta}, \sin\beta x = 0 and \cos\beta x = \pm 1, so U has infinitely many roots and is the zero polynomial. Then V vanishes at every \displaystyle x = \frac{(2k + 1)\pi}{2\beta}, so V = 0. This lemma is also what makes matching coefficients legitimate.

Existence. The real part of (P_1 - iP_2)(\cos\beta x + i\sin\beta x) is P_1\cos\beta x + P_2\sin\beta x, so f = \operatorname{Re}\big((P_1 - iP_2)\,e^{sx}\big), and P_1 - iP_2 has degree d. Result 3 gives Q with L[x^m Q e^{sx}] = (P_1 - iP_2)\,e^{sx}. Because a, b, c are real, L[\operatorname{Re} z] = \operatorname{Re} L[z] (Result 3 of Complex and repeated roots), so y_p = \operatorname{Re}(x^m Q e^{sx}) solves L[y] = f. Writing Q = A - iB with A, B real polynomials of degree at most d,

y_p = x^m e^{\alpha x}\big(A\cos\beta x + B\sin\beta x\big).

Uniqueness. If two pairs work, their difference gives real A, B of degree at most d with L[D] = 0, where D = x^m e^{\alpha x}(A\cos\beta x + B\sin\beta x) = \operatorname{Re}(x^m\tilde{Q}\,e^{sx}) and \tilde{Q} = A - iB. By Result 2, L[D] = \operatorname{Re}\big(T(x^m\tilde{Q})\,e^{sx}\big). Writing T(x^m\tilde{Q}) = U - iV with U, V real, this is e^{\alpha x}(U\cos\beta x + V\sin\beta x) = 0. The lemma gives U = V = 0, so T(x^m\tilde{Q}) = 0 and \tilde{Q} = 0 by the last sentence of the proof of Result 3. For \beta = 0, Result 3 with real s is the claim. \square

Result 5. By Results 3, 4 and 1, y_p solves L[y] = f on \mathbb{R}. If y solves it on an open interval K, then y - y_p solves L[y] = 0 on K, so it is the restriction of c_1y_1 + c_2y_2 for exactly one pair (Result 5 of Constant-coefficient equations, Results 4 and 5 of Complex and repeated roots). The initial conditions read

\begin{aligned} c_1y_1(x_0) + c_2y_2(x_0) &= k_0 - y_p(x_0), \\ c_1y_1'(x_0) + c_2y_2'(x_0) &= k_1 - y_p'(x_0), \end{aligned}

with determinant the Wronskian W(y_1, y_2)(x_0) \ne 0, so exactly one pair fits (Cramer's rule). \square

Worked examples

Example 1

A polynomial forcing with initial conditions

Solve y'' + 3y' + 2y = 4x^2, y(0) = 0, y'(0) = 0.

Solution

Here p(r) = (r + 1)(r + 2), with roots -1 and -2. The forcing is one block with s = 0 and d = 2, and p(0) = 2 \ne 0, so m = 0 and the trial is a full quadratic A(x) = A_2x^2 + A_1x + A_0: the lower terms belong in it although 4x^2 has none. With A' = 2A_2x + A_1 and A'' = 2A_2,

\begin{aligned} L[A] &= 2A_2x^2 + (6A_2 + 2A_1)\,x \\ &\quad + (2A_2 + 3A_1 + 2A_0). \end{aligned}

Matching with 4x^2 from the top degree down, as in the proof of Result 3: 2A_2 = 4, so A_2 = 2; 12 + 2A_1 = 0, so A_1 = -6; 4 - 18 + 2A_0 = 0, so A_0 = 7. By Result 5,

y = c_1e^{-x} + c_2e^{-2x} + 2x^2 - 6x + 7.

The conditions give c_1 + c_2 + 7 = 0 and -c_1 - 2c_2 - 6 = 0, so c_2 = 1 and c_1 = -8:

y = -8e^{-x} + e^{-2x} + 2x^2 - 6x + 7.
Example 2

Two forcing terms, one of them a sine

Find the general solution of y'' - 2y' - 3y = 8e^{x} + 10\sin x.

Solution

Here p(r) = (r - 3)(r + 1). The forcing has two blocks, with exponents 1 and i, neither a root, so m = 0 for both, and by Result 1 they can be handled one at a time.

For 8e^{x} the trial is A\,e^{x}, and Result 2 gives L[A\,e^{x}] = p(1)\,A\,e^{x} = -4A\,e^{x}, so A = -2.

For 10\sin x the trial is A\cos x + B\sin x; the cosine is needed because differentiating the sine produces it. Substituting,

\begin{aligned} L[A\cos x + B\sin x] &= (-4A - 2B)\cos x \\ &\quad + (2A - 4B)\sin x, \end{aligned}

so -4A - 2B = 0 and 2A - 4B = 10, which give A = 1 and B = -2. The route of Result 4 does the same in one line: 10\sin x = \operatorname{Re}(-10i\,e^{ix}) and p(i) = -4 - 2i, so \displaystyle Q = \frac{-10i}{-4 - 2i} = 1 + 2i, and \operatorname{Re}\big((1 + 2i)(\cos x + i\sin x)\big) = \cos x - 2\sin x. So

\begin{aligned} y &= c_1e^{3x} + c_2e^{-x} \\ &\quad - 2e^{x} + \cos x - 2\sin x. \end{aligned}
Example 3

When the forcing solves the homogeneous equation

Find a particular solution of y'' - y' - 6y = 10x\,e^{3x}.

Solution

Here p(r) = (r - 3)(r + 2), so s = 3 is a simple root and m = 1, with p(3) = 0 and p'(3) = 5. The guess without the factor x fails: by Result 2,

L\big[(A_1x + A_0)\,e^{3x}\big] = 5A_1\,e^{3x},

which has no x\,e^{3x} term, because A_0e^{3x} solves the homogeneous equation. The trial with m = 1 is x(A_1x + A_0)\,e^{3x}. With R = A_1x^2 + A_0x, Result 2 gives

\begin{aligned} L[R\,e^{3x}] &= (R'' + 5R')\,e^{3x} \\ &= \big(10A_1x + 2A_1 + 5A_0\big)\,e^{3x}. \end{aligned}

So 10A_1 = 10 and 2A_1 + 5A_0 = 0, that is A_1 = 1 and \displaystyle A_0 = -\frac{2}{5}:

y_p = \left(x^2 - \tfrac{2}{5}x\right)e^{3x}.
Example 4

A short proof: one exponential on the right

Prove that for a complex number s and a constant k, the equation L[y] = k\,e^{sx} has the particular solution

y_p = \frac{k\,e^{sx}}{p(s)}, \quad \frac{k\,x\,e^{sx}}{p'(s)}, \quad \frac{k\,x^2e^{sx}}{2a}

when m = 0, 1, 2 respectively, and use it to solve y'' + 6y' + 9y = 12e^{-3x}.

Solution

Apply Result 2. For m = 0, \displaystyle Q = \frac{k}{p(s)} is constant, so Q' = Q'' = 0 and L[Qe^{sx}] = p(s)\,Q\,e^{sx} = k\,e^{sx}. For m = 1, p(s) = 0 and \displaystyle Q = \frac{kx}{p'(s)} has Q'' = 0, so L[Qe^{sx}] = p'(s)\,Q'\,e^{sx} = k\,e^{sx}. For m = 2, p(s) = p'(s) = 0 and \displaystyle Q = \frac{kx^2}{2a}, so L[Qe^{sx}] = a\,Q''\,e^{sx} = k\,e^{sx}. \square

For y'' + 6y' + 9y = 12e^{-3x}, p(r) = (r + 3)^2 has the double root -3, so m = 2, a = 1, and \displaystyle y_p = \frac{12}{2}x^2e^{-3x} = 6x^2e^{-3x}. With the homogeneous solutions of Complex and repeated roots,

y = (c_1 + c_2x + 6x^2)\,e^{-3x}.
Example 5

Sophia’s buoy in the wave pool

At a water park, Sophia tests a marker buoy in the wave pool. Its height y(t) in meters above its resting level, t seconds after the wave machine starts, obeys

y'' + 4y' + 5y = 0.8\cos t,

where, per kilogram of buoy and in newtons per kilogram, 5y is the restoring force of buoyancy, 4y' the drag of the water and 0.8\cos t the push of the waves. The buoy starts at rest at its resting level. Find y(t), the amplitude and lag of the motion it settles into, and a time after which it stays within 1 millimeter of that motion.

Solution

Here p(r) = (r + 2)^2 + 1, with roots -2 \pm i, and the exponent i of the forcing is not a root. Since 0.8\cos t = \operatorname{Re}(0.8\,e^{it}) and p(i) = 4 + 4i, Result 3 gives \displaystyle Q = \frac{0.8}{4 + 4i} = 0.1 - 0.1i, and

\begin{aligned} y_p &= \operatorname{Re}\big((0.1 - 0.1i)\,e^{it}\big) \\ &= 0.1\cos t + 0.1\sin t. \end{aligned}

So y = e^{-2t}(c_1\cos t + c_2\sin t) + y_p, and the conditions give c_1 + 0.1 = 0 and -2c_1 + c_2 + 0.1 = 0, so c_1 = -0.1 and c_2 = -0.3:

\begin{aligned} y &= 0.1(\cos t + \sin t) \\ &\quad - e^{-2t}(0.1\cos t + 0.3\sin t). \end{aligned}

The steady periodic solution \displaystyle 0.1(\cos t + \sin t) = 0.1\sqrt{2}\cos\left(t - \frac{\pi}{4}\right) has amplitude \displaystyle 0.1\sqrt{2} \approx 0.1414 meters, and the buoy’s crest comes \displaystyle \frac{\pi}{4} \approx 0.785 seconds after the push peaks (Combining a sine and a cosine into one wave). The rest, the transient, is \displaystyle -\sqrt{0.1}\,e^{-2t}\cos(t - \delta) for a fixed \delta, so its size is at most \displaystyle \sqrt{0.1}\,e^{-2t}. That is below 0.001 once \displaystyle e^{2t} > 100\sqrt{10}, that is for \displaystyle t > \frac{1}{2}\ln\left(100\sqrt{10}\right) \approx 2.878 seconds (Figure 1).

2 4 6 8 −0.1 0.1 t y
Figure 1Sophia’s buoy from Example 5: the solution y(t) (solid) and the steady periodic solution 0.1(\cos t + \sin t) (dashed). Their difference shrinks like e^{-2t}.
Example 6

A product forcing at a root

Find a particular solution of y'' + 9y = 12x\sin 3x.

Solution

The roots of r^2 + 9 are \pm 3i. The forcing is one block with exponent s = 3i and degree d = 1, and s is a simple root, so m = 1. The real trial x\big((A_1x + A_0)\cos 3x + (B_1x + B_0)\sin 3x\big) has four unknowns; the route of Result 4 needs two complex ones. Here 12x\sin 3x = \operatorname{Re}(-12ix\,e^{3ix}) and p'(3i) = 6i. With R = Q_1x^2 + Q_0x, Result 2 gives

\begin{aligned} L[R\,e^{3ix}] &= (R'' + 6iR')\,e^{3ix} \\ &= \big(12iQ_1x + 2Q_1 + 6iQ_0\big)\,e^{3ix}. \end{aligned}

Matching with -12ix: 12iQ_1 = -12i, so Q_1 = -1, and 2Q_1 + 6iQ_0 = 0, so \displaystyle Q_0 = \frac{2}{6i} = -\frac{i}{3}. The real part of (-x^2 - \tfrac{i}{3}x)(\cos 3x + i\sin 3x) is

y_p = -x^2\cos 3x + \tfrac{x}{3}\sin 3x.

The forcing grows like x but this response like x^2.

Boundaries

  • The forcing must be an exponential polynomial. If y is one, so is L[y] (Exercise 19), so a forcing such as \tan x, \displaystyle \frac{1}{x} or e^{x^2} has no particular solution of any trial shape; Variation of parameters handles every continuous forcing.
  • The coefficients must be constant. For y'' + x\,y = 1 no polynomial solves: if y has degree k \ge 0, then x\,y has degree k + 1 and y'' degree at most k - 2, so the left side has degree k + 1 \ge 1; and y = 0 gives 0.
  • The multiplicity belongs to the whole exponent s. For y'' + 2y' + 17y, with roots -1 \pm 4i, the forcing e^{-x}\cos 4x needs the factor x, while e^{-x}\cos x and e^{x}\cos 4x do not.
  • Higher order is stated, not proved, here. For order n, Result 2 becomes \displaystyle L[Qe^{sx}] = \sum_{j=0}^{n} \frac{p^{(j)}(s)}{j!}\,Q^{(j)}\,e^{sx} (Exercise 21 proves order 3), m runs up to n, and the triangular argument of Result 3 goes through.

Where this goes next

Variation of parameters finds a particular solution for any continuous forcing, at the cost of two integrals. Springs, damping and resonance reads the steady periodic solution of Sophia’s buoy as an amplitude and phase that depend on the forcing frequency, and the case m = 1 for a cosine as resonance. The Laplace transform solves the same initial value problems by algebra, with m returning as a repeated factor in partial fractions.

Exercises

Choosing the trial

For each equation, give the multiplicity m and the trial function, then find its coefficients.

  1. y'' + 5y' + 4y = x\,e^{-4x}
  2. y'' + 36y = e^{x}\cos 6x
  3. y'' - 2y' + y = x^2e^{x} + 5
  4. y'' + 6y' + 13y = e^{-3x}\sin 2x

Solving

  1. Find the general solution of y'' - 5y' + 6y = 6x + 1.
  2. Solve y'' + 2y' + y = 9e^{2x}, y(0) = 1, y'(0) = 0.
  3. Find a particular solution of y'' + 4y' = 8x.
  4. Solve y'' + y = 6\cos 2x, y(0) = 0, y'(0) = 1.
  5. Find a particular solution of y'' - 2y' + 2y = e^{x}\sin x.
  6. Find a particular solution of y'' + 3y' + 4y = 6\sin x.
  7. Find a particular solution of y'' - 4y = x\,e^{2x}.
  8. Find a particular solution of y'' + 4y' + 4y = 3x\,e^{-2x}.
  9. Solve y'' + 16y = 16x + 17e^{x}, y(0) = 1, y'(0) = 0.
  10. Find a particular solution of y'' - 4y' + 13y = 40\cos x.

The wave pool

  1. Isaac sets the wave machine to add a second motion, so that Sophia’s buoy obeys y'' + 4y' + 5y = 0.8\cos t + 0.65\sin 2t. Find the steady periodic solution and the amplitude of its new part.
  2. With the machine as in Example 5, Grace holds the buoy 0.3 meters above its resting level and lets go from rest at t = 0. Find y(t).

Proofs

  1. Let c \ne 0 and let P be a real polynomial of degree d. Prove that a y'' + b y' + c y = P(x) has exactly one polynomial solution, and that its degree is d.
  2. Let a, b, c > 0 and let k be a constant. Prove that every solution of a y'' + b y' + c y = k tends to \displaystyle \frac{k}{c} as x \to \infty.
  3. Prove that if y is an exponential polynomial, so is L[y], and deduce that y'' + y = \tan x has no solution on \displaystyle \left(-\frac{\pi}{2}, \frac{\pi}{2}\right) that is the restriction of an exponential polynomial.
  4. Let s be a root of p of multiplicity m \ge 1 and Q a polynomial of degree less than m. Prove that L[Q\,e^{sx}] = 0.
  5. Let M[y] = y''' + b_2y'' + b_1y' + b_0y with real b_j and p(r) = r^3 + b_2r^2 + b_1r + b_0. Prove that M[Q\,e^{sx}] = \big(Q''' + \tfrac{1}{2}p''(s)\,Q'' + p'(s)\,Q' + p(s)\,Q\big)\,e^{sx} for complex s and three times differentiable Q, and deduce that if p(r) = (r - s)^3, then \displaystyle \frac{x^3}{6}e^{sx} solves M[y] = e^{sx}.
Answers
  1. m = 1, since p(r) = (r + 1)(r + 4); trial x(A_1x + A_0)\,e^{-4x}. With p'(-4) = -3, R'' - 3R' = -6A_1x + 2A_1 - 3A_0 = x, so \displaystyle y_p = -\left(\frac{x^2}{6} + \frac{x}{9}\right)e^{-4x}.
  2. m = 0, since s = 1 + 6i has p(s) = 1 + 12i; trial e^{x}(A\cos 6x + B\sin 6x). Then \displaystyle Q = \frac{1}{1 + 12i} = \frac{1 - 12i}{145}, so \displaystyle y_p = \frac{e^{x}(\cos 6x + 12\sin 6x)}{145}.
  3. Two blocks. For x^2e^{x}, s = 1 is a double root of (r - 1)^2, so m = 2, the trial is x^2(A_2x^2 + A_1x + A_0)\,e^{x}, and Result 2 reduces it to R'' = x^2, so \displaystyle R = \frac{x^4}{12}. For 5, m = 0 and the trial is a constant C = 5. So \displaystyle y_p = \frac{x^4e^{x}}{12} + 5.
  4. m = 1, since the roots are -3 \pm 2i; trial x\,e^{-3x}(A\cos 2x + B\sin 2x). With e^{-3x}\sin 2x = \operatorname{Re}(-i\,e^{sx}) and p'(s) = 4i, Example 4 gives \displaystyle \frac{-i\,x\,e^{sx}}{4i}, so \displaystyle y_p = -\frac{x}{4}e^{-3x}\cos 2x.
  5. y = c_1e^{2x} + c_2e^{3x} + x + 1. The trial A_1x + A_0 gives 6A_1x - 5A_1 + 6A_0 = 6x + 1.
  6. y = e^{2x} - 2x\,e^{-x}. Here \displaystyle y_p = \frac{9e^{2x}}{p(2)} = e^{2x}, and y = (c_1 + c_2x)\,e^{-x} + e^{2x} with c_1 + 1 = 1, -c_1 + c_2 + 2 = 0.
  7. \displaystyle y_p = x^2 - \frac{x}{2}. Here s = 0 is a simple root of r^2 + 4r, so the trial is x(A_1x + A_0), and 2A_1 + 4(2A_1x + A_0) = 8x.
  8. y = 2\cos x + \sin x - 2\cos 2x. Here \displaystyle y_p = \frac{6\cos 2x}{p(2i)} = -2\cos 2x, then c_1 - 2 = 0 and c_2 = 1.
  9. \displaystyle y_p = -\frac{1}{2}x\,e^{x}\cos x. The roots are 1 \pm i, so s = 1 + i has m = 1, p'(s) = 2i, and \displaystyle \operatorname{Re}\left(\frac{-i\,x\,e^{sx}}{2i}\right) = -\frac{x}{2}e^{x}\cos x.
  10. y_p = \sin x - \cos x. Here p(i) = 3 + 3i and \displaystyle Q = \frac{-6i}{3 + 3i} = -1 - i.
  11. \displaystyle y_p = \frac{(2x^2 - x)\,e^{2x}}{16}. Here m = 1, p'(2) = 4, and R = A_1x^2 + A_0x needs 8A_1x + 2A_1 + 4A_0 = x.
  12. \displaystyle y_p = \frac{1}{2}x^3e^{-2x}. Here p(r) = (r + 2)^2, m = 2, and R'' = 3x with R = x^2(A_1x + A_0) gives \displaystyle R = \frac{x^3}{2}.
  13. \displaystyle y = x + e^{x} - \frac{1}{2}\sin 4x. The trial A_1x + A_0 needs 16A_1x + 16A_0 = 16x, and Example 4 gives \displaystyle \frac{17e^{x}}{p(1)} = e^{x}; then c_1 + 1 = 1 and 4c_2 + 2 = 0.
  14. y_p = 3\cos x - \sin x. Here p(i) = 12 - 4i and \displaystyle Q = \frac{40}{12 - 4i} = 3 + i.
  15. y_p = 0.1\cos t + 0.1\sin t - 0.08\cos 2t + 0.01\sin 2t and the new part has amplitude \displaystyle \sqrt{0.0065} \approx 0.0806 meters. For the new block, 0.65\sin 2t = \operatorname{Re}(-0.65i\,e^{2it}), p(2i) = 1 + 8i, and \displaystyle Q = \frac{-0.65i}{1 + 8i} = -0.08 - 0.01i.
  16. y = 0.1(\cos t + \sin t) + e^{-2t}(0.2\cos t + 0.3\sin t). The conditions give c_1 + 0.1 = 0.3 and -2c_1 + c_2 + 0.1 = 0.
  17. Existence: p(0) = c \ne 0, so s = 0 has m = 0, and Result 3 gives a polynomial Q of degree d with L[Q] = P. Uniqueness: if R is a polynomial solution, L[R - Q] = 0. If R - Q \ne 0, with degree j and leading coefficient \ell, the coefficient of x^j in L[R - Q] is c\,\ell \ne 0, since the derivative terms have degree less than j; a contradiction. So R = Q. \square
  18. The constant \displaystyle \frac{k}{c} solves, so by Result 5 every solution is \displaystyle \frac{k}{c} + y_h with L[y_h] = 0. If b^2 < 4ac, |y_h| \le (|c_1| + |c_2|)\,e^{-bx/(2a)} \to 0. If b^2 = 4ac, y_h = (c_1 + c_2x)\,e^{r_1x} with \displaystyle r_1 = -\frac{b}{2a} < 0, which tends to 0. If b^2 > 4ac, the roots have sum \displaystyle -\frac{b}{a} < 0 and product \displaystyle \frac{c}{a} > 0, so both are negative and y_h \to 0. In every case \displaystyle y \to \frac{k}{c}. \square
  19. Differentiation is linear, so one term at a time suffices. \left(x^ke^{\alpha x}\cos\beta x\right)' = k\,x^{k-1}e^{\alpha x}\cos\beta x + \alpha\,x^ke^{\alpha x}\cos\beta x - \beta\,x^ke^{\alpha x}\sin\beta x (no first term when k = 0), and the sine term gives the same with +\beta\,x^ke^{\alpha x}\cos\beta x. So y', y'' and L[y] are exponential polynomials. These are continuous on \mathbb{R}, so L[y] is bounded on \displaystyle \left[0, \frac{\pi}{2}\right] by the extreme value theorem, while \tan x \to \infty as \displaystyle x \to \frac{\pi}{2} from the left. So L[y] \ne \tan x somewhere in \displaystyle \left[0, \frac{\pi}{2}\right). \square
  20. By Result 2, L[Qe^{sx}] = \big(aQ'' + p'(s)\,Q' + p(s)\,Q\big)\,e^{sx}. If m = 1, Q is constant, so Q' = Q'' = 0, and p(s) = 0. If m = 2, Q has degree at most 1, so Q'' = 0, and p(s) = p'(s) = 0. In both cases L[Qe^{sx}] = 0. \square
  21. Differentiating (Qe^{sx})'' = (Q'' + 2sQ' + s^2Q)\,e^{sx} once more gives (Qe^{sx})''' = (Q''' + 3sQ'' + 3s^2Q' + s^3Q)\,e^{sx}. Adding b_2, b_1, b_0 times the lower derivatives, the coefficient of Q'' is \displaystyle 3s + b_2 = \frac{1}{2}p''(s), since p''(r) = 6r + 2b_2; that of Q' is 3s^2 + 2b_2s + b_1 = p'(s); that of Q is p(s). If p(r) = (r - s)^3, then p(s) = p'(s) = p''(s) = 0, and \displaystyle Q = \frac{x^3}{6} has Q''' = 1, so \displaystyle M\left[\frac{x^3}{6}e^{sx}\right] = e^{sx}. \square
math.wiki · CC-BY-SA-4.0