Differential Equations › Section G · Higher order differential equations › Topic G.3

Undetermined coefficients and variation of parameters

Both methods for finding a particular solution carry over to any order. Undetermined coefficients needs only one extension, a modification power that can now exceed 2, while variation of parameters becomes a determinant formula.

Finding one particular solution

The topic Basic concepts for nth order equations showed that every solution of a nonhomogeneous linear equation is \(y = y_c + y_p\): the complementary solution plus any one particular solution. The topic Homogeneous nth order equations supplies \(y_c\) for constant coefficients. What remains is \(y_p\), and the two second order methods, from the topics Undetermined coefficients and Variation of parameters, both still work. Undetermined coefficients is quick but limited to constant coefficients and forcing terms built from polynomials, exponentials, sines and cosines. Variation of parameters always gives a formula, at the cost of determinants and integrals.

Undetermined coefficients and the modification power

Guess \(y_p\) in the same form as the forcing \(g(t)\), exactly as before: \(Ae^{2t}\) for \(e^{2t}\), a full polynomial of the same degree for a polynomial, and both \(A\cos \beta t + B\sin \beta t\) for either trig function. Always find \(y_c\) first. If any term of a guess already appears in \(y_c\), multiply that part of the guess by \(t^s\), with \(s\) the smallest power that removes every overlap. That \(s\) equals the multiplicity of the matching characteristic root: of \(\alpha\) for \(e^{\alpha t}\) times a polynomial, and of \(\alpha + \beta i\) when sines and cosines are involved. In second order equations \(s\) was at most 2; now it can be as large as the order.

Worked example: a triple root

Solve

y''' - 3y'' + 3y' - y = 12e^t + 4t

1. The characteristic polynomial is \((r - 1)^3\), so \(y_c = (c_1 + c_2t + c_3t^2)e^t\). 2. For \(12e^t\), the guesses \(Ae^t\), \(Ate^t\) and \(At^2e^t\) all lie in \(y_c\), so use \(At^3e^t\) (\(s = 3\)). For \(4t\), use \(Bt + C\); 0 is not a root, so no change. 3. Differentiating \(At^3e^t\) three times and substituting, every term cancels except \(6Ae^t\), so \(6A = 12\) and \(A = 2\). 4. For \(Bt + C\), only \(3y' - y\) survives: \(3B - Bt - C = 4t\), so \(B = -4\) and \(C = 3B = -12\). 5. Answer:

y = (c_1 + c_2t + c_3t^2)e^t + 2t^3e^t - 4t - 12

Worked example: two separate modifications

Find \(y_p\) for \(y^{(4)} - y'' = t + e^{-t}\). 1. The characteristic polynomial is \(r^2(r - 1)(r + 1)\): root 0 is double, and \(\pm 1\) are simple, so \(y_c\) is built from \(1\), \(t\), \(e^t\) and \(e^{-t}\). 2. For \(t\), the plain guess \(At + B\) duplicates both 1 and \(t\); multiplying by \(t^2\) gives \(At^3 + Bt^2\). 3. For \(e^{-t}\), the root \(-1\) is simple, so use \(Cte^{-t}\). 4. Substitute: \(At^3 + Bt^2\) gives \(-6At - 2B = t\), so \(A = -\frac{1}{6}\) and \(B = 0\). Since \((te^{-t})'' = (t - 2)e^{-t}\) and \((te^{-t})^{(4)} = (t - 4)e^{-t}\), the last part gives \(-2Ce^{-t} = e^{-t}\), so \(C = -\frac{1}{2}\). 5. So

y_p = -\frac{t^3}{6} - \frac{1}{2}te^{-t}

Each part of the guess gets its own power.

Worked example: a cantilever beam

A horizontal beam of length \(L\), clamped at \(x = 0\) and free at \(x = L\), carries a uniform load \(w\) per unit length. Its downward deflection satisfies \(EI\,y^{(4)} = w\), where \(E\) measures the material's stiffness and \(I\) depends on the cross-section. 1. Root 0 has multiplicity 4, so \(y_c\) is a cubic, \(c_1 + c_2x + c_3x^2 + c_4x^3\), and the constant forcing needs \(s = 4\): \(y_p = \frac{w x^4}{24EI}\). 2. The clamped end gives \(y(0) = y'(0) = 0\), so \(c_1 = c_2 = 0\). 3. The free end carries no bending moment or shear: \(y''(L) = y'''(L) = 0\), which give \(c_4 = -\frac{wL}{6EI}\) and \(c_3 = \frac{wL^2}{4EI}\). 4. So

y = \frac{w x^2(x^2 - 4Lx + 6L^2)}{24EI}

and the tip sags by \(\frac{wL^4}{8EI}\). Doubling the length multiplies that by 16.

Variation of parameters for order n

Start from a fundamental set \(y_1, \ldots, y_n\) of the homogeneous equation in standard form, with leading coefficient 1, and look for \(y_p = u_1y_1 + \cdots + u_ny_n\). Imposing \(n - 1\) simplifying conditions on the \(u_k'\), as in the second order derivation, plus the equation itself, gives \(n\) linear equations for \(u_1', \ldots, u_n'\). Cramer's rule solves them:

u_k' = \frac{g(t)\,W_k(t)}{W(t)}

where \(W\) is the Wronskian and \(W_k\) is \(W\) with its \(k\)th column replaced by \((0, \ldots, 0, 1)\). Integrate each \(u_k'\), with no constants needed, and assemble \(y_p\). If the leading coefficient is not 1, divide it out before reading off \(g\).

Worked example: a forcing that guessing cannot handle

Find \(y_p\) for \(y''' + y' = \sec t\) on the interval \((-\frac{\pi}{2}, \frac{\pi}{2})\). No guess works for \(\sec t\). 1. The roots of \(r^3 + r\) are \(0, \pm i\), giving \(y_1 = 1\), \(y_2 = \cos t\), \(y_3 = \sin t\). 2. Expanding along the first column, \(W = \sin^2 t + \cos^2 t = 1\). 3. Replacing columns by \((0, 0, 1)\): \(W_1 = 1\), \(W_2 = -\cos t\), \(W_3 = -\sin t\). 4. So \(u_1' = \sec t\), \(u_2' = -1\) and \(u_3' = -\tan t\), which integrate to \(\ln|\sec t + \tan t|\), \(-t\) and \(\ln|\cos t|\). 5. Assemble:

y_p = \ln|\sec t + \tan t| - t\cos t + \sin t\,\ln|\cos t|

Substituting this back confirms it.

Which method to use

Use undetermined coefficients whenever the coefficients are constant and the forcing is a polynomial, exponential, sine or cosine, or a sum or product of these: it needs only differentiation and algebra. Use variation of parameters for anything else, such as \(\sec t\), \(\ln t\) or \(\frac{1}{t}\), or when the coefficients vary but a fundamental set is known. Its price grows quickly with the order: \(n + 1\) determinants of size \(n\), then \(n\) integrals, some of which may have no closed form. In that case the formula still defines \(y_p\), and the integrals can be evaluated numerically.

Common mistakes to avoid

First, choosing the guess before finding \(y_c\), which hides the overlaps. Second, stopping at \(t\) or \(t^2\) out of second order habit: with a triple root you need \(t^3\). Third, multiplying the whole guess by \(t^s\) when only one part overlaps; in the two-modification example the \(e^{-t}\) part needed only \(t\). Fourth, forgetting to divide by the leading coefficient before using the variation of parameters formula. Fifth, placing the 1 in the wrong row of \(W_k\): it goes in the last row, the row of \((n - 1)\)th derivatives.

Key takeaways

Every solution is \(y_c + y_p\), and \(y_c\) comes first. Undetermined coefficients guesses \(y_p\) in the form of the forcing and multiplies any overlapping part by \(t^s\), where \(s\) is the multiplicity of the matching root and can exceed 2. Variation of parameters gives \(u_k' = \frac{gW_k}{W}\) for any continuous forcing, with \(W_k\) the Wronskian with column \(k\) replaced by \((0, \ldots, 0, 1)\). Fourth order beam problems show both ideas at work: \(EIy^{(4)} = w\) needs \(s = 4\), and a cantilever's tip deflection is \(\frac{wL^4}{8EI}\).

Flashcards

Flashcard

General solution

y = y_c + y_p

Find y_c first, always.

Test your understanding

Question 1 of 5

What is the correct form of the guess for a particular solution of y‴ - 3y″ + 3y′ - y = 5eᵗ?

Question 2 of 5

What is a particular solution of y⁽⁴⁾ - y″ = 6?

Question 3 of 5

In variation of parameters for order n, what is uₖ′?

Question 4 of 5

A uniformly loaded cantilever has tip deflection wL⁴/(8EI). If its length doubles and nothing else changes, the tip deflection is multiplied by what?

Question 5 of 5

Which method suits y‴ + y′ = sec t, and why?

← Previous topicNext topic →