Differential Equations › Section G · Higher order differential equations › Topic G.4

Laplace, systems, and series for higher order

Laplace transforms, first order systems and power series were developed mostly for second order problems, yet each handles higher order with only small changes. Seeing them side by side also shows that the three methods describe the same underlying structure.

Three tools, one more derivative

Earlier sections built three tools on second order examples: the Laplace transform, in the topic Solving initial value problems; first order systems with eigenvalues, from the topic Systems of differential equations onwards; and power series, in the topic Series solutions near ordinary points. None of them depended on the order being 2. This topic runs each on third order equations and points out the one or two places where something new appears: an extra term in a transform, a larger matrix, an extra free coefficient. Throughout, an equation of order \(n\) still needs exactly \(n\) pieces of initial data.

Transforms of higher derivatives

The rule for transforming derivatives extends one step at a time. Writing \(Y(s)\) for the transform of \(y(t)\),

\mathcal{L}\{y'''\} = s^3Y - s^2y(0) - s\,y'(0) - y''(0)

and in general the transform of \(y^{(n)}\) is \(s^nY\) minus one term for each initial value: \(s^{n-1}y(0)\), then \(s^{n-2}y'(0)\), down to \(y^{(n-1)}(0)\) with no power of \(s\). The powers of \(s\) fall as the derivatives rise. Every initial condition enters automatically, which is the method's great advantage at higher order: there is no \(3 \times 3\) system to solve for constants at the end.

Worked example: a third order problem by Laplace

Solve \(y''' - 4y' = 8\) with \(y(0) = 1\), \(y'(0) = 0\) and \(y''(0) = 4\). 1. Transform:

(s^3Y - s^2 - 4) - 4(sY - 1) = \frac{8}{s}

2. Collect: \((s^3 - 4s)Y\) equals \(s^2 + \frac{8}{s}\), so

Y = \frac{s^3 + 8}{s^2(s - 2)(s + 2)}

3. Since \(s^3 + 8\) factors as \((s + 2)(s^2 - 2s + 4)\), the factor \(s + 2\) cancels. 4. Partial fractions on what remains give

Y = \frac{1}{s - 2} - \frac{2}{s^2}

5. Invert: \(y = e^{2t} - 2t\). Check: \(y(0) = 1\), \(y'(0) = 2 - 2 = 0\), \(y''(0) = 4\), and \(y''' - 4y'\) equals \(8e^{2t} - 8e^{2t} + 8 = 8\).

From one equation to a system

Any equation of order \(n\) can be rewritten as \(n\) first order equations. For \(y''' - 2y'' - y' + 2y = 0\), set \(x_1 = y\), \(x_2 = y'\) and \(x_3 = y''\). Then \(x_1' = x_2\), \(x_2' = x_3\), and the equation itself gives \(x_3' = -2x_1 + x_2 + 2x_3\). In matrix form \(\mathbf{x}' = A\mathbf{x}\) with

A = \begin{pmatrix} 0 & 1 & 0 \\ 0 & 0 & 1 \\ -2 & 1 & 2 \end{pmatrix}

This companion matrix has 1s just above the diagonal and the negated coefficients in its last row. Its characteristic polynomial, in the variable \(\lambda\), is the equation's own characteristic polynomial, so its eigenvalues \(1, -1, 2\) are the characteristic roots found in the topic Basic concepts for nth order equations.

Three by three systems

For \(\mathbf{x}' = A\mathbf{x}\) with a real \(3 \times 3\) matrix, there are three eigenvalues counted with multiplicity, and each simple eigenvalue \(\lambda\) with eigenvector \(\mathbf{v}\) gives the solution \(e^{\lambda t}\mathbf{v}\). Because complex eigenvalues of a real matrix come in conjugate pairs, the possibilities are: three real eigenvalues; or one real eigenvalue and one complex pair. Never exactly one complex eigenvalue. Repeated real eigenvalues need the extra care from the topic Repeated eigenvalues, depending on how many independent eigenvectors they have. A complex pair is handled as in the topic Complex eigenvalues: take the real and imaginary parts of \(e^{\lambda t}\mathbf{v}\) for one eigenvalue of the pair.

Worked example: a real and a complex eigenvalue

Solve \(\mathbf{x}' = A\mathbf{x}\) with rows \((2, 0, 0)\), \((0, 1, -1)\) and \((0, 1, 1)\). 1. Expanding along the first row, the characteristic polynomial is

(2 - \lambda)\big[(1 - \lambda)^2 + 1\big]

so \(\lambda = 2\) or \(\lambda = 1 \pm i\). 2. For \(\lambda = 2\), the eigenvector is \((1, 0, 0)\). 3. For \(\lambda = 1 + i\), an eigenvector is \((0, i, 1)\). 4. Multiply out \(e^{(1+i)t}(0, i, 1)\) with Euler's formula; its imaginary part is \(e^t(0, \cos t, \sin t)\) and its real part is \(e^t(0, -\sin t, \cos t)\). 5. General solution:

\mathbf{x} = c_1e^{2t}\begin{pmatrix}1\\0\\0\end{pmatrix} + c_2e^t\begin{pmatrix}0\\ \cos t\\ \sin t\end{pmatrix} + c_3e^t\begin{pmatrix}0\\ -\sin t\\ \cos t\end{pmatrix}

Check the second: \(x_2' = e^t(\cos t - \sin t)\), which is \(x_2 - x_3\).

Series for a third order equation

The series method needs no change either. For \(y''' - xy = 0\) about the ordinary point 0, substitute \(y = \sum c_n x^n\); the third derivative shifts to \(\sum (n+3)(n+2)(n+1)c_{n+3}x^n\). 1. The \(x^0\) coefficient gives \(6c_3 = 0\). 2. For \(n \ge 1\):

c_{n+3} = \frac{c_{n-1}}{(n+3)(n+2)(n+1)}

3. So \(c_4 = \frac{c_0}{24}\), \(c_5 = \frac{c_1}{60}\), \(c_6 = \frac{c_2}{120}\) and \(c_7 = 0\). The new feature is a third free coefficient: \(c_0 = y(0)\), \(c_1 = y'(0)\) and \(c_2 = \frac{y''(0)}{2}\), the last divided by \(2!\) as every Taylor coefficient must be. The solution splits into three series, one built on each free coefficient.

Choosing a method

For constant coefficients with given initial values and a forcing that has a transform, especially a switched or impulsive one, Laplace is usually fastest: initial conditions are built in. The systems view is the natural home for several coupled quantities and is how most numerical software handles higher order equations, because standard solvers expect first order systems. Series are the tool for variable coefficients near an ordinary point, at any order. Whichever you use, the answer must contain \(n\) free constants, or satisfy \(n\) initial conditions, for an equation of order \(n\).

Common mistakes to avoid

First, wrong powers of \(s\) in the transform of \(y'''\): \(y(0)\) goes with \(s^2\), \(y'(0)\) with \(s\), and \(y''(0)\) stands alone. Second, forgetting to negate the coefficients in the last row of the companion matrix, or putting them in the wrong order. Third, using both the real and imaginary parts of both members of a complex pair; one eigenvalue of the pair gives both real solutions. Fourth, treating \(c_2\) as \(y''(0)\) in a series; it is \(\frac{y''(0)}{2}\). Fifth, missing the low-order equation \(6c_3 = 0\) that sits outside the general recurrence.

Key takeaways

The Laplace transform of \(y^{(n)}\) is \(s^nY\) minus the initial values multiplied by falling powers of \(s\), so all \(n\) initial conditions are built in. An equation of order \(n\) becomes an \(n \times n\) first order system by taking \(y\) and its first \(n - 1\) derivatives as unknowns; the companion matrix's eigenvalues are the characteristic roots. Real \(3 \times 3\) systems have three real eigenvalues or one real eigenvalue and a complex pair. Series solutions of order \(n\) have \(n\) free coefficients, with \(c_k = \frac{y^{(k)}(0)}{k!}\).

Flashcards

Flashcard

ℒ{y‴}

s³Y - s²y(0) - s y′(0) - y″(0)

Powers of s fall as the derivatives rise.

Test your understanding

Question 1 of 5

What is the Laplace transform of y‴?

Question 2 of 5

What is the last row of the companion matrix of y‴ + 5y″ - 2y′ + 7y = 0, with x₁ = y, x₂ = y′, x₃ = y″?

Question 3 of 5

Which list cannot be the eigenvalues of a real 3 × 3 matrix?

Question 4 of 5

For the series solution of y‴ - xy = 0 with y(0) = 2, y′(0) = 0 and y″(0) = 0, what is c₄?

Question 5 of 5

In a power series solution of a third order equation about an ordinary point 0, which coefficients are free?

← Previous topicNext topic →