Differential Equations › Section E · Systems of differential equations › Topic E.9

Laplace transforms for systems

The Laplace transform turns a single differential equation into algebra, and it does the same for a whole system at once. The initial conditions are built in from the start, and switched or sudden inputs cause no extra trouble.

Why transform a system

The eigenvalue method finds the general solution of a system first and fits the initial conditions at the end. Laplace transforms reverse that order. Transform every equation, and each unknown function \(x_k(t)\) becomes an unknown transform \(X_k(s)\), with its initial value already included. What remains is a set of linear algebraic equations in \(X_1(s), X_2(s), \ldots\), solved by ordinary elimination. Inverse transforms, using partial fractions, then give the solution. The method suits initial value problems with forcing, especially forcing that switches on or off, which the topics Problems with step functions and The Dirac delta function handled for single equations.

Transforming each equation

Two facts from the topic Computing Laplace transforms do almost all the work. The transform is linear, so it can be applied term by term. And a derivative becomes

\mathcal{L}\{x'\} = sX(s) - x(0)

For a system such as \(x_1' = 3x_1 - 2x_2\), \(x_2' = 2x_1 - 2x_2\), each equation transforms separately: the first becomes \(sX_1 - x_1(0) = 3X_1 - 2X_2\). Gather the unknowns on the left and the known terms on the right. The system need not be written in matrix form for this, and it need not be first order: a second derivative becomes \(s^2X - sx(0) - x'(0)\), so mixed-order systems transform just as easily.

The algebraic system and its determinant

For a first order system \(\mathbf{x}' = A\mathbf{x} + \mathbf{g}(t)\), transforming every row gives

(sI - A)\mathbf{X}(s) = \mathbf{x}(0) + \mathbf{G}(s)

so \(\mathbf{X}(s) = (sI - A)^{-1}[\mathbf{x}(0) + \mathbf{G}(s)]\). For two equations, elimination or Cramer's rule is quickest. Either way, every \(X_k\) has the denominator \(\det(sI - A)\), which is the characteristic polynomial of \(A\). Its roots are the eigenvalues, so the exponentials in the answer are exactly the ones the eigenvalue method would give, plus any coming from the forcing. In fact \((sI - A)^{-1}\) is the transform of the matrix exponential \(e^{At}\) from the topic Nonhomogeneous systems.

Worked example: an unforced system

Solve \(x_1' = 3x_1 - 2x_2\), \(x_2' = 2x_1 - 2x_2\), with \(x_1(0) = 3\) and \(x_2(0) = 3\). 1. Transform and rearrange: \((s - 3)X_1 + 2X_2 = 3\) and \(-2X_1 + (s + 2)X_2 = 3\). 2. The determinant is

(s - 3)(s + 2) + 4 = (s - 2)(s + 1)

3. Cramer's rule:

X_1 = \frac{3(s + 2) - 6}{(s - 2)(s + 1)} = \frac{3s}{(s - 2)(s + 1)}

4. Partial fractions: \(X_1 = \frac{2}{s - 2} + \frac{1}{s + 1}\), so \(x_1 = 2e^{2t} + e^{-t}\). 5. Check: \(x_1(0) = 3\), and the exponents 2 and \(-1\) are the eigenvalues of the coefficient matrix, as the determinant predicted.

Saving a second inversion

Once one unknown is found, an original equation often gives the other without more partial fractions. In the example, the first equation rearranges to \(x_2 = \frac{1}{2}(3x_1 - x_1')\). With \(x_1' = 4e^{2t} - e^{-t}\), this gives

x_2 = \tfrac{1}{2}(6e^{2t} + 3e^{-t} - 4e^{2t} + e^{-t}) = e^{2t} + 2e^{-t}

and \(x_2(0) = 3\), as required. If the only available equation gives \(x_2'\) rather than \(x_2\), integrate instead, and use the initial value \(x_2(0)\) to fix the constant of integration. Solving for \(X_2\) directly, \(X_2 = \frac{3s - 3}{(s - 2)(s + 1)}\), gives the same answer and makes a useful check.

Worked example: a constant input

Two compartments start empty. A substance enters the first at a constant rate: \(x_1' = -3x_1 + x_2 + 6\), \(x_2' = x_1 - 3x_2\). 1. Transform with zero initial values: \((s + 3)X_1 - X_2 = \frac{6}{s}\) and \(-X_1 + (s + 3)X_2 = 0\). 2. The second gives \(X_1 = (s + 3)X_2\); substituting, \([(s + 3)^2 - 1]X_2 = \frac{6}{s}\), so \(X_2 = \frac{6}{s(s + 2)(s + 4)}\). 3. Partial fractions:

X_2 = \frac{3/4}{s} - \frac{3/2}{s + 2} + \frac{3/4}{s + 4}

4. Invert: \(x_2 = \frac{3}{4} - \frac{3}{2}e^{-2t} + \frac{3}{4}e^{-4t}\). 5. From the second equation,

x_1 = x_2' + 3x_2 = \frac{9}{4} - \frac{3}{2}e^{-2t} - \frac{3}{4}e^{-4t}

6. Check: both start at 0, and as \(t \to \infty\) they approach \((\frac{9}{4}, \frac{3}{4})\), the equilibrium \(-A^{-1}\mathbf{b}\).

Switched and impulsive inputs

Suppose the inflow in the last example switches on only at \(t = 2\), so the forcing is \(6u_2(t)\), where \(u_c(t)\) is the unit step that turns on at \(t = c\). Its transform is \(\frac{6e^{-2s}}{s}\), so every transform gains the factor \(e^{-2s}\). Because the initial values are zero, the solution is the earlier one delayed by 2: \(x_k(t) = u_2(t)f_k(t - 2)\), where \(f_k\) is the solution just found. An impulse \(k\delta(t - c)\) in an equation transforms to \(ke^{-cs}\). An impulse of size \(k\) at \(t = 0\) in the \(x_1\) equation has the same effect as adding \(k\) to \(x_1(0)\): the dose arrives all at once.

Choosing a method

For a homogeneous system where the general solution is wanted, eigenvalues and eigenvectors are usually quicker, and they show the geometry of the phase plane. Laplace transforms are the better tool for an initial value problem with step, impulse or piecewise forcing, and for systems that mix first and second derivatives. Both methods must give the same answer, so solving a small problem both ways is a strong check. The Laplace method works most smoothly with constant coefficients and initial values given at \(t = 0\), and it produces one particular solution rather than a general family, so it complements the eigenvalue method rather than replacing it.

Common mistakes to avoid

First, a sign slip in the derivative rule: it is \(sX - x(0)\), not \(sX + x(0)\). Second, dropping a forcing term while transforming, or transforming a constant input 6 as 6 rather than \(\frac{6}{s}\). Third, forgetting that a step-delayed answer must carry \(u_c(t)\) and the shift \(t - c\) together. Fourth, recovering \(x_2\) by integration and leaving out the constant fixed by \(x_2(0)\). Fifth, misreading a root: a factor \(s + 2\) in the denominator gives \(e^{-2t}\), not \(e^{2t}\).

Key takeaways

Transform each equation using linearity and \(\mathcal{L}\{x'\} = sX - x(0)\), so the initial conditions enter at once. Solve the resulting linear equations for the transforms by elimination or Cramer's rule; in matrix form \(\mathbf{X} = (sI - A)^{-1}[\mathbf{x}(0) + \mathbf{G}]\), and the denominator is the characteristic polynomial. Invert with partial fractions, and use an original equation to get the second unknown cheaply. Step forcing introduces \(e^{-cs}\) and a delayed solution; impulses transform to \(ke^{-cs}\).

Flashcards

Flashcard

ℒ{x′}

sX(s) − x(0)

Each initial value enters as soon as you transform.

Test your understanding

Question 1 of 5

In a system, x₁(0) = 4. What is the Laplace transform of x₁′?

Question 2 of 5

A system x′ = Ax with x(0) = x₀ is transformed. Which expression is X(s)?

Question 3 of 5

The coefficient matrix A of a homogeneous system has eigenvalues 1 and −4. Which exponentials can appear in the solution found by Laplace transforms?

Question 4 of 5

Solving a transformed system gives X₁(s) = 5 / ((s − 1)(s + 4)). What is x₁(t)?

Question 5 of 5

In a system with zero initial values, a forcing term 6 switches on at t = 2. What does it transform to?

← Previous topicNext topic →