Differential Equations › Section D · Laplace transforms › Topic D.8

The Dirac delta function

A hammer blow, a lightning strike on a power line, a sudden dose of medicine: some inputs are huge but last almost no time. The Dirac delta function models them as perfect instantaneous impulses, and its Laplace transform is the simplest of all.

Sudden blows

Step functions model inputs that switch on and stay on. They cannot describe a force that acts for a fraction of a second, such as a bat hitting a ball. What matters about such a force is not its exact shape, which is hard to measure anyway, but its total effect: the area under the force-time graph, which physicists call the impulse. The Dirac delta function is an idealisation that keeps only that total. It describes a unit impulse delivered at a single instant. Named after the physicist Paul Dirac, it is the standard model for shocks in mechanics, circuits and signal processing.

A tall, thin pulse

Build the delta as a limit. For a small \(\varepsilon > 0\), let

d_\varepsilon(t) = \frac{1}{\varepsilon}\big(u_a(t) - u_{a + \varepsilon}(t)\big)

a pulse of height \(\frac{1}{\varepsilon}\) and width \(\varepsilon\) starting at \(t = a\). Its area is always 1. As \(\varepsilon \to 0\) the pulse gets taller and thinner but keeps unit area. The delta, written \(\delta(t - a)\), is what this family approaches. Its transform shows how the limit behaves: by the topic Step functions,

\mathcal{L}\{d_\varepsilon\} = e^{-as}\,\frac{1 - e^{-\varepsilon s}}{\varepsilon s}

and the fraction tends to 1 as \(\varepsilon \to 0\), for example by L'Hôpital's rule.

The defining properties

The delta is pinned down by three properties. First, \(\delta(t - a) = 0\) for every \(t \ne a\). Second, its integral over any interval with \(a\) strictly inside equals 1. Third, the sifting property: for any \(f\) continuous at \(a\),

\int_{a - \varepsilon}^{a + \varepsilon} f(t)\,\delta(t - a)\,dt = f(a)

for every \(\varepsilon > 0\), and the same holds over any interval with \(a\) inside. No ordinary function can do this: a function that is zero except at one point has integral 0. The delta is a generalised function, or distribution, defined by what it does inside integrals. It is used only in that role.

Worked example: sifting

Evaluate three integrals. 1. \(\int_0^{2\pi} \cos t\,\delta(t - \pi)\,dt\): the impulse at \(\pi\) lies inside \([0, 2\pi]\), so the integral picks out \(\cos\pi = -1\). 2. \(\int_0^5 t^2\,\delta(t - 3)\,dt\): 3 is inside \([0, 5]\), so the value is \(3^2 = 9\). 3. \(\int_0^2 t^2\,\delta(t - 3)\,dt\): the impulse at 3 lies outside \([0, 2]\), where the delta is zero, so the integral is 0. Check: in each case only the value of the other factor at the impulse time matters; its behaviour anywhere else is irrelevant.

The transform of the delta

The sifting property makes the transform immediate. For \(a > 0\),

\mathcal{L}\{\delta(t - a)\} = \int_0^\infty e^{-st}\delta(t - a)\,dt = e^{-as}

which agrees with the limit of the pulses. For an impulse at \(t = 0\), the convention is that the whole impulse counts as inside the integral, as if the integral began just before 0, giving \(\mathcal{L}\{\delta(t)\} = 1\). Notice that 1 does not tend to 0 as \(s \to \infty\). Nonconstant coefficient problems showed that every piecewise continuous function of exponential order has a transform that does, so this is one more sign that the delta is not an ordinary function.

Steps and deltas

A step is flat except at its switch, where it rises by 1 instantly. Its slope is therefore zero everywhere except at \(t = c\), where it is infinite in such a way that the total rise is 1. In the generalised sense,

u_c'(t) = \delta(t - c)

Transforms agree, at least formally (the step is not continuous, so this is a consistency check rather than a proof): the derivative rule gives \(s \cdot \frac{e^{-cs}}{s} - u_c(0) = e^{-cs}\) for \(c > 0\), exactly \(\mathcal{L}\{\delta(t - c)\}\). The relationship runs the other way too: integrating a delta from 0 gives a step. A unit impulse is the rate of change of a unit switch.

Worked example: a struck spring

A spring obeys \(y'' + 4y = 2\delta(t - 1)\), with \(y(0) = 1\) and \(y'(0) = 0\). 1. Transform: \(s^2Y - s + 4Y = 2e^{-s}\). 2. Solve: \(Y = \frac{s}{s^2 + 4} + e^{-s}\frac{2}{s^2 + 4}\). 3. Invert: \(\frac{s}{s^2 + 4}\) gives \(\cos 2t\), and \(\frac{2}{s^2 + 4}\) gives \(\sin 2t\), shifted by 1. 4. So

y = \cos 2t + u_1(t)\sin\big(2(t - 1)\big)

Check: the new term is 0 at \(t = 1\), so \(y\) is continuous, but its slope there is \(2\cos 0 = 2\). The blow adds 2 to the velocity, the strength of the impulse.

Why the velocity jumps

The last check is no accident. Integrate \(y'' + 4y = k\delta(t - a)\) over a tiny interval around \(a\). The integral of \(4y\) shrinks to zero because \(y\) is bounded, the integral of the delta side is \(k\), and the integral of \(y''\) is the change in \(y'\). So, for a leading coefficient of 1, an impulse of strength \(k\) makes \(y'\) jump by exactly \(k\) while \(y\) stays continuous. Physically this is the impulse-momentum principle: a blow of impulse \(J\) on a mass \(m\) changes its velocity by \(\frac{J}{m}\) without moving it in that instant.

Worked example: a damped system struck once

Solve \(y'' + 3y' + 2y = \delta(t - 1)\) from rest. 1. Transform: \((s^2 + 3s + 2)Y = e^{-s}\). 2. Factor and split: \(Y = e^{-s}\big(\frac{1}{s + 1} - \frac{1}{s + 2}\big)\). 3. Invert and shift by 1:

y = u_1(t)\big(e^{-(t - 1)} - e^{-2(t - 1)}\big)

Check: nothing happens until \(t = 1\). There the bracket is 0, so \(y\) is continuous, and its derivative is \(-1 + 2 = 1\), so \(y'\) jumps from 0 to 1, the impulse strength. The response rises, peaks, and decays back to 0, as an overdamped system should.

Impulse response

The response of a system at rest to \(\delta(t)\) is its impulse response \(h(t)\). From Solving initial value problems, starting at rest gives \(Y = H(s)G(s)\), and here \(G = 1\), so \(\mathcal{L}\{h\} = H(s)\): the impulse response is the inverse transform of the transfer function. For \(y'' + 3y' + 2y\), it is \(h(t) = e^{-t} - e^{-2t}\), and the last example is just \(h\) delayed by 1. Engineers measure \(h\) by striking a structure or sending a click through a room, because knowing \(h\) determines the response to every input, as Convolution integrals explains.

Common mistakes to avoid

Writing \(\mathcal{L}\{\delta(t - a)\} = \frac{e^{-as}}{s}\), which is the transform of the step, not the impulse. Sifting with an impulse outside the interval of integration, where the integral is 0. Expecting \(y\) itself to jump at an impulse: for second order equations the jump is in \(y'\). Forgetting the strength: \(5\delta(t - 1)\) transforms to \(5e^{-s}\). Treating \(\delta(t - a)\) as having a value at \(a\) that can be substituted; it makes sense only inside an integral or a transform.

Key takeaways

\(\delta(t - a)\) models a unit impulse at time \(a\): zero away from \(a\), unit area, and \(\int f(t)\delta(t - a)\,dt = f(a)\) over any interval containing \(a\). It is the limit of unit-area pulses and a generalised function, not an ordinary one. Its transform is \(e^{-as}\), with \(\mathcal{L}\{\delta(t)\} = 1\), and \(u_c' = \delta(t - c)\). An impulse of strength \(k\) makes \(y'\) jump by \(k\) when the leading coefficient is 1. The impulse response is the inverse transform of the transfer function.

Flashcards

Flashcard

δ(t - a)

A unit impulse at time a

Zero for t ≠ a, with total area 1.

Test your understanding

Question 1 of 5

What is the Laplace transform of δ(t - 4)?

Question 2 of 5

What is the value of the integral of (t² + 1) δ(t - 3) from t = 0 to t = 10?

Question 3 of 5

Solve y″ + 9y = δ(t - 2) with y(0) = 0 and y′(0) = 0.

Question 4 of 5

For y″ + y = 5δ(t - 1), what happens to the solution at t = 1?

Question 5 of 5

What is the impulse response of y″ + 3y′ + 2y, that is, its solution from rest with forcing δ(t)?

← Previous topicNext topic →