What does it mean to multiply two transforms?
By now the rhythm is in your bones: transform, do algebra in the s-domain, invert. The algebra step constantly hands you *products*. When you solve a linear ODE you reach Y(s) = H(s) X(s) — the transform of the output is the transfer function times the transform of the input — and to invert it you must answer one sharp question: if F(s) and G(s) are the transforms of f(t) and g(t), what time function does the product F(s) G(s) belong to? The tempting guess is f(t) g(t), the pointwise product. That guess is wrong, and seeing why opens the door to one of the most useful ideas in the whole subject.
A quick sanity check kills the wrong guess. Take f(t) = g(t) = 1, whose integrals over the transform give F(s) = G(s) = 1/s. If the rule were pointwise multiplication, the inverse of F(s) G(s) = 1/s^2 would be 1 * 1 = 1. But from the table, the inverse of 1/s^2 is t, not 1. So the product of transforms answers to *something else entirely* — and that something is built by a sliding, overlapping integral, not by multiplying values at matching instants.
Convolution: blending one function's past into another
The correct partner of multiplication in the s-domain is an operation in time called convolution, written with a star: (f * g)(t) = integral from 0 to t of f(tau) g(t - tau) d tau. Read the integral as a recipe. Let tau range over all past instants from 0 up to the present t. At each past instant tau, take the value f(tau), and weight it by g(t - tau) — the second function evaluated at the elapsed time since tau. Then accumulate. The convolution theorem makes the clean promise: L{f * g} = F(s) G(s). A product in the s-world is a convolution in the t-world, full stop.
Picture the mechanism. To form g(t - tau) you take g, flip it left-to-right (the minus sign), and slide it so its origin sits at the present moment t; then you let it overlap f and you integrate the product of the two graphs across their shared region. As t advances, the flipped g slides further along, sweeping over more of f's history — which is exactly why convolution feels like an echo or a smearing: the present output is a weighted sum of the whole past. This is the same flip-slide-accumulate picture you will meet again with the Fourier convolution theorem; the operation is one idea, shared across transforms.
Now the payoff for inversion. Suppose you finish an ODE and hold Y(s) = 1/(s(s^2 + 1)). You may not find this in any table, but you recognize each factor: 1/s is the transform of 1, and 1/(s^2 + 1) is the transform of sin(t). The theorem says the inverse of the product is the convolution of the inverses, so y(t) = 1 * sin(t) = integral from 0 to t of sin(tau) d tau = 1 - cos(t). You inverted a product you could not look up, by convolving two pieces you could. That is the everyday use of the theorem: split, recognize, convolve.
The Heaviside step: a mathematical switch
Real inputs rarely run from the dawn of time; they get switched on. A voltage is applied, a valve opens, a force begins at t = 2 seconds. The mathematical switch is the Heaviside step function u(t - a): it is 0 before the instant a and 1 after it, a clean jump from off to on. In the transform world it is the building block from which every piecewise, on-off input is assembled. Its transform is delightfully simple — L{u(t - a)} = e^{-a s}/s, and the step at the origin u(t) transforms to 1/s.
Steps combine into anything piecewise. A rectangular pulse that is 1 only between t = a and t = b is u(t - a) - u(t - b): the first step switches it on, the second switches it back off. A staircase is a sum of steps; a signal that turns on and off many times is a sum of such pulses. To transform a *delayed and shifted* signal you use the second shifting theorem: L{f(t - a) u(t - a)} = e^{-a s} F(s). In words, delaying a function by a in time multiplies its transform by e^{-a s} — that exponential factor is the unmistakable signature of a pure time delay, and seeing it lets you read a delay backward when you invert.
This is why the Laplace method crushes the classical approach to piecewise forcing. The old way solved the ODE separately on each interval and then painfully matched the solution and its derivative at every breakpoint to keep things continuous. The transform way encodes the entire forcing function as one expression in steps, transforms once, solves the algebra once, and inverts once — the switching information rides through the transform intact, and the physical continuity falls out automatically. One honest caution lives in the argument: the theorem needs the genuinely shifted f(t - a), not f(t) merely multiplied by a step. If you have g(t) u(t - a) with g not yet shifted, first rewrite g in terms of (t - a) before applying the rule.
The Dirac delta: an idealized kick
Some inputs are not switches but kicks: a hammer blow, a spark of charge dumped in, a collision — concentrated entirely at one instant. The model is the Dirac delta delta(t - a): an input of zero duration but unit total strength, infinitely tall and infinitely thin, yet enclosing area exactly one. Its defining behavior is the *sifting property*: integral of delta(t - a) g(t) dt = g(a). The delta reaches into any function and plucks out its value at the point a. From this its Laplace transform follows in one line — L{delta(t - a)} = e^{-a s}, and the delta at the origin transforms to the bare constant 1.
Feed an impulse to a system and something striking happens. Solve y'' + y = delta(t) with y(0) = 0 and y'(0) = 0. Transform: because L{delta(t)} = 1, the equation becomes (s^2 + 1) Y = 1, so Y = 1/(s^2 + 1) and y(t) = sin(t). Notice that the same system with no forcing and the same zero initial conditions would just sit still at y = 0 forever. The instantaneous kick at t = 0 changed everything: it deposits its strength between 0- and 0+, effectively jumping the velocity y' from 0 to 1, and the system rings thereafter. An impulse is a sudden injection that the algebra handles without you ever splitting the timeline.
Impulse response: where everything ties together
Hit a bell once, sharply, and listen. The ringing tone that swells and fades is the bell's signature — and remarkably, it tells you everything about how the bell will respond to any sound at all. That signature has a precise name: the impulse response h(t), the output of a linear system at rest when its input is a single unit impulse, a Dirac delta. Because L{delta(t)} = 1, the s-domain output is simply H(s) times 1, so h(t) and the transfer function H(s) are a transform pair: the impulse response is exactly the inverse Laplace transform of the transfer function.
Now watch the threads of this guide braid together. For an arbitrary input x(t), the output is Y(s) = H(s) X(s) in the s-domain — a product. By the convolution theorem, that product inverts to a convolution in time: y(t) = (h * x)(t) = integral from 0 to t of h(tau) x(t - tau) d tau. Read it physically. Treat the input as a dense train of scaled, delayed impulses; the system answers each one with a scaled, delayed copy of its impulse response; you add up all the echoes. The convolution integral *is* superposition for a continuous input. This is why the bell's single ring tells the whole story: knowing the echo from one kick, you know the echo from any input, by convolving.
Two costumes for one fact (linear, time-invariant system at rest):
s-domain: Y(s) = H(s) * X(s) ( ordinary product )
|
invert | convolution theorem
v
t-domain: y(t) = (h * x)(t)
= integral_0^t h(tau) x(t - tau) d tau
where h(t) = inverse-Laplace{ H(s) } = impulse response
( forcing = delta(t), zero initial conditions )
Worked impulse: y'' + y = delta(t), y(0)=y'(0)=0
L: (s^2 + 1) Y = L{delta} = 1
Y = 1/(s^2 + 1)
h(t) = sin(t) <- the impulse response itself