Laplace transform

The formula

The unilateral transform, the one engineering runs on:

F(s) = ∫₀^∞ f(t)·e^(−s·t) dt, with s = σ + jω complex — weigh the signal against a decaying complex exponential and integrate; each value of s asks "how much of this decay-and-spin is in the signal?" ^[12]

The integral converges when the signal grows no faster than an exponential: if |f(t)| ≤ M·e^(α·t), then F(s) exists and is analytic for Re(s) > α ^[12]. That half-plane is the region of convergence, and it matters more than most people remember.

The rule that pays the rent:

L{f′(t)} = s·F(s) − f(0⁻) — differentiation in time becomes multiplication by s, and the initial condition walks in as an algebraic term instead of a separate boundary-value chore ^[1]^[12]. Apply it twice and a second-order ODE becomes a quadratic in s.

Convolution collapses the same way:

y(t) = h(t) * x(t) ⇔ Y(s) = H(s)·X(s) — a system's output is its transfer function times the input transform. Cascade two blocks, multiply two functions ^[12].

The bookends:

f(0⁺) = lim(s→∞) s·F(s) and f(∞) = lim(s→0) s·F(s) — the initial and final value theorems, the second valid only when every pole of s·F(s) sits strictly in the left half-plane ^[1].

The pairs worth knowing cold (each checked numerically):

u(t)              ⇔  1/s
e^(−a·t)·u(t)     ⇔  1/(s + a)
t·u(t)            ⇔  1/s²
sin(ω·t)·u(t)     ⇔  ω/(s² + ω²)
cos(ω·t)·u(t)     ⇔  s/(s² + ω²)
δ(t)              ⇔  1

Where you meet it

Any transfer function you have ever written down. H(s) = 1/(1 + s·R·C) for the RC low-pass on your signal-conditioning board is a Laplace-domain object: one pole at s = −1/(RC), step response 1 − e^(−t/RC), 63.2% of final value at one time constant (checked: 1 − e⁻¹ = 0.632). SPICE's .AC sweep is this function evaluated along s = jω.
Control-loop design on a test stand. The plant, the actuator, the sensor, and the PID block are each an H(s); close the loop and the denominator 1 + G(s)·H(s) tells you everything. Tuning a hydraulic load frame or a thrust-stand servo is pole placement whether or not anyone in the room says the words.
The stability chart at a design review. Gain margin and phase margin come off G(jω) — the Laplace transfer function evaluated on the imaginary axis. A pole in the right half-plane is a finding, not a discussion. When the board asks "where are your poles," this is the mathematics they are asking about.
Transient prediction before hardware exists. Relay coil turn-off, inrush into a filter cap, the snubber across a switching FET — write the network equations in s, invert, and you know the peak and the ring frequency before the first board is fabbed.

How it works

The mechanism is integration by parts. Multiply f′(t) by e^(−st) and integrate, and the derivative slides off the signal onto the exponential — which just hands back a factor of s. Do that to every term of an LTI differential equation and the whole thing becomes polynomial algebra. Solve for the output transform, split it into partial fractions, and read the time response off the pair table. Nobody on a bench evaluates the formal inverse (a contour integral up a vertical line in the complex plane); partial fractions and a table cover everything a rational transfer function can produce.

The payoff is that poles are the system. Every pole is a natural mode:

Real pole at s = −1/τ → decay e^(−t/τ). Further left is faster.
Complex pair at s = −ζ·ωₙ ± j·ωₙ·√(1−ζ²) → a damped ring. The angle off the negative real axis sets overshoot: ζ = 0.5 gives 16.3% overshoot on a step, ζ = 0.707 gives 4.3% (both verified by simulation).
Any pole with Re(s) > 0 → growth without bound. That is the entire definition of instability, in one sentence.

Zeros do not add modes; they reweight them — a right-half-plane zero is what makes a system step the wrong way before it goes the right way.

The Fourier transform is the special case: set s = jω and you have the frequency response, legitimate whenever the jω axis lies inside the region of convergence. Laplace earns its keep on everything Fourier chokes on — growing signals, unstable plants, and problems where initial conditions are the point. The σ in s = σ + jω is exactly the convergence knob Fourier doesn't have.

The mistakes people actually make:

Applying the final value theorem without checking poles. F(s) = ω/(s² + ω²) gives lim(s→0) s·F(s) = 0, and yet sin(ω·t) never settles to anything. Poles on the jω axis void the theorem silently — it returns a confident, wrong number. Check pole locations first, every time.
Cancelling an unstable pole with a zero on paper. The algebra allows it; the hardware does not. The unstable mode is still in the plant, now unobservable, still integrating toward the rail. Pole-zero cancellation in the right half-plane is how block-diagram algebra burns people.
Forgetting the transform is an LTI tool. The moment an element saturates, switches, or drifts with temperature, Y(s) = H(s)·X(s) stops being true. It still works as a small-signal linearization about an operating point — but that's a modeling decision someone has to actually make, not a default.
Sloppy 0⁻ versus 0⁺. The unilateral transform takes initial conditions at 0⁻, before the input hits. Feed a system an impulse and the state jumps across t = 0; mix the two conventions and your predicted transient is off by exactly the jump.

History

The transform carries Pierre-Simon Laplace's name because the integral shows up in his probability work — he was manipulating expressions of this form in his memoirs of the early 1780s and built related machinery into his 1812 Théorie analytique des probabilités ^[1]^[2]^[3]. But Laplace was doing probability, not circuits, and the tool sat in the mathematics literature for a century.

The person who made it an engineering method never used the integral at all. Oliver Heaviside — a self-taught former telegraph operator working alone in a room, no degree, no post — spent 1880 to 1887 building what he called operational calculus: replace d/dt with a symbol p, treat it as an algebraic quantity, solve the telegraph and circuit equations that were resisting everyone else, and convert back by rules he compiled himself ^[4]^[5]^[8]. It worked. He could not prove why, and famously did not care; when mathematicians attacked the lack of rigor, his answer was, "Shall I refuse my dinner because I do not fully understand the process of digestion?" ^[4]^[6]. The establishment was not amused — one of his Royal Society papers was rejected outright, the referee citing errors of substance and inadequate proof ^[5].

Vindication came in stages. Thomas Bromwich showed that Heaviside's operator answers were what you get from inverting a Laplace-type integral by contour integration — the vertical-line inverse still carries Bromwich's name ^[7]^[8]. John R. Carson at AT&T and Bell Labs spent the 1920s putting the method on an integral-equation footing for telephone-network transient analysis, and his 1926 Electric Circuit Theory and the Operational Calculus carried it into practice ^[8]^[9]. Gustav Doetsch supplied the full rigor and, in 1937, the first systematic text applying the Laplace transform to engineering problems ^[10]^[11]. During and just after World War II, the transform proper displaced Heaviside's p-operators in the textbooks ^[1]^[10] — same answers, defensible foundations. The controls curriculum every Huntsville GNC engineer sat through is Heaviside's method wearing Laplace's name and Doetsch's proofs.

Related tools

/tools/rc-filter — H(s) = 1/(1 + sRC) is the one-pole example above, live
/tools/rc-charge-time — the 1 − e^(−t/τ) step response and the 63.2% point
/tools/series-rlc-impedance — Z(s) = R + sL + 1/(sC) evaluated at s = jω
/tools/lc-resonance — where the complex pole pair of an underdamped circuit lands
/tools/q-factor-bandwidth — Q = 1/(2ζ): pole damping read as resonance width
/tools/rc-snubber — pole placement to kill a switching ring, in component values
/tools/vibration-natural-freq — the mechanical twin of the same second-order denominator
/tools/reactance — the jω-axis slice of sL and 1/(sC)

Sources

Written by HE in our own words from the cited sources — engineering judgment included, your stamp still required. All entries →