One machine, four tools
You have spent this whole rung building a toolkit, one tool per guide, and now they all belong to a single picture. The Heaviside step taught you to write an input that switches on. The Dirac delta gave you the idealised instantaneous kick. Convolution showed you how a system smears any input against its own memory. And the transfer function H(s) packaged the system itself into one rational expression in s. Separately they are clever tricks; together they are a way of reading a real machine. This guide is where they meet a circuit.
Take the workhorse RLC circuit: a resistor, inductor, and capacitor in a loop, driven by a voltage. Kirchhoff's law turns it into a y'' + b y' + c y = E(t), where the coefficients are the component values and E(t) is the applied voltage. That is the same constant-coefficient equation you have solved a dozen ways, but now we treat it as an input-output system: feed in a voltage E(t), read out a current or charge y(t). Everything interesting is hidden in two questions — what does the circuit do when you flip its switch, and what does it do when the load suddenly changes?
Flip the switch: a windowed step
Connect a battery of voltage V to the loop at t = 0 and disconnect it at t = 5. The driving voltage is not a formula valid for all time — it is on, then off. With the step in hand you write it in one line: E(t) = V * [u(t) - u(t - 5)], a force that switches on at 0 and off at 5. This is the exact shape of a switched circuit input, and it is why the second shifting theorem earns its keep. Transform the equation, and each switching time a leaves behind a tidy e^(-a s) factor — the algebraic fingerprint of the delay.
Solving for Y(s) and inverting, the answer comes out already in switched form: a base response that begins charging the circuit at t = 0, plus a delayed correction that switches on exactly at t = 5 to undo the source. You never glue intervals by hand, and the continuity of y is built in. There is one detail worth flagging: the step input is discontinuous, so the current through a capacitor (which follows the voltage) can jump, while the current through an inductor cannot change instantaneously — physics constrains which variables may leap at a switch and which must stay continuous.
Drop the load: an impulse
Now the second question. A capacitor is suddenly connected and dumps its charge into the loop in a flash; a hammer strikes a structure; a heavy box is dropped onto a sprung platform. Each delivers a large force over a tiny time, and what matters is not its shape but its total — the area under the force, the transferred impulse. That is precisely what the delta captures: V0 * delta(t - 2) is a unit of action concentrated at t = 2. The system's reply to a single unit kick is its impulse response h(t), and h(t) is nothing other than the inverse transform of the transfer function H(s).
Here the abstract suddenly becomes intensely physical. The impulse response is the system's signature — its free ringing after a single tap. In an RLC loop you can literally watch it: a sharp kick sets the circuit ringing at its damped natural frequency, the oscillation decaying as the resistor bleeds energy away. Underdamped, you see a fading sine; overdamped, a sluggish slump back to rest. The kick does not impose its own rhythm; it merely excites the rhythm the circuit already owns. That is why h(t) is the most honest fingerprint a system has.
Convolution: any input at all
The deepest reward of this rung is that the impulse response answers everything, not just impulses. Decompose any input E(t) into an unbroken stream of tiny kicks — one delta of height E(tau) at each instant tau. The system answers each kick with a shifted copy of h, and because it is linear, the full response is the sum of all those answers. That sum, taken to the limit, is the convolution y(t) = (h * E)(t) = the integral of h(t - tau) E(tau) over tau from 0 to t. Read it plainly: the output now is the input weighted by how the system remembers each past instant, fading as h fades.
And the convolution theorem makes this practical instead of merely beautiful. That awkward overlap integral in time becomes a plain product in the s-domain: Y(s) = H(s) * E(s). This is the engine of the whole machine. To predict a circuit's response to a complicated drive you never grind out the convolution by hand — you multiply two transforms and invert. The transfer function is the multiplier, the input transform is the input, and the output transform falls out as their product. This is exactly Duhamel's principle, read forward.
RLC loop: a y'' + b y' + c y = E(t)
Transfer function H(s) = 1 / (a s^2 + b s + c)
Impulse response h(t) = L^(-1){ H(s) } (the circuit's signature ring)
Any input Y(s) = H(s) * E(s) (convolution theorem)
y(t) = (h * E)(t) (same thing, in time)
Poles = roots of a s^2 + b s + c = 0 -> all in left half-plane => stableWill it settle? Poles decide
There is one more question a real engineer must answer before trusting any of this: after the switching and the kicks, does the circuit settle down or does it run away? The transfer function answers it at a glance. The poles of H(s) are the roots of its denominator — for the RLC loop, the roots of a s^2 + b s + c = 0, which are exactly the roots of the characteristic equation you met long ago. Each pole p contributes an e^(p t) term to the natural behaviour, so a pole's real part is the growth or decay rate, and its imaginary part is a ringing frequency.
So the rule is sharp: the system is stable when every pole lies strictly in the left half-plane — every real part negative — because then every natural term decays and the response settles into its forced steady state. A pole in the right half-plane means a term that grows without bound; the circuit saturates or burns. A pole exactly on the imaginary axis is the knife-edge — undamped oscillation that neither grows nor decays. For a passive RLC loop with real resistance, the poles always sit left of the axis, which is why such a circuit always rings down; active or feedback circuits can push poles rightward, and that is where instability is engineered or feared.
Reading a machine in one breath
Step back and see what you can now do without solving a single integral by hand. Hand me a switched circuit and I write its input with steps and deltas, package the circuit as H(s), read its stability off the poles, and get any output as H(s) times the input transform. The same four tools describe a struck bell, a suspension hitting a pothole, a thermostat pulsing a heater, or a drug given as repeated doses — all are an input-output system driven by switches and kicks, and all yield to the same pipeline.
Two honest cautions close the rung. First, this entire framework is the gift of linearity: convolution, the product Y = H*E, and the pole test all rest on superposition, and a genuinely nonlinear circuit (a diode, a saturating amplifier) obeys none of them. Second, the step and the delta are idealised limits — fine when the switching or the kick is fast against the system's time constant, but a story you must abandon when the input's own duration matters. Within those honest limits, though, you now hold a remarkably complete way of reading what a switched, kicked, real machine will do — and whether you can trust it.