Visual walkthrough — Transfer function — Laplace domain, poles and zeros

We will follow one concrete machine the whole way: a mass hanging on a spring with some friction, described by $\ddot{y}+3\dot{y}+2y=u(t).$ Don't panic at the symbols — the very first step defines every one of them on a picture.

---

Step 1 — Draw the machine and name every symbol

WHAT. A block of mass sits on a spring. A hand pushes it with a force. We watch how far it moves.

WHY. Before any algebra, we must know what each letter is physically. A symbol you can't point to on a picture is a symbol you'll misuse later.

PICTURE.

• $t$ — time, in seconds. The horizontal thing that keeps ticking.
• $y(t)$ — the output: how far the block has moved from rest, at time $t$. (The green measuring arrow.)
• $u(t)$ — the input: the force the hand applies (the yellow push arrow). "$u$" is the standard control-engineering name for the command you send in.
• $\dot{y}$ — a shorthand for $\frac{dy}{dt}$, the velocity: how fast $y$ is changing. The single dot means "one time-derivative."
• $\ddot{y}$ — two dots: $\frac{d^{2}y}{d{t}^2}$, the acceleration: how fast the velocity is changing.

The problem: to know $y(t)$ for a new push $u(t)$, you must re-solve this differential equation every single time. That is the pain we are about to kill.

---

Step 2 — The escape hatch: trade calculus for algebra

WHAT. We apply the Laplace transform — an operation $\mathcal{L}$ that takes a time-signal $f(t)$ and returns a new function $F(s)$ of a new variable $s$.

WHY. Because of one magic property (built in the next step): inside this new world, taking a derivative becomes multiplying by $s$. Derivatives are the hard part of a differential equation; if they turn into simple multiplication, the whole equation collapses into ordinary algebra.

PICTURE.

Think of the left room (time domain) as full of springs, dots, and calculus. The doorway $\mathcal{L}$ leads to the right room (the $s$-domain), where those same springs are just polynomials. We solve in the easy room, then walk back out.

We don't even need to compute the integral by hand today — we only need one consequence of it, which is Step 3.

---

Step 3 — The one rule that does all the work: $\frac{d}{dt}\to s$

WHAT. We show that transforming a derivative $f^{\prime }(t)$ gives $sF(s)$ (with a zero start).

WHY. This is the hinge the entire method swings on. If we can trust "derivative → multiply by $s$", every derivative in Step 1 becomes a power of $s$, and the ODE becomes a polynomial.

PICTURE.

Starting from the definition and using integration by parts (the reverse of the product rule — the standard trick when you have a product like $f^{\prime }(t)e^{-st}$ under an integral):

$\mathcal{L}\{f^{\prime }(t)\}=\underset{\text{boundary term}=-f(0)}{\underbrace{[f(t)e^{-st}{]}_0^{\infty }}}+s\underset{=F(s)}{\underbrace{{\int }_0^{\infty }f(t)e^{-st}dt}}=sF(s)-f(0)$

• $[f(t)e^{-st}{]}_0^{\infty }$ — the value at the two ends. At $t=0$ it leaves $-f(0)$; at $t\to \infty$ it vanishes provided the probe wins (see below).
• $f(0)$ — the initial condition: where the signal started.
• $sF(s)$ — the payoff: the derivative became $s$ times the transform.

---

Step 4 — Push the machine through the doorway

WHAT. Transform every term of $\ddot{y}+3\dot{y}+2y=u$ using Step 3.

WHY. Now the calculus vanishes. Each dot becomes a power of $s$.

PICTURE.

Term by term, with zero initial conditions ($Y(s)$ is the transform of $y$, $U(s)$ of $u$):

$\underset{\ddot{y}\to s^{2}Y}{\underbrace{s^{2}Y}}+\underset{3\dot{y}\to 3sY}{\underbrace{3sY}}+\underset{2y\to 2Y}{\underbrace{2Y}}=\underset{u\to U}{\underbrace{U}}$

• $\ddot{y}$ (two dots) → $s^{2}Y$ (two factors of $s$),
• $3\dot{y}$ (one dot) → $3sY$ (one factor of $s$),
• $2y$ (no dots) → $2Y$ (no $s$),
• $u$ → $U$.

Notice: the number of dots became the power of $s$. A second-order machine gives an $s^2$. That is not a coincidence — it is Step 3 applied twice.

---

Step 5 — Isolate the machine: form the ratio $G(s)=Y/U$

WHAT. Factor $Y$ out of the left side and divide to get output-over-input.

WHY. We want a single object that means "for any input $U$, the output is $Y=G\cdot U$." That object is the transfer function — the machine's fingerprint.

PICTURE.

Factor $Y$: $(s^2+3s+2)Y=U$

Divide both sides by $U$ and by $(s^2+3s+2)$:

Now $Y(s)=G(s)U(s)$: multiply the fingerprint by any input and you get the output. No more re-solving the ODE. This is the central result of the parent note — and we built it from a hanging block.

---

Step 6 — Crack the denominator into roots: poles

WHAT. Factor the bottom polynomial and find where it equals zero.

WHY. Because dividing by zero blows $G$ up to infinity, and — as Step 8 shows — those blow-up points are exactly the machine's natural behaviours. They deserve a name: poles.

PICTURE.

Factor: $s^2+3s+2=(s+1)(s+2)$ Set each factor to zero: $s+1=0\Rightarrow s=-1,s+2=0\Rightarrow s=-2$

Because $s=\sigma +j\omega$ is a complex number, we plot each pole as a dot on a 2-D map: horizontal axis $\sigma$ (real part), vertical axis $\omega$ (imaginary part). Both our poles sit on the negative-real axis at $-1$ and $-2$. Remember those spots — the next steps give them meaning.

• $\sigma$ (real part) — how fast a mode grows or decays.
• $\omega$ (imaginary part) — how fast a mode oscillates.

---

Step 7 — Split the fraction into single-pole pieces (partial fractions)

WHAT. Break $G(s)=\frac{1}{(s+1)(s+2)}$ into a sum of one-pole chunks $\frac{A}{s+1}+\frac{B}{s+2}$, and solve for the numbers $A,B$.

WHY. We are about to walk each pole back to the time room. But we only know how to invert a single pole $\frac{1}{s-p}$. So first we must chop the product into a sum of such singletons. That chopping is called partial fractions.

PICTURE.

Guess the shape (one term per pole) and find the unknown weights: $\frac{1}{(s+1)(s+2)}=\frac{A}{s+1}+\frac{B}{s+2}$

• $A,B$ — the weights (how much of each pole is present).
• Cover-up trick: to get $A$, multiply both sides by $(s+1)$ and set $s=-1$ (which kills the $B$ term): $A={\frac{1}{s+2}|}_{s=-1}=\frac{1}{1}=1$.
• For $B$: multiply by $(s+2)$, set $s=-2$: $B={\frac{1}{s+1}|}_{s=-2}=\frac{1}{-1}=-1$.

So $G(s)=\frac{1}{s+1}-\frac{1}{s+2}.$ The single fraction has become two clean singletons — each ready to invert.

---

Step 8 — What a pole means: the impulse response is $\sum A_k{e}^{p_{k}t}$

WHAT. Feed the machine a sharp kick (an impulse) and watch its natural motion — this is the response that reveals what each pole does.

WHY. To make the poles show their personality with nothing else in the way, we pick the cleanest possible input: the unit impulse $u(t)=\delta (t)$, whose transform is simply $U(s)=1$. Then $Y(s)=G(s)\cdot 1=G(s)$, so inverting $G(s)$ itself gives the impulse response $y(t)$. This is why we can walk $G$ directly back to the time domain — and because the impulse only "lights the fuse" at $t=0$ and then lets go, what we see afterward is exactly the system's natural (homogeneous) modes, one per pole.

PICTURE.

Take the two singletons from Step 7 and use the single-pole rule ${\mathcal{L}}^{-1}\{\frac{1}{s-p}\}=e^{pt}$: $y(t)=\underset{\text{pole }s=-1}{\underbrace{(+1)e^{-1\cdot t}}}+\underset{\text{pole }s=-2}{\underbrace{(-1)e^{-2\cdot t}}}=e^{-t}-e^{-2t}$

• $A=+1$ at pole $s=-1$ → $e^{-t}$: dies with time constant $\tau =1$ s (the slow, dominant mode),
• $B=-1$ at pole $s=-2$ → $-e^{-2t}$: dies in $0.5$ s (faster, fades sooner).

The weights $A,B$ came from the zeros/numerator (Step 7); the exponents $p$ came from the poles. That is the whole "poles decide whether, zeros shape how" story in one line.

Both exponents are negative, so both modes shrink → the block wobbles once and settles. Stable.

---

Step 9 — Cover the other cases: oscillation, instability, repeated poles, and zeros

We derived a stable, non-oscillating machine. But the reader must never meet a case we skipped. Here are the ones the map handles.

PICTURE.

Case A — Complex poles (oscillation). The standard second-order form $G(s)=\frac{{\omega }_n^2}{s^2+2\zeta {\omega }_{n}s+{\omega }_n^2},\text{poles }s=-\zeta {\omega }_n\pm j{\omega }_n\sqrt{1-{\zeta }^2}.$

• ${\omega }_n$ — natural frequency, $\zeta$ — damping ratio.
• Real part $-\zeta {\omega }_n<0$ → still decays.
• Imaginary part $\pm {\omega }_n\sqrt{1-{\zeta }^2}\ne 0$ → rings while decaying. Push the poles further left (more damping) → faster settling, less ringing.

Case B — A pole on the right (instability). Take $G(s)=\frac{5}{s^2-s-6}=\frac{5}{(s-3)(s+2)}$. Poles at $s=+3$ and $s=-2$. The dot at $+3$ is on the right → the term $e^{+3t}$ grows forever → unstable, no matter what the other pole does. One bad dot ruins the machine — this feeds directly into the Stability - Routh-Hurwitz criterion and Root locus ideas.

Case C — On the axis (marginal). A pole exactly at $\sigma =0$ (e.g. $s=\pm j\omega$) neither grows nor decays → oscillates forever at constant amplitude. The knife-edge between stable and unstable.

Case D — Repeated (double) poles. If two poles land on exactly the same spot, e.g. $G(s)=\frac{1}{(s+1{)}^2}$, partial fractions can no longer give one clean $\frac{A}{s+1}$ — a doubled pole needs the rule ${\mathcal{L}}^{-1}\{\frac{1}{(s-p{)}^2}\}=t{e}^{pt}$. So the mode is $t{e}^{-t}$: an extra factor of $t$ that first rises before the exponential drags it down. A pole of order $n$ gives $t^{n-1}{e}^{pt}$. Qualitatively new — but the sign of the real part still rules stability (if $p$ is on the axis, $t\cdot e^0=t$ grows, so a repeated axis pole is unstable, unlike the single one in Case C).

Case E — A zero. $G(s)=\frac{s+3}{(s+1)(s+2)}$ has a zero at $s=-3$. It does not move any pole, so it cannot change stability — but it reweights the coefficients $A,B$ (recall Step 7), changing the shape (overshoot, or "wrong-way" first motion if the zero is on the right).

---

The one-picture summary

The whole journey on one canvas: ODE → (doorway $\mathcal{L}$, dots become powers of $s$) → ratio $G(s)=N/D$ → (factor the bottom) → poles as dots on a map → (split into singletons, kick with an impulse, walk back) → each dot is a motion $e^{pt}$. Left of the vertical line = calm; right = catastrophe.

Recall Feynman retelling — the whole walkthrough in plain words

We had a block on a springy, sticky spring, and an annoying equation with dots (speeds) and double-dots (accelerations) that we'd have to solve fresh for every new push. So we walked through a magic doorway where the rule is: every dot turns into a multiply-by-$s$ — but only for well-behaved signals that don't grow faster than the doorway's shrinking probe (that's the "region of convergence" small print). Suddenly the equation had no calculus in it — just $s^2+3s+2$ times the output equals the input. We shuffled that into a single fraction, $\frac{1}{s^2+3s+2}$: the block's ID card. Then we cracked the bottom into $(s+1)(s+2)$ and found the two special numbers $-1$ and $-2$ — the poles. We split the single fraction into two easy pieces (partial fractions), gave the block one sharp kick so the input was just "1", and walked each piece back through the doorway: a dot at $-1$ became a wiggle dying in 1 second, and $-2$ one dying in half a second. Both to the left of the middle line → the block calms down → good. If a dot sneaks to the right, its wiggle grows forever → broken. Two dots stacked on the same spot add an extra "×time" that makes even an on-the-line pole grow. And zeros? Just volume knobs on the wiggles — they change the shape but can never break the machine.

Connections

• Parent topic
• Laplace transform — the doorway used in Steps 2–3
• Second-order systems - damping ratio and natural frequency — Case A
• Stability - Routh-Hurwitz criterion — Case B without factoring
• Root locus — how poles move as you tune a gain
• Block diagrams and feedback loops — where these $G(s)$'s multiply together
• Bode plot, State-space representation — other views of the same machine