Helpful context:

Differential equations are hard. Consider $y'' + 3y' + 2y = e^{-t}$ with $y(0) = 0$ and $y'(0) = 0$. To solve this directly, you find the homogeneous solution, guess a particular solution of the right form, combine them, and then match the initial conditions. Each step requires care.

But there is a completely different approach. Transform the problem into a different domain where differentiation disappears. In this new domain, the differential equation becomes an algebraic equation - just polynomial arithmetic. Solve the algebra. Transform back. This is the Laplace transform.

The idea of transforming a hard problem into an easier domain is not new. Logarithms turn multiplication into addition: $\log(ab) = \log a + \log b$. Fourier series turn periodic functions into lists of amplitudes. The Laplace transform turns differential equations into rational functions. The payoff in each case: you trade one operation for a simpler one, do the work there, and translate back.

This post develops the Laplace transform from scratch - what it is, why differentiation becomes multiplication, how to invert it, and why it is the natural language of systems and control theory.

Section 1: The Transform

Definition. The Laplace transform of $f(t)$, defined for $t \geq 0$, is:

$$\mathcal{L}\{f(t)\}(s) = F(s) = \int_0^\infty f(t) e^{-st} dt$$

The input is a function of time $t$. The output is a function of a new variable $s$, which can be complex: $s = \sigma + i\omega$.

Discomfort check. What does the complex variable $s$ mean?

Write out the factor: $e^{-st} = e^{-(\sigma + i\omega)t} = e^{-\sigma t} e^{-i\omega t}$. The real part $\sigma$ controls exponential damping - the factor $e^{-\sigma t}$ suppresses $f(t)$ before integrating, which is what makes the integral converge. The imaginary part $\omega$ provides oscillation: $e^{-i\omega t} = \cos(\omega t) - i\sin(\omega t)$, which extracts the frequency-$\omega$ component of $f$. When you evaluate $F(s)$ at a particular $s = \sigma + i\omega$, you are asking: how much does $f(t)$ participate at frequency $\omega$, after damping it at rate $\sigma$? The entire complex $s$-plane is a map of all decay rates and oscillation frequencies simultaneously.

Why the integral converges. For the integral to converge, the $e^{-st}$ factor must eventually suppress $f(t)$. If $f(t)$ grows no faster than $e^{at}$ for some $a$ (formally: $|f(t)| \leq M e^{at}$ for large $t$), then for $\text{Re}(s) > a$, the integrand $|f(t) e^{-st}| \leq M e^{-(s-a)t}$ decays exponentially and the integral converges. The set of $s$ for which the integral converges is a right half-plane $\{s : \text{Re}(s) > a\}$, called the region of convergence (ROC).

What the transform does. The Laplace transform maps a function of time to a function of the complex frequency variable $s$. This is the move from the time domain to the $s$-domain (also called the frequency domain or complex frequency domain).

Section 2: Computing Basic Transforms

Let us build a table by direct computation, working each case out carefully.

$\mathcal{L}\{1\}$:

$$F(s) = \int_0^\infty e^{-st} dt = \left[\frac{-e^{-st}}{s}\right]_0^\infty = 0 - \left(\frac{-1}{s}\right) = \frac{1}{s}, \quad \text{Re}(s) > 0$$

The constant function $1$ transforms to the simple pole $1/s$ at $s = 0$.

$\mathcal{L}\{e^{at}\}$:

$$F(s) = \int_0^\infty e^{at} e^{-st} dt = \int_0^\infty e^{-(s-a)t} dt = \frac{1}{s-a}, \quad \text{Re}(s) > a$$

The exponential $e^{at}$ transforms to a pole at $s = a$. This is the fundamental transform: every other transform can be derived from this one.

$\mathcal{L}\{\sin(\omega t)\}$ and $\mathcal{L}\{\cos(\omega t)\}$.

Apply the formula for $e^{at}$ with $a = i\omega$:

$$\mathcal{L}\{e^{i\omega t}\} = \frac{1}{s - i\omega}$$

Rationalize: $\frac{1}{s - i\omega} = \frac{s + i\omega}{s^2 + \omega^2}$. Use $e^{i\omega t} = \cos(\omega t) + i\sin(\omega t)$ and take real and imaginary parts:

$$\mathcal{L}\{\cos(\omega t)\} = \frac{s}{s^2 + \omega^2}, \quad \mathcal{L}\{\sin(\omega t)\} = \frac{\omega}{s^2 + \omega^2}, \quad \text{Re}(s) > 0$$

$\mathcal{L}\{t^n\}$.

Integrate by parts $n$ times (or use the substitution $u = st$ to get $\Gamma(n+1)/s^{n+1} = n!/s^{n+1}$):

$$\mathcal{L}\{t^n\} = \frac{n!}{s^{n+1}}, \quad \text{Re}(s) > 0$$

$\mathcal{L}\{\delta(t)\}$ (Dirac delta).

The sifting property gives:

$$\int_0^\infty \delta(t) e^{-st} dt = e^{-s \cdot 0} = 1$$

The Dirac delta transforms to $1$, valid for all $s$.

Frequency shifting. Multiplying $f(t)$ by $e^{at}$ shifts $F(s)$ to $F(s - a)$:

$$\mathcal{L}\{e^{at} f(t)\} = F(s - a)$$

This follows immediately from substituting $s - a$ for $s$ in the definition. Combined with the sine and cosine formulas:

$$\mathcal{L}\{e^{at}\sin(\omega t)\} = \frac{\omega}{(s-a)^2 + \omega^2}, \quad \mathcal{L}\{e^{at}\cos(\omega t)\} = \frac{s-a}{(s-a)^2 + \omega^2}$$

These are the time-domain functions that appear when you invert transfer functions with complex poles.

The complete table:

$f(t)$ $F(s)$ ROC
$1$ $\dfrac{1}{s}$ $\text{Re}(s) > 0$
$e^{at}$ $\dfrac{1}{s-a}$ $\text{Re}(s) > a$
$t^n$ $\dfrac{n!}{s^{n+1}}$ $\text{Re}(s) > 0$
$\sin(\omega t)$ $\dfrac{\omega}{s^2 + \omega^2}$ $\text{Re}(s) > 0$
$\cos(\omega t)$ $\dfrac{s}{s^2 + \omega^2}$ $\text{Re}(s) > 0$
$e^{at}\sin(\omega t)$ $\dfrac{\omega}{(s-a)^2 + \omega^2}$ $\text{Re}(s) > a$
$e^{at}\cos(\omega t)$ $\dfrac{s-a}{(s-a)^2 + \omega^2}$ $\text{Re}(s) > a$
$te^{at}$ $\dfrac{1}{(s-a)^2}$ $\text{Re}(s) > a$
$t^n e^{at}$ $\dfrac{n!}{(s-a)^{n+1}}$ $\text{Re}(s) > a$
$\delta(t)$ $1$ all $s$

Section 3: The Key Property - Differentiation Becomes Multiplication

This is the theorem that makes Laplace transforms useful.

Theorem. If $\mathcal{L}\{f(t)\} = F(s)$, then:

$$\mathcal{L}\{f'(t)\} = sF(s) - f(0)$$

Proof. Integrate by parts with $u = e^{-st}$ and $dv = f'(t) dt$:

$$\int_0^\infty f'(t) e^{-st} dt = \left[f(t) e^{-st}\right]_0^\infty - \int_0^\infty f(t)(-s) e^{-st} dt$$

$$= \left(0 - f(0) \cdot 1\right) + s \int_0^\infty f(t) e^{-st} dt = sF(s) - f(0)$$

The boundary term at $\infty$ vanishes because $e^{-st}$ decays to zero in the region of convergence. The initial condition $f(0)$ appears explicitly. $\blacksquare$

Discomfort check. Why does differentiation become multiplication by $s$?

Integration by parts is the mechanical derivation, but the deeper reason is that exponentials $e^{st}$ are eigenfunctions of the derivative operator: $\frac{d}{dt}e^{st} = s \cdot e^{st}$. The Laplace transform expresses $f(t)$ as a superposition of exponentials $e^{st}$ weighted by $F(s)$. Differentiating each exponential $e^{st}$ just multiplies it by its eigenvalue $s$. So differentiating $f$ multiplies each $e^{st}$ component by $s$, which multiplies $F(s)$ by $s$. The subtracted term $f(0)$ is a correction for the integration boundary.

Higher derivatives. Apply the rule repeatedly:

$$\mathcal{L}\{f''(t)\} = s(sF(s) - f(0)) - f'(0) = s^2 F(s) - sf(0) - f'(0)$$

$$\mathcal{L}\{f^{(n)}(t)\} = s^n F(s) - s^{n-1}f(0) - s^{n-2}f'(0) - \cdots - f^{(n-1)}(0)$$

Every derivative becomes a factor of $s$, plus terms from the initial conditions. Apply this to an ODE and every derivative vanishes in favor of algebra.

Linearity. Since the integral is linear: $\mathcal{L}\{af + bg\} = a\mathcal{L}\{f\} + b\mathcal{L}\{g\}$. This means the Laplace transform respects superposition, and you can transform each term of an ODE separately.

Section 4: Solving ODEs from Start to Finish

Let us solve $y'' + 3y' + 2y = e^{-t}$ with $y(0) = 0$, $y'(0) = 0$.

Step 1: Apply $\mathcal{L}$ to both sides.

Use linearity and the derivative rule. Let $Y(s) = \mathcal{L}\{y(t)\}$.

Left side (using both initial conditions equal to zero):

$$\mathcal{L}\{y''\} + 3\mathcal{L}\{y'\} + 2\mathcal{L}\{y\} = (s^2 Y - 0 - 0) + 3(sY - 0) + 2Y = (s^2 + 3s + 2)Y$$

Right side:

$$\mathcal{L}\{e^{-t}\} = \frac{1}{s+1}$$

The ODE becomes: $(s^2 + 3s + 2)Y = \frac{1}{s+1}$.

Step 2: Solve algebraically for $Y(s)$.

$$Y(s) = \frac{1}{(s+1)(s^2 + 3s + 2)}$$

Factor: $s^2 + 3s + 2 = (s+1)(s+2)$. So:

$$Y(s) = \frac{1}{(s+1)^2(s+2)}$$

The differential equation is solved. What remains is algebra.

Step 3: Partial fraction decomposition.

The poles are at $s = -1$ (double) and $s = -2$ (simple). Write:

$$Y(s) = \frac{A}{s+1} + \frac{B}{(s+1)^2} + \frac{C}{s+2}$$

Multiply through by $(s+1)^2(s+2)$:

$$1 = A(s+1)(s+2) + B(s+2) + C(s+1)^2$$

Set $s = -1$: $1 = A(0)(1) + B(1) + C(0)$, so $\mathbf{B = 1}$.

Set $s = -2$: $1 = A(-1)(0) + B(0) + C(-1)^2 = C$, so $\mathbf{C = 1}$.

Set $s = 0$ (any third point): $1 = A(1)(2) + B(2) + C(1) = 2A + 2 + 1 = 2A + 3$. So $A = -1$.

Therefore:

$$Y(s) = \frac{-1}{s+1} + \frac{1}{(s+1)^2} + \frac{1}{s+2}$$

Step 4: Invert each term.

From the table: $\mathcal{L}^{-1}\{1/(s-a)\} = e^{at}$ and $\mathcal{L}^{-1}\{1/(s-a)^2\} = te^{at}$.

$$y(t) = -e^{-t} + te^{-t} + e^{-2t} = (t-1)e^{-t} + e^{-2t}$$

Verification. $y(0) = (0-1)(1) + 1 = 0$. Check. $y'(t) = e^{-t} + (t-1)(-e^{-t}) - 2e^{-2t} = te^{-t} - 2e^{-2t}$. So $y'(0) = 0$. Check. Both initial conditions satisfied.

The Laplace transform converted a second-order differential equation into a quadratic - something anyone who has taken algebra can handle.

Section 5: Partial Fractions and the Inverse Transform

Formally, the inverse Laplace transform is a complex contour integral - the Bromwich integral:

$$f(t) = \frac{1}{2\pi i} \int_{\sigma - i\infty}^{\sigma + i\infty} F(s) e^{st} ds, \quad t > 0$$

This integrates along a vertical line in the complex plane, to the right of all singularities of $F(s)$. The residue theorem from complex analysis gives: this equals the sum of residues of $F(s)e^{st}$ at each pole of $F$.

For practical purposes, when $F(s)$ is a rational function (which it always is for ODE problems), the procedure is:

  1. Factor the denominator of $F(s)$.
  2. Decompose into partial fractions.
  3. Read off each term from the table.

Standard partial fraction cases:

Simple real pole at $s = a$: $\frac{A}{s-a}$ inverts to $Ae^{at}$.

Repeated pole $(s-a)^n$: $\frac{A_1}{s-a} + \frac{A_2}{(s-a)^2} + \cdots + \frac{A_n}{(s-a)^n}$ inverts to $(A_1 + A_2 t + \cdots + A_n \frac{t^{n-1}}{(n-1)!})e^{at}$.

Complex conjugate poles from $(s-\alpha)^2 + \beta^2$ in denominator: contribute damped sinusoids $e^{\alpha t}(A\cos\beta t + B\sin\beta t)$.

Finding residues directly. For a simple pole at $s = a$: $\text{Res}{s=a} F(s) e^{st} = \lim{s \to a} (s-a) F(s) e^{at}$. The residue is the coefficient of $e^{at}$ in the time-domain solution.

Example with complex poles. Find the inverse of $F(s) = \frac{s}{s^2 + 2s + 5}$.

The denominator has roots at $s = -1 \pm 2i$ (from $s^2 + 2s + 5 = (s+1)^2 + 4$). Write as:

$$F(s) = \frac{s}{(s+1)^2 + 4} = \frac{(s+1) - 1}{(s+1)^2 + 4} = \frac{s+1}{(s+1)^2 + 4} - \frac{1}{2} \cdot \frac{2}{(s+1)^2 + 4}$$

From the table with $a = -1$ and $\omega = 2$:

$$f(t) = e^{-t}\cos(2t) - \frac{1}{2}e^{-t}\sin(2t)$$

The complex poles at $s = -1 \pm 2i$ produce damped oscillations with decay rate $1$ and oscillation frequency $2$.

Section 6: The Convolution Theorem

Definition. The convolution of two causal functions $f$ and $g$ is:

$$(f * g)(t) = \int_0^t f(\tau) g(t - \tau) d\tau$$

This computes a weighted average of $f$ over the past, weighted by $g$.

Theorem (Convolution Theorem).

$$\mathcal{L}\{f * g\} = F(s) \cdot G(s)$$

Convolution in time equals multiplication in the $s$-domain.

Proof sketch. Substitute the definition of convolution into the Laplace integral:

$$\int_0^\infty \left(\int_0^t f(\tau) g(t-\tau) d\tau\right) e^{-st} dt$$

Swap the order of integration (valid by Fubini’s theorem under absolute convergence), substitute $u = t - \tau$ to get $g(u) e^{-su}$, collect the remaining exponential to get $e^{-s\tau}$, and the two integrals separate into $F(s) \cdot G(s)$. $\blacksquare$

Why this is central to systems theory. A linear time-invariant (LTI) system is any system that: (1) is linear (superposition holds) and (2) does not change with time (time-shift the input, time-shift the output by the same amount). Ohm’s law, Newton’s law, most physical laws describing equilibria - these give LTI systems.

Every LTI system is completely characterized by its impulse response $h(t)$: the output when the input is $\delta(t)$. For any other input $x(t)$, the output is:

$$y(t) = (h * x)(t) = \int_0^t h(\tau) x(t-\tau) d\tau$$

Taking the Laplace transform:

$$Y(s) = H(s) \cdot X(s)$$

where $H(s) = \mathcal{L}\{h(t)\}$ is the transfer function. In the $s$-domain, the output is simply the product of the transfer function and the input transform. The entire system behavior - how it responds to any input, at any frequency - is encoded in this single rational function $H(s)$.

This is why engineers work in the $s$-domain. Designing a filter, an amplifier, a controller - all reduce to designing $H(s)$ as a rational function, then implementing it as a physical system. The convolution theorem makes this possible.

Section 7: Transfer Functions and System Stability

For a system governed by the ODE $ay'' + by' + cy = x(t)$ with zero initial conditions, taking the Laplace transform gives:

$$(as^2 + bs + c)Y(s) = X(s)$$

$$H(s) = \frac{Y(s)}{X(s)} = \frac{1}{as^2 + bs + c}$$

The transfer function is a ratio of polynomials in $s$.

Poles. The poles of $H(s)$ are the values of $s$ where $H(s)$ blows up - the roots of the denominator polynomial. These are exactly the roots of the characteristic equation $as^2 + bs + c = 0$.

Zeros. The zeros of $H(s)$ are the values where $H(s) = 0$ - the roots of the numerator polynomial.

Poles determine the system’s behavior. Each pole $s = \alpha + i\beta$ of $H(s)$ corresponds to a natural mode $e^{(\alpha + i\beta)t}$ of the system’s impulse response. This mode:

  • Decays to zero if $\alpha < 0$
  • Grows exponentially if $\alpha > 0$
  • Oscillates without decay if $\alpha = 0$

The stability criterion. A system is BIBO stable (bounded input, bounded output stable) if and only if all poles of $H(s)$ lie in the open left half-plane - strictly to the left of the imaginary axis.

Why: a bounded input $|x(t)| \leq M$ produces a bounded output if and only if the impulse response is absolutely integrable: $\int_0^\infty |h(t)| dt < \infty$. This holds if and only if every mode $e^{\alpha t}$ decays, which requires $\alpha < 0$ for all poles. If any pole has $\alpha > 0$, the corresponding mode grows exponentially and the output is unbounded even for bounded inputs.

Reading stability from the $s$-plane. The complex $s$-plane is divided by the imaginary axis into left (stable) and right (unstable) half-planes:

  • Poles in the left half-plane: stable, decaying modes.
  • Poles on the imaginary axis: purely oscillatory modes (marginally stable - not BIBO stable).
  • Poles in the right half-plane: growing modes, unstable.

Example: The spring-mass-damper. Transfer function $H(s) = \frac{1}{ms^2 + cs + k}$ with $m, c, k > 0$. Poles:

$$s = \frac{-c \pm \sqrt{c^2 - 4mk}}{2m}$$

The real part is always $-c/(2m) < 0$. All poles are in the left half-plane, regardless of whether they are real (overdamped), repeated (critically damped), or complex (underdamped). A spring-mass-damper with positive mass, damping, and spring constant is always stable. This is physically obvious, now mathematically confirmed.

Example: An unstable system. $H(s) = \frac{1}{s-2}$. The pole at $s = 2$ is in the right half-plane. The impulse response is $h(t) = e^{2t}$, which grows without bound. Any bounded input will eventually produce an unbounded output.

Section 7b: A Harder ODE Example

Let us solve $y'' + 4y = \sin(2t)$, $y(0) = 1$, $y'(0) = 0$.

This is resonance: the forcing frequency ($\omega = 2$) matches the natural frequency of $y'' + 4y = 0$ (characteristic roots $\pm 2i$, natural frequency $2$).

Apply $\mathcal{L}$:

$$(s^2 Y - s \cdot 1 - 0) + 4Y = \frac{2}{s^2 + 4}$$

$$(s^2 + 4)Y = s + \frac{2}{s^2 + 4}$$

$$Y(s) = \frac{s}{s^2 + 4} + \frac{2}{(s^2 + 4)^2}$$

Invert the first term. $\mathcal{L}^{-1}\{s/(s^2 + 4)\} = \cos(2t)$.

Invert the second term. We need $\mathcal{L}^{-1}\{2/(s^2+4)^2\}$. Use the standard identity: $\mathcal{L}^{-1}\{2\omega^3 / (s^2 + \omega^2)^2\} = \sin(\omega t) - \omega t\cos(\omega t)$. Here $\omega = 2$, so $2\omega^3 = 16$. Therefore:

$$\mathcal{L}^{-1}\left\{\frac{2}{(s^2 + 4)^2}\right\} = \mathcal{L}^{-1}\left\{\frac{16}{16} \cdot \frac{2}{(s^2+4)^2}\right\} = \frac{1}{8}(\sin(2t) - 2t\cos(2t))$$

Full solution:

$$y(t) = \cos(2t) + \frac{1}{8}\sin(2t) - \frac{t}{4}\cos(2t)$$

The term $-\frac{t}{4}\cos(2t)$ grows linearly in amplitude - the resonance. The Laplace transform handles this automatically via the $1/(s^2 + 4)^2$ term; no special modification of the method is needed (contrast with undetermined coefficients, which requires modifying the guess).

Section 8: Properties of the Transform

Time shift. Delaying $f$ by $a \geq 0$ seconds:

$$\mathcal{L}\{f(t-a) u(t-a)\} = e^{-as} F(s)$$

where $u(t)$ is the Heaviside step function. A time delay of $a$ seconds multiplies the $s$-domain function by $e^{-as}$.

Frequency shift (already covered). $\mathcal{L}\{e^{at}f(t)\} = F(s-a)$.

Scaling. $\mathcal{L}\{f(at)\} = \frac{1}{a} F(s/a)$ for $a > 0$. Compressing time stretches the $s$-domain, and vice versa.

Initial value theorem. Extract $f(0^+)$ directly from $F(s)$ without inverting:

$$\lim_{t \to 0^+} f(t) = \lim_{s \to \infty} s F(s)$$

Check with $F(s) = 1/(s-a)$: $\lim s \cdot \frac{1}{s-a} = \lim \frac{s}{s-a} = 1$. And indeed $e^{at}|_{t=0} = 1$. Correct.

Final value theorem. If all poles of $sF(s)$ are in the left half-plane:

$$\lim_{t \to \infty} f(t) = \lim_{s \to 0} s F(s)$$

Check with $F(s) = \frac{1}{s(s+2)}$: $\lim_{s \to 0} s \cdot \frac{1}{s(s+2)} = \frac{1}{2}$. The time-domain function approaches $1/2$ as $t \to \infty$. (The time-domain function is $\frac{1}{2}(1 - e^{-2t})$, which indeed approaches $1/2$.)

The final value theorem is the most useful in control design: given a system’s transfer function and a reference input, compute the steady-state output directly from $\lim_{s \to 0} s \cdot H(s) \cdot X(s)$.

Section 9: Region of Convergence

The integral $\int_0^\infty f(t) e^{-st} dt$ does not converge for all $s$. The region of convergence (ROC) is the set of $s$ for which the Laplace transform is defined.

For causal functions of exponential order: if $|f(t)| \leq M e^{at}$, the ROC is the right half-plane $\text{Re}(s) > a$.

Why the ROC matters. Given a formula $F(s)$, multiple time-domain functions can have the same formula with different ROCs - corresponding to causal vs. anti-causal signals or bilateral (two-sided) transforms. For engineering applications with causal signals (starting at $t = 0$), the ROC is always a right half-plane and the inverse is unique within that class.

ROC and stability. For a causal LTI system, all poles of $H(s)$ must lie inside the ROC. If the ROC includes the imaginary axis $\text{Re}(s) = 0$, the Fourier transform exists and the system is stable. So:

ROC of $H(s)$ includes the imaginary axis $\iff$ all poles in left half-plane $\iff$ system is BIBO stable.

These are all equivalent conditions, linked through the ROC.

Abscissa of convergence. The boundary of the ROC is the vertical line $\text{Re}(s) = \sigma_c$, where $\sigma_c$ is the abscissa of convergence. This equals the real part of the rightmost pole of $F(s)$.

Section 9b: Bode Plots and Frequency Response

When a stable LTI system with transfer function $H(s)$ receives a sinusoidal input $x(t) = \cos(\omega t)$, the steady-state output is:

$$y_{ss}(t) = |H(i\omega)| \cos(\omega t + \angle H(i\omega))$$

The system scales the amplitude by $|H(i\omega)|$ and shifts the phase by $\angle H(i\omega)$. This is the frequency response: the system’s behavior at each frequency $\omega$ is completely determined by evaluating $H$ on the imaginary axis.

Bode plots display $|H(i\omega)|$ (in decibels: $20 \log_{10} |H(i\omega)|$) and $\angle H(i\omega)$ (in degrees) versus $\omega$ on a logarithmic frequency axis. The logarithmic scale is crucial: poles and zeros create characteristic breakpoints where the gain changes slope by $\pm 20$ dB per decade (for simple poles/zeros) or $\pm 40$ dB per decade (for double poles/zeros).

Example: First-order lowpass filter. $H(s) = \frac{1}{1 + s/\omega_0}$, a pole at $s = -\omega_0$.

  • For $\omega \ll \omega_0$: $|H(i\omega)| \approx 1$ (0 dB). The filter passes low frequencies unchanged.
  • For $\omega = \omega_0$ (the cutoff frequency): $|H(i\omega_0)| = 1/\sqrt{2} \approx -3$ dB.
  • For $\omega \gg \omega_0$: $|H(i\omega)| \approx \omega_0/\omega \to 0$. The filter attenuates high frequencies, decaying at $-20$ dB per decade.

The transition at $\omega_0$ is sharp in the Bode plot. Audio equalizers, anti-aliasing filters, and smoothing circuits are all designed by placing poles to achieve desired Bode plot shapes.

Poles near the imaginary axis create resonance peaks. If $H(s) = \frac{1}{(s+\alpha)^2 + \beta^2}$ with small $\alpha$, the pole is close to $i\beta$ on the imaginary axis. The frequency response $|H(i\omega)|$ has a tall, narrow peak near $\omega = \beta$. The height of the peak is approximately $1/(2\alpha\beta)$, which diverges as $\alpha \to 0$. This is the frequency-domain manifestation of resonance.

Section 10: Connection to the Fourier Transform

The Fourier transform of $f(t)$:

$$\hat{f}(\omega) = \int_{-\infty}^\infty f(t) e^{-i\omega t} dt$$

For a causal function ($f(t) = 0$ for $t < 0$), this is $\int_0^\infty f(t) e^{-i\omega t} dt$.

Compare to the Laplace transform at $s = i\omega$:

$$F(i\omega) = \int_0^\infty f(t) e^{-i\omega t} dt = \hat{f}(\omega)$$

The Fourier transform is the Laplace transform on the imaginary axis. Setting $\sigma = 0$ removes the damping factor $e^{-\sigma t}$, leaving only the pure oscillation $e^{-i\omega t}$.

Why Laplace is more general. The Fourier transform requires $\int |f(t)| dt < \infty$ for convergence. Functions like $e^t$, $t$, $\sin(t)$ (over $[0, \infty)$), the unit step - all fail this test. The Laplace transform handles all of them: the damping factor $e^{-\sigma t}$ with $\sigma$ large enough forces convergence. The Laplace transform extends Fourier to exponentially growing functions, constant functions, and step functions - exactly the signals that appear in control systems and circuit analysis.

Poles, frequencies, and decay. A pole at $s = \alpha + i\beta$ with $\alpha < 0$:

  • Lies in the left half-plane (stable).
  • Contributes a damped oscillation $e^{\alpha t}\cos(\beta t)$ to the time-domain response.
  • The Fourier transform of this signal exists and peaks near $\omega = \beta$.
  • The closer the pole is to the imaginary axis (smaller $|\alpha|$), the narrower and taller the peak in $|H(i\omega)|$.

Bode plots. For a stable system, $H(i\omega)$ is the frequency response: how the system responds to a sinusoidal input $x(t) = \cos(\omega t)$. The steady-state output is $|H(i\omega)| \cos(\omega t + \angle H(i\omega))$. A Bode plot shows $|H(i\omega)|$ and $\angle H(i\omega)$ versus $\omega$ on a log scale. Poles create resonance peaks (gain increases near the pole’s imaginary part); zeros create notches (gain decreases near the zero’s imaginary part).

Section 11: Applications Beyond ODEs

Control theory. Every classical control design method works in the $s$-domain. A PID controller has transfer function:

$$C(s) = K_p + \frac{K_i}{s} + K_d s$$

The integral term $K_i/s$ introduces a pole at $s = 0$ (which forces the steady-state error to zero). The derivative term $K_d s$ introduces a zero (which speeds up the response). Designing the controller means placing poles and zeros in the $s$-plane to achieve the desired closed-loop behavior.

Circuit analysis. In the $s$-domain, impedances generalize: resistor $Z_R = R$, capacitor $Z_C = 1/(sC)$, inductor $Z_L = sL$. Kirchhoff’s laws apply to these $s$-domain impedances exactly as in DC analysis. The result is the circuit’s transfer function directly, without setting up and solving differential equations.

Signal processing and filters. A lowpass filter attenuates high frequencies and passes low ones. In the $s$-domain, this means $|H(i\omega)| \approx 1$ for small $\omega$ and $|H(i\omega)| \approx 0$ for large $\omega$. Butterworth filters place all poles on a circle in the left half-plane; Chebyshev filters place them on an ellipse for a steeper rolloff. All filter design is pole-zero design in the $s$-plane.

Partial differential equations. The Laplace transform in time converts a PDE in $(x, t)$ into an ODE in $x$ alone (with $s$ as a parameter). This reduces the PDE to a solvable ODE, then inverts. This is the basis for solving the heat equation and wave equation on semi-infinite domains.

Z-transform. The discrete-time analog of the Laplace transform. For a sequence $\{x[n]\}$:

$$X(z) = \sum_{n=0}^\infty x[n] z^{-n}$$

Under the substitution $z = e^{sT}$ (with $T$ the sampling period), poles in the left half-plane of the $s$-domain map to poles inside the unit circle $|z| < 1$ in the $z$-domain. Stability condition for discrete systems: all poles of $H(z)$ must lie strictly inside the unit circle. Digital filter design, difference equations, and discrete-time control all use the Z-transform as the Laplace transform of discrete-time signals.

Section 12: The Rigorous Underpinning

Existence. If $f$ is piecewise continuous on $[0, \infty)$ and of exponential order ($|f(t)| \leq Me^{at}$), then $F(s) = \mathcal{L}\{f\}$ exists for $\text{Re}(s) > a$ and is analytic (complex-differentiable) there. Analyticity follows from differentiating under the integral sign.

Uniqueness. If $F(s) = G(s)$ on some right half-plane, and $f, g$ are piecewise continuous and of exponential order, then $f(t) = g(t)$ at all points of continuity. The Laplace transform is injective on this class.

The inverse as residues. The Bromwich integral $\frac{1}{2\pi i} \int_{\sigma - i\infty}^{\sigma + i\infty} F(s) e^{st} ds$ is computed by closing the contour to the left (for $t > 0$, the factor $e^{st}$ decays in the left half-plane) and applying the residue theorem:

$$f(t) = \sum_{\text{poles of } F} \text{Res}_{s = s_k} [F(s) e^{st}]$$

For a simple pole at $s = s_k$: $\text{Res} = \lim_{s \to s_k} (s - s_k) F(s) e^{s_k t}$. For a double pole at $s = s_k$: $\text{Res} = \lim_{s \to s_k} \frac{d}{ds}[(s-s_k)^2 F(s) e^{st}]$. These residues are exactly what partial fractions compute, organized differently.

Laplace vs. Fourier in $L^2$. For functions in $L^2(\mathbb{R}^+)$ (square-integrable), the Laplace transform on the line $\text{Re}(s) = \sigma$ is related to the Fourier transform by the Paley-Wiener theorem: $F(\sigma + i\omega)$ as a function of $\omega$ is in $L^2(\mathbb{R})$, and the map $f \mapsto F$ is an isometry (up to a constant). This connects Laplace theory to Fourier analysis on $L^2$ rigorously.

Summary

Concept Key fact
Laplace transform $F(s) = \int_0^\infty f(t) e^{-st} dt$; time domain to $s$-domain
$s = \sigma + i\omega$ $\sigma$ = decay rate, $\omega$ = oscillation frequency
ROC $F(s)$ converges for $\text{Re}(s) > \sigma_c$ (right half-plane)
Differentiation rule $\mathcal{L}\{f'\} = sF(s) - f(0)$; derivatives become powers of $s$
Second derivative $\mathcal{L}\{f''\} = s^2 F(s) - sf(0) - f'(0)$
ODE solving strategy Apply $\mathcal{L}$, solve algebraically for $Y(s)$, partial fractions, invert
Convolution theorem $\mathcal{L}\{f * g\} = F(s) G(s)$; multiplication in $s$ = convolution in time
Transfer function $H(s) = Y(s)/X(s)$; characterizes any LTI system
Poles Roots of denominator; location in $s$-plane determines stability
BIBO stability All poles in open left half-plane $\text{Re}(s) < 0$
Fourier connection $\hat{f}(\omega) = F(i\omega)$; Fourier is Laplace on the imaginary axis
Initial value theorem $f(0^+) = \lim_{s \to \infty} sF(s)$
Final value theorem $\lim_{t \to \infty} f(t) = \lim_{s \to 0} sF(s)$ (if stable)

The Laplace transform is the natural language for linear time-invariant systems. It converts differential equations into algebra, convolution into multiplication, and system behavior into pole-zero geometry. Once you understand where the poles are, you understand the system.

Read Next: