Boo-AI — Master Artificial Intelligence by Building from Scratch

Learning Objectives

By the end of this section, you will be able to:

Define the inverse Laplace transform and explain its role in solving differential equations
Apply partial fraction decomposition to break complex rational functions into invertible components
Use tables of standard Laplace transforms to find inverse transforms
Apply linearity, shifting, and scaling properties to compute inverse transforms
Complete the square to handle quadratic denominators with complex roots
Interpret poles in the s-domain as decay rates and oscillation frequencies in time
Connect inverse Laplace transforms to control systems, signal processing, and machine learning

The Big Picture: Returning to the Time Domain

"The art of problem-solving is the art of finding the right representation. Transform methods let us solve in one domain and convert back to another." — Richard Hamming

In the previous sections, we learned how the Laplace transform converts differential equations into algebraic equations—a remarkable simplification. But to answer real questions about physical systems, we need to convert our s-domain solutions back to the time domain. This is the job of the inverse Laplace transform.

1️⃣

Transform

Convert differential equation from time domain $f(t)$ to s-domain $F(s)$

2️⃣

Solve Algebraically

Manipulate $F(s)$ using algebra (no derivatives or integrals needed)

3️⃣

Inverse Transform

Convert back: $f(t) = \\mathcal{L}^{-1}\\{F(s)\\}$

Why Inverse Transforms Matter

The inverse Laplace transform is the final step in the Laplace method for solving differential equations. Without it, we have solutions in an abstract mathematical space that don't directly tell us what happens in time.

In circuit analysis: What voltage appears across a capacitor after switching on?
In mechanical systems: How does a spring-mass system oscillate after being displaced?
In control systems: How does an aircraft's position respond to a pilot's input?

Historical Context

The inverse Laplace transform has a fascinating mathematical history that spans centuries of development.

1737: Euler's Gamma Integral

Leonhard Euler studied integrals of the form $\\int_0^\\infty e^{-st} f(t) \\, dt$ , laying the groundwork for what would become the Laplace transform.

1812: Laplace's Treatise

Pierre-Simon Laplace developed the transform systematically in his work on probability theory and celestial mechanics. However, he didn't work out the inverse transform formula.

1820s: Fourier and Cauchy

Augustin-Louis Cauchy developed the complex contour integral formula for the inverse transform. Joseph Fourier's work on integral transforms influenced this development significantly.

1892: Bromwich Integral

Thomas John I'Anson Bromwich rigorously established the inverse transform formula as a contour integral in the complex plane, now called the Bromwich integral.

1940s: Engineering Revolution

Oliver Heaviside's operational calculus and the needs of World War II radar systems made Laplace methods standard in electrical engineering. Tables of transforms became essential references.

Definition of the Inverse Transform

The inverse Laplace transform recovers a time-domain function $f(t)$ from its s-domain representation $F(s)$ .

The Inverse Laplace Transform

f(t) = \\mathcal{L}^{-1}\\{F(s)\\} = \\frac{1}{2\\pi i} \\int_{\\gamma - i\\infty}^{\\gamma + i\\infty} F(s) e^{st} \\, ds

This is the Bromwich integral (or Mellin's inverse formula), where $\\gamma$ is a real number greater than the real parts of all singularities of $F(s)$ .

Practical Reality

In practice, we almost never evaluate the Bromwich integral directly! Instead, we use:

Tables of known transform pairs
Partial fractions to decompose complex functions
Properties like linearity and shifting
Completing the square for quadratic denominators

Notation

We write the inverse Laplace transform as:

f(t) = \\mathcal{L}^{-1}\\{F(s)\\}

This notation indicates:

$\\mathcal{L}^{-1}$ — the inverse Laplace transform operator
$F(s)$ — the function in the s-domain (typically a rational function)
$f(t)$ — the recovered time-domain function

Uniqueness and Existence

A fundamental question: if two different functions have the same Laplace transform, how do we know which one to recover?

Lerch's Theorem (Uniqueness)

If $f(t)$ and $g(t)$ are continuous for $t \\geq 0$ and have the same Laplace transform, then $f(t) = g(t)$ for all $t \\geq 0$ .

This guarantees that the inverse Laplace transform is essentially unique for continuous functions.

Technical Note

Functions that differ only at isolated points (a "set of measure zero") have the same Laplace transform. For practical purposes, this distinction rarely matters since physical quantities are continuous.

Table of Basic Inverse Transforms

The most practical approach to finding inverse Laplace transforms is to match $F(s)$ with entries in a table of known transform pairs.

Using the Table

The strategy is to manipulate $F(s)$ until it matches one or more entries in the table. This often requires:

Partial fraction decomposition
Completing the square
Algebraic manipulation (factoring, combining terms)
Using properties like shifting and scaling

Properties of Inverse Transforms

The inverse Laplace transform inherits key properties from the forward transform that make computation tractable.

1. Linearity

\\mathcal{L}^{-1}\\{aF(s) + bG(s)\\} = a \\cdot \\mathcal{L}^{-1}\\{F(s)\\} + b \\cdot \\mathcal{L}^{-1}\\{G(s)\\}

Inverse transforms distribute over addition and scalar multiplication.

Example: Find $\\mathcal{L}^{-1}\\left\\{\\frac{3}{s} + \\frac{5}{s^2}\\right\\}$

By linearity: $3 \\cdot \\mathcal{L}^{-1}\\left\\{\\frac{1}{s}\\right\\} + 5 \\cdot \\mathcal{L}^{-1}\\left\\{\\frac{1}{s^2}\\right\\} = 3 \\cdot 1 + 5 \\cdot t = 3 + 5t$

2. First Shifting Theorem (s-shift)

If $\\mathcal{L}^{-1}\\{F(s)\\} = f(t)$ , then:

\\mathcal{L}^{-1}\\{F(s - a)\\} = e^{at} f(t)

Shifting in s-domain corresponds to multiplication by $e^{at}$ in time domain.

Example: Find $\\mathcal{L}^{-1}\\left\\{\\frac{1}{(s-3)^2}\\right\\}$

We know $\\mathcal{L}^{-1}\\left\\{\\frac{1}{s^2}\\right\\} = t$ . Replacing $s$ with $s - 3$ and applying the shift theorem:

\\mathcal{L}^{-1}\\left\\{\\frac{1}{(s-3)^2}\\right\\} = e^{3t} \\cdot t = te^{3t}

3. Scaling in s

If $\\mathcal{L}^{-1}\\{F(s)\\} = f(t)$ , then:

\\mathcal{L}^{-1}\\{F(as)\\} = \\frac{1}{a} f\\left(\\frac{t}{a}\\right), \\quad a > 0

4. Derivative in s (Multiplication by t)

\\mathcal{L}^{-1}\\{F'(s)\\} = -t \\cdot f(t)

More generally: $\\mathcal{L}^{-1}\\{F^{(n)}(s)\\} = (-1)^n t^n f(t)$

Partial Fraction Decomposition

Partial fraction decomposition is the most important technique for computing inverse Laplace transforms. It breaks a complicated rational function into simpler pieces that match table entries.

The Strategy

If $F(s) = \\frac{N(s)}{D(s)}$ is a proper rational function (degree of $N$ less than degree of $D$ ), we can decompose it based on the factors of $D(s)$ .

Case 1: Distinct Real Roots

If $D(s) = (s - r_1)(s - r_2) \\cdots (s - r_n)$ with distinct roots:

\\frac{N(s)}{D(s)} = \\frac{A_1}{s - r_1} + \\frac{A_2}{s - r_2} + \\cdots + \\frac{A_n}{s - r_n}

Each term $\\frac{A_k}{s - r_k}$ has inverse transform $A_k e^{r_k t}$ .

Case 2: Repeated Real Roots

If $(s - r)$ appears with multiplicity $m$ :

\\frac{\\cdots}{(s-r)^m} = \\frac{A_1}{s-r} + \\frac{A_2}{(s-r)^2} + \\cdots + \\frac{A_m}{(s-r)^m}

Each term $\\frac{A_k}{(s-r)^k}$ has inverse transform $A_k \\frac{t^{k-1}}{(k-1)!} e^{rt}$ .

Case 3: Complex Conjugate Roots

For quadratic factors $s^2 + bs + c$ that don't factor over the reals:

\\frac{\\cdots}{s^2 + bs + c} = \\frac{As + B}{s^2 + bs + c}

We then complete the square to match standard forms involving sines and cosines.

Completing the Square

When the denominator contains an irreducible quadratic (complex roots), we complete the square to get it into the form $(s - a)^2 + \\omega^2$ .

Completing the Square Method

To rewrite $s^2 + bs + c$ :

s^2 + bs + c = \\left(s + \\frac{b}{2}\\right)^2 - \\frac{b^2}{4} + c

= \\left(s + \\frac{b}{2}\\right)^2 + \\left(c - \\frac{b^2}{4}\\right)

Let $a = -\\frac{b}{2}$ and $\\omega^2 = c - \\frac{b^2}{4}$ . Then:

s^2 + bs + c = (s - a)^2 + \\omega^2

Example: Damped Oscillator

Find $\\mathcal{L}^{-1}\\left\\{\\frac{1}{s^2 + 2s + 5}\\right\\}$

Step 1: Complete the square in the denominator.

s^2 + 2s + 5 = (s^2 + 2s + 1) + 4 = (s + 1)^2 + 4 = (s + 1)^2 + 2^2

Step 2: Recognize the shifted form.

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

Step 3: Apply the table entry for $\\frac{\\omega}{(s-a)^2 + \\omega^2}$ .

\\mathcal{L}^{-1}\\left\\{\\frac{\\omega}{(s-a)^2 + \\omega^2}\\right\\} = e^{at} \\sin(\\omega t)

Step 4: With $a = -1$ and $\\omega = 2$ :

\\mathcal{L}^{-1}\\left\\{\\frac{1}{s^2 + 2s + 5}\\right\\} = \\frac{1}{2} e^{-t} \\sin(2t)

Physical Interpretation

This result describes a damped oscillation:

$e^{-t}$ — exponential decay (damping factor)
$\\sin(2t)$ — oscillation at angular frequency $\\omega = 2$

The poles at $s = -1 \\pm 2i$ encode both the decay rate (real part = -1) and oscillation frequency (imaginary part = ±2).

Interactive Transform Visualizer

Explore the relationship between s-domain functions and their time-domain representations. See how poles in the s-plane determine the behavior of the time-domain signal.

Worked Examples

Example 1: Distinct Real Poles

Find $\\mathcal{L}^{-1}\\left\\{\\frac{2s + 5}{(s + 1)(s + 3)}\\right\\}$

Step 1: Set up partial fractions.

\\frac{2s + 5}{(s + 1)(s + 3)} = \\frac{A}{s + 1} + \\frac{B}{s + 3}

Step 2: Find A by the cover-up method. Set $s = -1$ :

A = \\frac{2(-1) + 5}{(-1) + 3} = \\frac{3}{2}

Step 3: Find B. Set $s = -3$ :

B = \\frac{2(-3) + 5}{(-3) + 1} = \\frac{-1}{-2} = \\frac{1}{2}

Step 4: Take the inverse transform.

f(t) = \\frac{3}{2} e^{-t} + \\frac{1}{2} e^{-3t}

Example 2: Repeated Root

Find $\\mathcal{L}^{-1}\\left\\{\\frac{s + 3}{(s + 2)^3}\\right\\}$

Step 1: Set up partial fractions for a triple root.

\\frac{s + 3}{(s + 2)^3} = \\frac{A}{s + 2} + \\frac{B}{(s + 2)^2} + \\frac{C}{(s + 2)^3}

Step 2: Multiply through by $(s+2)^3$ and expand, or use the substitution $u = s + 2$ :

s + 3 = (s + 2) + 1 = u + 1

\\frac{u + 1}{u^3} = \\frac{1}{u^2} + \\frac{1}{u^3}

Step 3: Back-substitute and invert.

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

f(t) = te^{-2t} + \\frac{t^2}{2}e^{-2t} = e^{-2t}\\left(t + \\frac{t^2}{2}\\right)

Example 3: Complex Conjugate Poles

Find $\\mathcal{L}^{-1}\\left\\{\\frac{2s + 3}{s^2 + 4s + 13}\\right\\}$

Step 1: Complete the square.

s^2 + 4s + 13 = (s + 2)^2 + 9 = (s + 2)^2 + 3^2

Step 2: Rewrite the numerator to match standard forms. We want forms involving $s + 2$ and constants.

2s + 3 = 2(s + 2) - 1

Step 3: Split into two fractions.

\\frac{2s + 3}{(s + 2)^2 + 9} = \\frac{2(s + 2)}{(s + 2)^2 + 9} - \\frac{1}{(s + 2)^2 + 9}

Step 4: Invert using table entries.

f(t) = 2e^{-2t}\\cos(3t) - \\frac{1}{3}e^{-2t}\\sin(3t)

Machine Learning Connections

The inverse Laplace transform and its underlying concepts appear throughout modern machine learning and signal processing.

1. Transfer Functions in Neural Networks

Continuous-depth neural networks (Neural ODEs) can be analyzed using Laplace methods:

The network's dynamics are described by differential equations
Transfer functions characterize input-output relationships
Poles determine stability and response characteristics

2. Control Theory and Reinforcement Learning

Modern RL systems often use linear quadratic regulators (LQR) where:

System dynamics are modeled as differential equations
Transfer functions describe how actions affect states
Inverse Laplace transforms give time-domain predictions

3. Signal Processing in Audio ML

Audio processing networks use concepts from Laplace/Fourier analysis:

Frequency-domain features: Spectrograms are related to the Fourier transform (imaginary axis of Laplace)
Filter design: IIR filters are designed using s-domain transfer functions, then converted to digital
Stability analysis: Pole locations determine if recursive networks are stable

4. Diffusion Models

The heat equation (diffusion) connection we saw in previous chapters extends here:

Forward diffusion adds noise following a specific schedule
The schedule is often designed using eigenvalue analysis
Denoising can be viewed as an inverse problem—recovering the original signal

5. System Identification

ML techniques for learning dynamical systems often estimate transfer functions:

Given input-output data, learn $H(s)$
Neural networks can approximate the inverse Laplace relationship
Poles/zeros of learned systems reveal physical interpretations

ML Application	Laplace Connection
Neural ODEs	State evolution via differential equations
Control systems (LQR)	Transfer function design
Audio processing	Filter design, stability analysis
Time series forecasting	System dynamics modeling
Diffusion models	Heat equation eigenvalue analysis

Python Implementation

Let's implement inverse Laplace transforms using both symbolic (SymPy) and numerical (SciPy) approaches.

Inverse Laplace Transforms in Python

🐍inverse_laplace.py

Explanation(7)

Code(154)

9Symbolic Inverse Laplace

SymPy can compute inverse Laplace transforms symbolically, giving exact analytical expressions. This is perfect for learning and verification.

19Simple Pole Example

A simple pole at s = -a in F(s) = 1/(s+a) gives f(t) = e^{-at}. The pole location directly determines the exponential decay rate.

30Complex Poles and Oscillation

Complex conjugate poles create oscillatory behavior. F(s) = ω/(s² + ω²) gives f(t) = sin(ωt). The imaginary part of the poles determines the frequency.

36Partial Fraction Decomposition

The KEY technique for inverse Laplace: break complicated F(s) into simpler fractions that match table entries. SymPy's apart() does this automatically.

55Numerical Inverse via Impulse Response

For engineering applications, scipy.signal computes the inverse Laplace numerically. The impulse response h(t) of a transfer function H(s) IS the inverse Laplace transform!

80Comparing Numerical and Analytical

The numerical and analytical methods should match. Discrepancies indicate either numerical errors or mistakes in the analytical solution.

93System Response Analysis

In control theory, the inverse Laplace transform converts transfer functions to time-domain responses. Poles determine stability and behavior characteristics.

147 lines without explanation

1import numpy as np
2import matplotlib.pyplot as plt
3from scipy import signal
4from sympy import symbols, inverse_laplace_transform, exp, sin, cos
5from sympy import apart, Heaviside
6
7# Define symbolic variables
8s, t = symbols('s t', positive=True, real=True)
9
10def symbolic_inverse_laplace(F_s):
11    """
12    Compute the inverse Laplace transform symbolically.
13
14    The inverse Laplace transform recovers f(t) from F(s):
15    f(t) = L^{-1}{F(s)}
16    """
17    f_t = inverse_laplace_transform(F_s, s, t)
18    return f_t
19
20# Example 1: Simple pole at s = -a
21# F(s) = 1/(s + a) → f(t) = e^{-at}
22F1 = 1 / (s + 2)
23f1 = symbolic_inverse_laplace(F1)
24print(f"L^{{-1}}{{1/(s+2)}} = {f1}")
25
26# Example 2: Double pole
27# F(s) = 1/(s + a)^2 → f(t) = t*e^{-at}
28F2 = 1 / (s + 3)**2
29f2 = symbolic_inverse_laplace(F2)
30print(f"L^{{-1}}{{1/(s+3)^2}} = {f2}")
31
32# Example 3: Complex poles (oscillation)
33# F(s) = ω/(s^2 + ω^2) → f(t) = sin(ωt)
34omega = 5
35F3 = omega / (s**2 + omega**2)
36f3 = symbolic_inverse_laplace(F3)
37print(f"L^{{-1}}{{ω/(s^2+ω^2)}} = {f3}")
38
39
40def partial_fraction_expansion(numerator, denominator):
41    """
42    Perform partial fraction decomposition for inverse Laplace.
43
44    This is the KEY technique: break F(s) into simpler fractions
45    that appear in standard Laplace transform tables.
46    """
47    F_s = numerator / denominator
48    # Use SymPy's apart() for partial fractions
49    decomposed = apart(F_s, s)
50    return decomposed
51
52
53# Example: Partial fraction decomposition
54# F(s) = (2s + 5) / (s^2 + 3s + 2)
55numerator = 2*s + 5
56denominator = s**2 + 3*s + 2  # = (s+1)(s+2)
57
58decomposed = partial_fraction_expansion(numerator, denominator)
59print(f"\nPartial fractions: {decomposed}")
60
61# Compute inverse of the decomposed form
62f_decomposed = symbolic_inverse_laplace(decomposed)
63print(f"Inverse: {f_decomposed}")
64
65
66def numerical_inverse_laplace(num_coeffs, den_coeffs, t_vals):
67    """
68    Numerically compute the inverse Laplace transform using
69    scipy's impulse response (for transfer functions).
70
71    H(s) = N(s)/D(s) represents a linear system.
72    The impulse response h(t) = L^{-1}{H(s)}.
73    """
74    # Create transfer function system
75    system = signal.TransferFunction(num_coeffs, den_coeffs)
76
77    # Compute impulse response (which IS the inverse Laplace)
78    t_out, y_out = signal.impulse(system, T=t_vals)
79
80    return t_out, y_out
81
82
83# Numerical example: H(s) = 1/(s^2 + 2s + 5)
84# This has complex poles at s = -1 ± 2j
85# So h(t) = (1/2)e^{-t}sin(2t)
86num = [1]
87den = [1, 2, 5]  # s^2 + 2s + 5
88t_vals = np.linspace(0, 10, 500)
89
90t_out, y_out = numerical_inverse_laplace(num, den, t_vals)
91
92# Analytical solution for comparison
93y_analytical = 0.5 * np.exp(-t_vals) * np.sin(2 * t_vals)
94
95# Plot comparison
96plt.figure(figsize=(10, 5))
97plt.plot(t_out, y_out, 'b-', linewidth=2, label='Numerical (scipy)')
98plt.plot(t_vals, y_analytical, 'r--', linewidth=2, label='Analytical')
99plt.xlabel('Time t')
100plt.ylabel('f(t)')
101plt.title(r'Inverse Laplace: $\mathcal{L}^{-1}\{1/(s^2+2s+5)\}$')
102plt.legend()
103plt.grid(True, alpha=0.3)
104plt.show()
105
106
107def analyze_system_response(num, den, input_type='impulse'):
108    """
109    Analyze a linear system's response using Laplace techniques.
110
111    In control theory:
112    - Transfer function H(s) = Y(s)/U(s)
113    - Output y(t) = L^{-1}{H(s) · U(s)}
114
115    This is why inverse Laplace transforms matter!
116    """
117    system = signal.TransferFunction(num, den)
118
119    t_vals = np.linspace(0, 10, 500)
120
121    if input_type == 'impulse':
122        # δ(t) → 1 in s-domain, so output = h(t)
123        t, y = signal.impulse(system, T=t_vals)
124        title = 'Impulse Response (inverse Laplace of H(s))'
125    elif input_type == 'step':
126        # u(t) → 1/s in s-domain, so output = L^{-1}{H(s)/s}
127        t, y = signal.step(system, T=t_vals)
128        title = 'Step Response (inverse Laplace of H(s)/s)'
129
130    # Find poles
131    poles = np.roots(den)
132
133    return t, y, poles, title
134
135
136# Analyze a second-order underdamped system
137num = [4]
138den = [1, 0.5, 4]  # s^2 + 0.5s + 4 (underdamped)
139
140t, y, poles, title = analyze_system_response(num, den, 'step')
141
142print(f"\nSystem poles: {poles}")
143print(f"Real part (decay rate): {poles[0].real:.3f}")
144print(f"Imaginary part (oscillation freq): {abs(poles[0].imag):.3f}")
145
146plt.figure(figsize=(10, 5))
147plt.plot(t, y, 'b-', linewidth=2)
148plt.xlabel('Time t')
149plt.ylabel('Response')
150plt.title(title)
151plt.grid(True, alpha=0.3)
152plt.axhline(y=1, color='r', linestyle='--', alpha=0.5, label='Steady state')
153plt.legend()
154plt.show()

Common Pitfalls

Forgetting the Region of Convergence

Different functions can have the same Laplace transform formula but different regions of convergence. Always check that your assumptions about $\\text{Re}(s) > a$ are satisfied.

Improper Fractions

If the degree of the numerator is greater than or equal to the denominator, you must perform polynomial long division first before applying partial fractions.

\\frac{s^2}{s + 1} = s - 1 + \\frac{1}{s + 1}

Mixing Up Shifting Theorems

The first shifting theorem (s-shift) and second shifting theorem (t-shift with unit step) are different:

s-shift: $F(s - a) \\leftrightarrow e^{at}f(t)$
t-shift: $e^{-as}F(s) \\leftrightarrow u(t-a)f(t-a)$

Sign Convention

Watch the sign in the shifting theorem carefully: $F(s - a)$ (replacing s with s - a) gives $e^{+at}$ , while $F(s + a)$ gives $e^{-at}$ .

Verification Strategy

Always verify your inverse transforms by checking:

Does $f(0^+)$ match the initial value theorem?
Does $\\lim_{t \\to \\infty} f(t)$ match the final value theorem (if applicable)?
Can you transform back to F(s) and get the original?

Test Your Understanding

Summary

The inverse Laplace transform is the essential final step that converts algebraic solutions back to time-domain functions. By mastering partial fractions and completing the square, you can invert most rational functions encountered in practice.

Key Concepts

Concept	Description
Inverse transform	Recovers f(t) from F(s): f(t) = L⁻¹{F(s)}
Linearity	L⁻¹{aF + bG} = aL⁻¹{F} + bL⁻¹{G}
s-shifting	F(s - a) ↔ eᵃᵗf(t)
Partial fractions	Decompose F(s) into table-matchable pieces
Completing square	Convert s² + bs + c to (s - a)² + ω²
Poles	Locations in s-plane determine time behavior

Key Takeaways

The Bromwich integral defines the inverse transform, but we almost never evaluate it directly
Partial fraction decomposition is the primary technique—break F(s) into simple pieces
Completing the square handles complex poles, converting to standard sine/cosine forms
Pole locations encode time-domain behavior: real parts give decay rates, imaginary parts give oscillation frequencies
The first shifting theorem handles exponentially modulated functions elegantly
Tables of transforms are essential references— memorize the common ones

The Essence of Inverse Transforms:

"Every pole tells a story about time. Real poles describe decay, imaginary poles describe oscillation, and their locations reveal the complete dynamics of the system."

Coming Next: In the next section, we'll put these techniques to work by Solving Initial Value Problems completely with Laplace transforms—transforming differential equations into algebra and back again.