SCHOOL OF ARTIFICIAL

INTELLIGENCE
PROBABLISTIC
DIFFERENTIAL EQUATION
THROUGH

KOOPMAN’S
OPERATOR
INSTITUTIONAL COURSE
23MAT112
MATHEMATICS FOR INTELLIGENT SYSTEMS -II
FACULTY INCHARGE

DR BISWAJIT DAS
BATCH A - 08
A.R. NIRMAL CB.AI.U4AID24001
N. SANJANA REDDY
CB.AI.U4AID24034
SRISHANTH S CB.AI.U4AID24056
T MUTHU RAAM
CB.AI.U4AID24058
1.Introduction
Nonlinear systems are found everywhere, from nature to
engineering, but analysing and controlling them is tough due to
their complex behaviour. The Koopman operator is a
mathematical tool that simplifies this by turning nonlinear
dynamics into a linear form in an infinite-dimensional space. This
makes it easier to analyse and control these systems, offering
powerful solutions in many fields.
2. Literature Review
Historical Background
the Koopman operator was first introduced by B.O. Koopman in
the 1930s to study how systems change over time. Over the
years, researchers have expanded its use, linking it to both
traditional physics and modern data-driven methods.
Major Developments
the Koopman operator has been successfully used in areas like
fluid dynamics, robotics, and control systems. For example,
Dynamic Mode Decomposition (DMD), a method based on
Koopman theory, helps find patterns in data like ocean waves or
airflow.
Recent Trends Today,
researchers are improving Koopman methods using machine
learning and data science. Applications include predicting
robotic movements and financial market trends, showing how
this approach can solve practical problems.
3. Mathematical Framework
What Is the Koopman Operator?
The Koopman operator works on functions, called observables,
that describe the state of a system. For instance, if a system
moves from state to , the Koopman operator transforms the
observable function into .
Key Ideas
 State Space: All possible states of the system.
 Observables: Features that describe system states (e.g.,
position or speed).
 Linearity: The Koopman operator lets us study nonlinear
systems with linear techniques.
4. Properties of the Koopman Operator
Eigenfunctions and Eigenvalues
Eigenfunctions and eigenvalues describe how the system
evolves. For example, eigenfunctions can represent stable
behaviours like steady population levels, while eigenvalues show
whether these behaviours grow or shrink.
Spectral Decomposition
The Koopman operator can break down into simpler parts using
its eigenfunctions and eigenvalues. This helps us understand
complex systems by focusing on key patterns, such as the stable
or unstable flow of water in a river.
5. Problem Definition
Identifying Nonlinear Systems Examples of nonlinear systems
include:
 Fluid Dynamics: Understanding airflow over an aircraft
wing.
 Robotics: Planning a robot’s movement.
 Electrical Circuits: Studying oscillations in power grids.
Setting Goals
The goal might be to predict future behaviour, control the
system, or analyse its stability. For instance, predicting
turbulence in an airplane's flight path can help design better
wings.
Project Scope
Clearly defining resources and methods is essential. For
example, analysing a robot’s movement might need sensor
data, Python for analysis, and realistic time limits for execution.
6. Data Collection
Sources of Data
Data can come from experiments, simulations, or existing
datasets. For example, wind tunnel experiments can provide
airflow data for fluid dynamics studies.
Data Preparation Steps include:
 Normalization: Ensures consistent data scales.
 Noise Removal: Cleans up errors.
 Time Sampling: Organizes data into regular intervals.
7. Approximation Methods
Dynamic Mode Decomposition (DMD)
DMD simplifies complex systems by finding key patterns, such
as wave movements in oceans.
Extended DMD (EDMD) EDMD
uses additional features to capture nonlinear behaviours, like a
robot’s joint angles and speeds.
Time-Delay Coordinates
This method uses past data to uncover hidden dynamics. For
instance, weather prediction systems benefit from analysing
past temperature trends.
Hankel-DMD
Organizing data into special matrices, Hankel-DMD is useful for
noisy datasets, like traffic flow studies, to extract reliable
patterns.
8. Implementing the Koopman Framework
Choosing Observables
Observables must highlight key system aspects, such as speed
or energy. For a robot, observables might be its joint angles and
speeds.
Numerical Tools
Tools like MATLAB or PyKoopman are used to compute and
analyse the Koopman operator.
Analysing
Modes Koopman modes describe system behaviours. For
example, in traffic analysis, slow modes show steady flows,
while fast modes signal congestion.
9. Data Science Integration
Using Machine Learning Machine learning can help identify the
best observables. For example, neural networks might uncover
hidden trends in stock prices for better predictions.
Dimensionality Reduction Techniques like Principal Component
Analysis (PCA) simplify complex data, making it easier to
analyse. This is useful in robotics, where PCA can focus on the
most important movements.
Validation Predictions from Koopman methods must be tested
against real-world data. For example, predicted robot
movements are compared to actual movements to check
accuracy.
10. Analysis of Results
Spectral Analysis By studying eigenvalues, we can understand
system stability and predict long-term behaviour. For instance, in
traffic management, eigenvalues can reveal whether congestion
will increase or decrease.
Koopman Modes These modes highlight dynamic patterns, like
eddies in water flows, providing actionable insights for system
design and control.
Real-Life Example: Simplifying Traffic Signals with
Koopman
Imagine analysing traffic at a busy intersection. The nonlinear
behaviour of traffic flow depends on many variables, such as
vehicle speed and signal timing. Using the Koopman operator,
we can transform this complexity into a linear model. This model
can predict when congestion will occur and suggest optimal
signal timings to reduce waiting times. Just like this, the
Koopman framework helps make complex systems easier to
understand and control.
Fundamentals of Koopman Operator
 Definition: The Koopman operator is an infinite-
dimensional linear operator that acts on observable
functions of the system's state, rather than the state
variables themselves.
 Key Idea: Nonlinear dynamics can be represented as linear
dynamics in an infinite-dimensional space of observables.
 Observable Functions: These are functions that map the
state of a system to measurable quantities, such as
velocity, position, or energy.
 Advantages: Provides a new perspective on system
dynamics, enabling linear analysis tools like eigenvalue
decomposition and modal analysis.
 Core Techniques:
o Dynamic Mode Decomposition (DMD): Extracts
dominant modes and frequencies from data.
o Extended DMD (EDMD): Enhances DMD by
incorporating a richer set of observables.

1. Predicting Turbulent Flow in Fluid Dynamics

  Scenario: Analysing airflows around an airplane wing.
Turbulence creates chaotic and nonlinear patterns, making
prediction challenging.
 Koopman Application: By selecting observables such as
velocity at key points in the flow, the Koopman operator
linearizes the airflow dynamics. Techniques like DMD extract
dominant flow modes, helping identify patterns like vortices
or laminar flow regions.
 Real-World Use: Aircraft designers use these insights to
reduce drag, enhance fuel efficiency, and improve stability
during flight.

2. Trajectory Planning for Robotic Arms

 Scenario: A robotic arm assembles a car by performing
precise, repetitive tasks. The nonlinear dynamics of joint
movements and external forces complicate control.
 Koopman Application: Observables such as joint angles,
velocities, and torque are fed into the Koopman framework.
EDMD is employed to approximate the operator, identifying
stable modes for accurate predictions of arm movements.
 Real-World Use: This enables smoother operations and
optimizations, such as reducing wear and tear on joints or
improving precision in high-speed manufacturing.

3. Monitoring Power Grids

 Scenario: Electrical grids experience fluctuations due to
varying demands, equipment failures, or environmental
conditions. These fluctuations are nonlinear and can
destabilize the grid.
 Koopman Application: Voltage, current, and frequency
data from grid nodes are used as observables. The
 Koopman operator helps identify slow dynamics (steady
load) and fast dynamics (sudden surges or faults).
 Real-World Use: Operators can predict and mitigate
instabilities, preventing blackouts and ensuring steady
electricity supply.

4. Modelling Traffic Patterns

 Scenario: Urban traffic exhibits nonlinear behaviour due to
congestion, traffic signals, and driver decisions. Predicting
traffic flow and identifying bottlenecks are critical for urban
planning.
 Koopman Application: Observables like vehicle speed,
density, and signal timings are analysed using Koopman-
based methods. Time-delay coordinates capture hidden
patterns in traffic flow.
 Real-World Use: City planners and traffic management
systems use these insights to optimize signal timings,
design better road networks, and reduce congestion.

5. Weather Forecasting
 Scenario: Weather systems exhibit chaotic dynamics due
to interactions between atmospheric pressure, temperature,
and wind patterns. Predicting storms or heatwaves requires
understanding these complex interactions.
 Koopman Application: Observables include pressure,
temperature, and wind speed at different altitudes. Time-
delay coordinates capture recurring patterns like seasonal
variations.
 Real-World Use: Meteorologists use Koopman-based
predictions to provide more accurate and timely weather
forecasts, improving disaster preparedness.
6. Stock Market Analysis
 Scenario: Stock prices fluctuate based on market trends,
company performance, and economic factors. These
nonlinear dynamics make accurate prediction difficult.
 Koopman Application: Observables like price movements,
trading volume, and market indices are analysed. Machine
learning-enhanced Koopman methods identify key modes of
growth or decline.
 Real-World Use: Investors and financial analysts use
Koopman predictions to optimize trading strategies and
manage risk.

7. Human Motion Analysis in Sports

 Scenario: Analysing the biomechanics of a sprinter involves
nonlinear dynamics due to muscle forces, joint angles, and
motion sequences.
 Koopman Application: Observables such as joint
velocities, ground reaction forces, and stride lengths are
modelled. The Koopman operator identifies inefficient
motion patterns or areas for improvement.
 Real-World Use: Coaches and athletes use these insights
to enhance performance and reduce injury risks.

8. Understanding Epidemic Spread

 Scenario: The spread of infectious diseases involves
nonlinear interactions between individuals and populations,
such as contact rates and immunity levels.
 Koopman Application: Observables include infection
rates, recovery rates, and population density. Koopman-
based spectral analysis identifies stable patterns (slow
spread) and unstable patterns (rapid outbreaks).
  Real-World Use: Health officials use this data to predict
outbreaks, design vaccination strategies, and allocate
resources effectively.

9. Simplifying Autonomous Vehicle Navigation

 Scenario: Self-driving cars navigate through nonlinear
interactions involving road geometry, traffic, and pedestrian
movements.
 Koopman Application: Observables include vehicle
position, speed, and sensor data (e.g., LiDAR). The Koopman
framework predicts nearby vehicle trajectories and adjusts
navigation strategies in real time.
 Real-World Use: Improved safety and efficiency in
autonomous vehicles, particularly in complex urban
environments.

Key Insights Across Examples

 Why Koopman Works: Nonlinear systems often contain
hidden linear structures in their observable space. The
Koopman operator exploits these structures to simplify
analysis.
 Tools for Implementation: Techniques like DMD, EDMD,
and machine learning enrich Koopman’s practicality across
domains.
 Impact: From improving flight efficiency to controlling
pandemics, the Koopman operator bridges theoretical
insights with practical applications.