Linear Systems [Control Bootcamp]

Steve Brunton

5 chapters7 takeaways12 key terms5 questions

Overview

This video reviews the fundamental concepts of linear systems of ordinary differential equations (ODEs), specifically the form x-dot = Ax. It explains how to solve these systems using matrix exponentials, the role of eigenvalues and eigenvectors in simplifying the analysis, and how to transform nonlinear systems into linear approximations. The video also introduces the concept of controllability in the context of adding control inputs (Bu) and previews upcoming topics like discrete-time systems and stability analysis.

How was this?

Save this permanently with flashcards, quizzes, and AI chat

Chapters

Linear systems of ODEs are represented as x-dot = Ax, where x is a state vector and A is a matrix.
The general solution is x(t) = exp(At) * x(0), where exp(At) is the matrix exponential.
The matrix exponential exp(At) can be defined using its Taylor series expansion.
While the Taylor series is a definition, it's often computationally impractical for large matrices.

Understanding the basic form of linear systems and their solutions is crucial because it forms the foundation for analyzing more complex systems, including those with nonlinearities or control inputs.

The system x-dot = Ax, where x is a vector and A is a matrix describing how the state's rate of change depends on the state itself.

Eigenvalues (lambda) and eigenvectors (xi) satisfy the equation A*xi = lambda*xi.
Eigenvectors represent special directions in the state space that are only scaled by the matrix A, not changed in direction.
Using a matrix T of eigenvectors and a diagonal matrix D of eigenvalues, we can relate A to D via A = TDT^-1.
Transforming the system into eigenvector coordinates (z = T^-1x) simplifies the dynamics to z-dot = Dz, where D is diagonal.

Eigenvalue decomposition transforms a complex, coupled system into a set of simple, decoupled scalar differential equations, making analysis and computation significantly easier.

The relationship A*xi = lambda*xi, where multiplying the eigenvector xi by matrix A results in a scaled version of xi.

In eigenvector coordinates (z), the system becomes z-dot = Dz, where D is a diagonal matrix of eigenvalues.
This diagonal form means each component of z evolves independently: z_i-dot = lambda_i * z_i.
The solution for each decoupled component is z_i(t) = exp(lambda_i * t) * z_i(0).
The matrix exponential of a diagonal matrix is easy to compute: exp(Dt) has exp(lambda_i * t) on the diagonal.

By decoupling the system into independent modes represented by eigenvalues, we can easily find the solution for each mode and thus the overall system behavior.

The decoupled equations z1_dot = lambda1*z1 and z2_dot = lambda2*z2, where the evolution of z1 only depends on z1 and z2 only depends on z2.

The solution in original coordinates x(t) can be found by transforming the solution in z-coordinates back: x(t) = T * z(t).
This involves using the relationship exp(At) = T * exp(Dt) * T^-1.
The overall solution process is: map initial conditions x(0) to z(0) using T^-1, solve for z(t) using the diagonal system, and map back to x(t) using T.
This eigenvector-based approach provides an intuitive understanding of system dynamics by breaking them into fundamental modes.

While analysis is easier in eigenvector coordinates, the ultimate goal is often to understand and control the system in its original physical coordinates, requiring a clear mapping between the two.

The formula x(t) = T * exp(Dt) * T^-1 * x(0), which shows how to compute the state in original coordinates using the transformed solution.

Control is introduced by adding a 'Bu' term to the system dynamics, resulting in x-dot = Ax + Bu.
Controllability asks whether it's possible to steer the system to any desired state using the available control input (u) and actuator (B).
Not all systems are controllable; some combinations of system dynamics (A) and control input (B) might prevent arbitrary state manipulation.
A simple test for controllability will be discussed later, applied to an example like the pendulum.

Understanding controllability is essential for designing effective control systems, as it determines whether the desired behavior can actually be achieved with the given system and actuators.

The pendulum example, where depending on the control input, it might be possible to stabilize it in an upright position or not.

Key takeaways

1Linear systems of ODEs (x-dot = Ax) can be solved using the matrix exponential exp(At).
2Eigenvalues and eigenvectors provide a coordinate transformation that decouples the system into independent modes.
3Analyzing and solving the system in eigenvector coordinates (z-dot = Dz) is significantly simpler than in the original coordinates.
4The solution in original coordinates can be recovered by transforming back from eigenvector coordinates.
5The matrix exponential exp(At) can be computed more easily using the eigenvalue decomposition: exp(At) = T * exp(Dt) * T^-1.
6Controllability is a fundamental property that determines if a system can be driven to any desired state using control inputs.
7Understanding linear systems is foundational for analyzing and controlling more complex, nonlinear systems.

Key terms

Linear SystemOrdinary Differential Equation (ODE)State Vector (x)System Matrix (A)Matrix Exponential (exp(At))Eigenvalue (lambda)Eigenvector (xi)Eigen decompositionDiagonalizationControllabilityControl Input (u)Actuator (B)

Test your understanding

1What is the general solution to a linear system of ODEs represented as x-dot = Ax?
2How do eigenvalues and eigenvectors help in simplifying the analysis of linear systems?
3Why is it beneficial to analyze a linear system in its eigenvector coordinates?
4What does the concept of controllability imply for a linear system with a control input (x-dot = Ax + Bu)?
5How can the matrix exponential exp(At) be computed more efficiently using eigenvalues and eigenvectors?