Linear Systems, Matrix Exponentials, and Modes

How first-order linear systems evolve through matrices, why the matrix exponential is the exact continuous-time evolution operator, and how eigenvectors reveal the modes that grow, decay, or oscillate.
Modified

April 26, 2026

Keywords

linear systems, matrix exponential, modes, eigenvalues, state space

1 Role

This is the third page of the ODEs and Dynamical Systems module.

Its job is to show how the state-space rewrite from the previous page becomes a genuine linear-algebra story:

matrix multiplication gives the local rule, and the matrix exponential gives the exact flow

2 First-Pass Promise

Read this page after Second-Order Systems, State Variables, and Reduction to First Order.

If you stop here, you should still understand:

  • what a linear time-invariant system y' = Ay is
  • why e^{tA} is the exact continuous-time evolution operator
  • how eigenvectors and eigenvalues become dynamical modes
  • why coupling can often be understood by changing coordinates to modal variables

3 Why It Matters

Once a model is written in state-space form, linear systems are the first major tractable class.

They appear in:

  • oscillators and coupled mechanical systems
  • circuits and filters
  • local models near equilibria
  • control and state estimation
  • numerical time stepping and stability analysis
  • diffusion, flow, and continuous-depth models in ML

The key step is to stop thinking only in coordinates and start thinking in operators:

  • the matrix A tells us the local rule for change
  • the matrix exponential e^{tA} tells us the exact finite-time evolution
  • eigenvectors tell us the preferred directions of motion
  • eigenvalues tell us whether those directions grow, decay, or oscillate

This is where linear algebra becomes dynamics instead of only static transformation.

4 Prerequisite Recall

  • the previous page rewrote second-order equations as first-order state-space systems
  • eigenvalues and eigenvectors describe invariant directions of a matrix
  • diagonalization is useful when a matrix has enough independent eigenvectors
  • numerical methods later approximate flows of systems, but here we are describing the exact linear flow

5 Intuition

5.1 A Linear System Evolves A Whole State Vector

A linear time-invariant system has the form

\[ y'(t)=Ay(t), \]

where y(t) is a vector and A is a constant matrix.

The derivative of the whole state is determined by multiplying the current state by A.

5.2 The Matrix Exponential Is The Right Exact Analogue Of e^{\\lambda t}

For the scalar equation

\[ x'=\lambda x, \]

the solution is

\[ x(t)=e^{\lambda t}x(0). \]

The matrix exponential is the system version of that formula:

\[ y(t)=e^{tA}y(0). \]

5.3 Modes Decompose Complicated Motion Into Simpler Pieces

If v is an eigenvector of A with eigenvalue \lambda, then motion started exactly in the direction v stays in that direction and evolves like

\[ e^{\lambda t}v. \]

That is a mode.

So each eigenpair gives a basic pattern:

  • real negative eigenvalue: decay
  • real positive eigenvalue: growth
  • purely imaginary or complex eigenvalues: oscillation or rotation

5.4 Coupling Often Looks Harder Than It Is

In the original coordinates, variables may appear strongly coupled.

But if we move to modal coordinates, the dynamics often decouple or nearly decouple.

That is why eigenvectors matter dynamically, not just algebraically.

6 Formal Core

Definition 1 (Definition: Linear Time-Invariant System) A linear time-invariant first-order system has the form

\[ y'(t)=Ay(t)+r(t), \]

where A is a constant matrix and r(t) is a forcing term.

The homogeneous case is

\[ y'(t)=Ay(t). \]

Definition 2 (Definition: Matrix Exponential) For a square matrix A, the matrix exponential is defined by the power series

\[ e^{tA} = I + tA + \frac{t^2}{2!}A^2 + \frac{t^3}{3!}A^3 + \cdots \]

This is the system-level version of the scalar exponential.

Theorem 1 (Theorem Idea: Exact Solution Of A Homogeneous Linear System) For

\[ y'(t)=Ay(t), \qquad y(0)=y_0, \]

the exact solution is

\[ y(t)=e^{tA}y_0. \]

This is the load-bearing fact behind linear systems.

Theorem 2 (Theorem Idea: Eigenvectors Produce Modes) If

\[ Av=\lambda v, \]

then

\[ y(t)=e^{\lambda t}v \]

is a solution of the homogeneous system y' = Ay.

So eigenvectors define invariant directions and eigenvalues define the time behavior along those directions.

Theorem 3 (Theorem Idea: Diagonalization Turns Coupled Dynamics Into Modal Dynamics) If A = V\Lambda V^{-1} is diagonalizable, then

\[ e^{tA}=V e^{t\Lambda} V^{-1}, \]

and the system can be understood by evolving each modal coordinate separately.

This is why modal decomposition is one of the cleanest first lenses for linear systems.

7 A Small Worked Example

Consider the system

\[ y'(t)= \begin{bmatrix} 0 & 1 \\ -2 & -3 \end{bmatrix} y(t). \]

This is exactly the state-space form of the second-order equation

\[ x''+3x'+2x=0. \]

7.1 Step 1: Find The Eigenvalues

The characteristic polynomial is

\[ \lambda^2 + 3\lambda + 2 = (\lambda+1)(\lambda+2), \]

so the eigenvalues are

\[ \lambda_1=-1,\qquad \lambda_2=-2. \]

7.2 Step 2: Read The Modes

Each eigenvalue gives a decaying exponential mode:

  • one decays like e^{-t}
  • the other decays like e^{-2t}

So every solution is a combination of two decaying modes.

7.3 Step 3: Read The Qualitative Dynamics

Because both eigenvalues are negative:

  • trajectories move toward the origin
  • there is no sustained oscillation
  • one mode decays faster than the other, so long-time behavior is eventually dominated by the slower mode e^{-t}

Even without writing the full closed-form solution, the modal picture already tells us what the system does.

8 Computation Lens

The matrix exponential is the exact finite-time flow of a linear system.

Numerical solvers such as Euler, Runge-Kutta, or exponential integrators are trying to approximate that exact flow.

This is why linear systems are such a useful benchmark:

  • exact evolution is available in principle through e^{tA}
  • numerical methods can be compared against that exact linear benchmark
  • spectral structure of A controls stiffness, transient decay, and long-time behavior

This page therefore sits directly between exact dynamics and Time-Stepping for ODEs and Stability.

9 Application Lens

9.1 Oscillators And Coupled Systems

Mass-spring and circuit models become matrix systems once several state variables evolve together.

9.2 Control And State-Space Models

Modern control begins with linear systems because modal structure, reachability, and stability are easiest to analyze there.

9.3 Continuous-Time ML

Linear flows are the simplest continuous-depth models, and they also provide the local template used to analyze nonlinear systems near equilibria.

10 Stop Here For First Pass

If you can now explain:

  • what a linear system y' = Ay is
  • why e^{tA} is the exact evolution operator
  • why eigenvectors are modes
  • why eigenvalues determine growth, decay, and oscillation

then this page has done its job.

11 Go Deeper

After this page, the next natural step is:

The strongest adjacent pages are:

12 Optional Deeper Reading After First Pass

The strongest current references connected to this page are:

13 Sources and Further Reading

Back to top