Stability in Linear Systems — Stability and Long-Term Behavior in Linear Systems (Intro) - Linear Algebra

SECTION 1

Historical Context & Motivation

For centuries, scientists and engineers have asked a simple but important question: will this system stay under control, or will it fly apart? Whether it's a bridge swaying in the wind, an economy reacting to a policy change, or a rocket adjusting its path in space, understanding stability — whether a system settles down or spirals out of control — is one of the most practical ideas in all of mathematics.

The story of stability in linear systems begins with early work on differential equations and matrix theory. Mathematicians realized that the eigenvalues (special numbers associated with a matrix) hold the key to predicting what happens to a system over a long stretch of time. This idea connects algebra, geometry, and real-world engineering in a powerful way.

1750s • Euler and Vibrating Systems

Leonhard Euler studied vibrating strings and rotating bodies. He discovered special values that describe how physical systems oscillate — the earliest ideas behind eigenvalues.

1855 • Cayley Formalizes Matrices

Arthur Cayley introduced the modern concept of a matrix, giving mathematicians a compact way to represent and study linear systems with many variables at once.

1892 • Lyapunov's Stability Theory

Aleksandr Lyapunov published his doctoral thesis on the stability of motion. He proved that eigenvalues determine whether small disturbances in a system grow or fade away.

1940s–60s • Control Theory Boom

During World War II and the space race, engineers used eigenvalue-based stability analysis to design autopilots, guided missiles, and the control systems for the Apollo spacecraft.

The central question these pioneers explored still drives research today: given a system described by a matrix, can we look at the eigenvalues and immediately tell if the system is stable? The answer, as we'll see, is a resounding yes.

SECTION 2

Core Principles & Definitions

Before we dive into diagrams and equations, let's nail down the core ideas. A linear system is a set of equations where the variables interact in a straightforward, proportional way — no squaring, no square roots, just multiplication by constants and addition. A matrix packages all those constants into a neat grid of numbers. The eigenvalues of that matrix are special numbers that tell us how the system stretches, shrinks, or rotates over time.

Stable System

All eigenvalues have an absolute value (or "magnitude") less than 1 (for discrete steps) or a negative real part (for continuous change). The system shrinks toward an equilibrium over time.

Unstable System

At least one eigenvalue has magnitude greater than 1 (discrete) or a positive real part (continuous). The system grows without bound — it "blows up."

Marginally Stable

All eigenvalues sit exactly on the boundary (magnitude = 1 or real part = 0). The system neither grows nor shrinks; it may oscillate forever.

Eigenvalue Magnitude

The magnitude (absolute value) of an eigenvalue acts like a multiplier each time step. Values below 1 shrink things; values above 1 grow things.

✦

KEY TAKEAWAY

Think of eigenvalues like a volume knob on a speaker. If the knob is turned below the halfway mark (magnitude < 1), the sound fades away — that's a stable system. If the knob is cranked past the halfway mark (magnitude > 1), the sound keeps getting louder until it's unbearable — that's an unstable system. Right at the halfway mark, the sound stays at a constant level — that's marginal stability.

SECTION 3

Visualizing Stability

One of the best ways to understand stability is to watch what happens to a point as a matrix acts on it again and again. Imagine placing a dot on a grid, then applying the matrix transformation repeatedly. Depending on the eigenvalues, the dot either spirals inward (stable), spirals outward (unstable), or traces a loop forever (marginally stable).

The left panel shows a stable system spiraling toward the origin. The center panel shows a marginally stable system looping around forever. The right panel shows an unstable system flying away from the origin.

In the diagram above, each dot represents the system's state after another step. Notice how the stable path shrinks toward the center (the origin), meaning the system calms down. The unstable path moves farther and farther away — the system is growing out of control. The marginally stable path never grows or shrinks; it just keeps circling. These three behaviors correspond directly to whether the eigenvalues are small, large, or right on the boundary.

SECTION 4

Mathematical Framework

Let's put some math behind the pictures. A discrete-time linear system updates its state by multiplying a vector by a matrix at each time step. If the system starts at state x₀, then after one step the state becomes A × x₀, after two steps it becomes A² × x₀, and so on.

STATE UPDATE RULE

x(n) = Aⁿ × x₀

x(n) = state after n steps, A = the system matrix, x₀ = the starting state, n = number of time steps.

As n gets very large, the behavior of Aⁿ is controlled entirely by the eigenvalues of A. If λ is an eigenvalue, then λⁿ appears in the formula for Aⁿ. This leads us to the stability rule.

EIGENVALUE EQUATION

A × v = λ × v

A = system matrix, v = eigenvector (the direction that A stretches or shrinks), λ = eigenvalue (the stretch/shrink factor).

STABILITY CRITERION (DISCRETE TIME)

|λ| < 1 for ALL eigenvalues → STABLE

If every eigenvalue λ has absolute value (magnitude) less than 1, then λⁿ → 0 as n → ∞, meaning the system's state approaches the origin. The system is asymptotically stable.

💡

Why absolute value?

Eigenvalues can be negative or even complex numbers. A negative eigenvalue like λ = −0.5 will flip the sign each step, but its magnitude (0.5) is less than 1, so it still shrinks. That's why we always check |λ| (the absolute value or magnitude) rather than λ itself.

SECTION 5

Classifying Stability by Eigenvalue Location

We can create a handy map by plotting eigenvalues on a number line (for real eigenvalues) or in the complex plane (when eigenvalues involve imaginary numbers). For discrete-time systems, the magic boundary is the unit circle — a circle of radius 1 centered at the origin. Eigenvalues inside the circle mean stability; eigenvalues outside mean instability.

The dashed yellow circle is the unit circle of radius 1. Green eigenvalues inside the circle produce stable behavior. The red eigenvalue outside produces instability. The yellow eigenvalue on the boundary is marginally stable.

Eigenvalue Condition	Long-Term Behavior	Stability Type
All \|λ\| < 1	State → 0 (system dies out)	Asymptotically Stable
All \|λ\| ≤ 1, at least one \|λ\| = 1	State stays bounded, may oscillate	Marginally Stable
Any \|λ\| > 1	State → ∞ (system blows up)	Unstable

Summary of stability types for discrete-time linear systems

SECTION 6

Worked Example

Let's work through a complete example to see how you determine the stability of a system from its matrix.

Is this 2×2 system stable?

1
Step 1 — Write Down the Matrix

Suppose our system is described by the matrix A = [[0.5, 0.2], [0.1, 0.3]]. This means the state updates each step according to x(n+1) = A × x(n).
2
Step 2 — Find the Characteristic Equation

To find the eigenvalues, we solve det(A − λI) = 0, where I is the identity matrix. This gives us: (0.5 − λ)(0.3 − λ) − (0.2)(0.1) = 0. Expanding: λ² − 0.8λ + 0.13 = 0.

Characteristic equation: λ² − 0.8λ + 0.13 = 0
3
Step 3 — Solve for Eigenvalues

Using the quadratic formula: λ = (0.8 ± √(0.64 − 0.52)) ÷ 2 = (0.8 ± √0.12) ÷ 2. Since √0.12 ≈ 0.346, we get: λ₁ ≈ (0.8 + 0.346) ÷ 2 ≈ 0.573 and λ₂ ≈ (0.8 − 0.346) ÷ 2 ≈ 0.227.

λ₁ ≈ 0.573, λ₂ ≈ 0.227
4
Step 4 — Check Magnitudes

Both eigenvalues are real and positive. We check: |λ₁| = 0.573 < 1 ✓ and |λ₂| = 0.227 < 1 ✓. Since both magnitudes are less than 1, the stability criterion is satisfied.

|λ₁| = 0.573 < 1, |λ₂| = 0.227 < 1
5
Step 5 — State the Conclusion

Because every eigenvalue of A has magnitude less than 1, the system is asymptotically stable. No matter what starting state x₀ you choose, after enough steps the system will settle down to the origin (the zero vector).

The system is asymptotically stable.

SECTION 7

Strengths and Limitations of Eigenvalue Stability Analysis

Eigenvalue-based stability analysis is incredibly useful, but it's important to know when it works perfectly and when you might need more advanced tools.

Strengths	Limitations
Quick and definitive: just compute eigenvalues and check magnitudes.	Only works for linear systems — most real-world systems are nonlinear.
Works for any size matrix (2×2, 3×3, 100×100, etc.).	Computing eigenvalues of very large matrices can be computationally expensive.
Gives you the exact long-term behavior, not just an approximation.	Doesn't tell you what happens during the short-term (transient behavior).
Applies to both discrete-time and continuous-time systems with slight rule changes.	Marginal stability (\|λ\| = 1) cases can be tricky — small errors in the matrix can flip the result.

Pros and cons of eigenvalue stability analysis

✦

KEY TAKEAWAY

Eigenvalue stability analysis is like checking the weather forecast. It gives you an excellent prediction of long-term trends (will it rain all week?), but it may not capture short-term surprises (a sudden afternoon shower). For linear systems, though, the forecast is exact — eigenvalues never lie about where the system ends up.

SECTION 8

Connection to Advanced Theory

The ideas in this lesson are the foundation for much deeper topics. As you advance in math and engineering, you'll encounter systems that change continuously (not in discrete steps) and systems that are nonlinear (the proportional relationship breaks down). The table below previews how the stability rules adapt.

Feature	This Lesson (Discrete, Linear)	Advanced (Continuous / Nonlinear)
System type	x(n+1) = A × x(n)	dx/dt = A × x or dx/dt = f(x)
Stability test	All \|λ\| < 1	All eigenvalues have negative real parts (continuous linear); Lyapunov methods (nonlinear)
Boundary	Unit circle (\|λ\| = 1)	Imaginary axis (Re(λ) = 0)
Typical applications	Population models, digital filters, economic models	Circuit design, robotics, climate models, biological networks

How stability analysis extends to more advanced settings

The key idea carries over perfectly: eigenvalues are always the gatekeepers of stability. The specific rule changes (unit circle vs. imaginary axis), but the concept — check the eigenvalues to predict the future — stays the same. In future courses, you'll learn about Lyapunov stability for nonlinear systems and Bode plots for frequency-domain analysis, both of which build on these foundations.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

A 2×2 matrix has eigenvalues λ₁ = 0.6 and λ₂ = −0.4. Without doing any calculations, is the corresponding discrete-time system stable, unstable, or marginally stable? Explain your reasoning.

Show answer

The system is asymptotically stable. Both eigenvalues have magnitudes less than 1: |0.6| = 0.6 < 1 and |−0.4| = 0.4 < 1. Even though λ₂ is negative (which means the system flips sign each step along that direction), the magnitude is what matters for stability. Since both magnitudes are below the threshold, the state vector shrinks toward the origin over time.

Question 2: BASIC CALCULATION

Find the eigenvalues of the matrix A = [[0.8, 0], [0, 0.3]]. Then determine whether the system x(n+1) = A × x(n) is stable.

Show answer

Since A is a diagonal matrix, the eigenvalues appear right on the diagonal: λ₁ = 0.8 and λ₂ = 0.3. Now check magnitudes: |0.8| = 0.8 < 1 and |0.3| = 0.3 < 1. Both are inside the unit circle, so the system is asymptotically stable. After many steps, each component of the state shrinks — the first component is multiplied by 0.8 each step and the second by 0.3, so both approach zero.

Question 3: INTERMEDIATE

A matrix has eigenvalues λ₁ = 1.1 and λ₂ = 0.5. Is the system stable? If not, describe what happens as n → ∞. What would need to change about λ₁ to make the system stable?

Show answer

The system is unstable because |λ₁| = 1.1 > 1, even though |λ₂| = 0.5 < 1. For stability, every eigenvalue must have magnitude less than 1 — even one "bad" eigenvalue ruins stability. As n → ∞, the component of the state along the eigenvector for λ₁ grows by a factor of 1.1 each step: after 50 steps it would be (1.1)⁵⁰ ≈ 117 times larger. To make the system stable, λ₁ would need to have |λ₁| < 1, meaning λ₁ would need to be between −1 and 1.

Question 4: APPLIED

A biologist models two competing animal populations using the system x(n+1) = A × x(n), where A = [[0.7, 0.1], [0.2, 0.6]]. The eigenvalues of this matrix are λ₁ = 0.8 and λ₂ = 0.5. Interpret what stability means in this biological context. Will the populations survive long-term?

Show answer

Both eigenvalues have magnitude less than 1 (|0.8| = 0.8 and |0.5| = 0.5), so the system is asymptotically stable. In biological terms, this means both populations are declining toward zero over time. This might represent a scenario where both species are losing habitat or food resources faster than they can reproduce. The population associated with λ₁ = 0.8 declines more slowly (losing 20% per step) while the one associated with λ₂ = 0.5 declines faster (losing 50% per step). Long-term, neither population survives — they both approach extinction in this model.

Question 5: CRITICAL THINKING

Consider a matrix where both eigenvalues are complex numbers: λ₁ = 0.6 + 0.6i and λ₂ = 0.6 − 0.6i. Calculate |λ₁| using the formula |a + bi| = √(a² + b²). Is the system stable? Describe the qualitative behavior you would expect to see (hint: complex eigenvalues often produce rotation or spiraling).

Show answer

First, calculate the magnitude: |λ₁| = √(0.6² + 0.6²) = √(0.36 + 0.36) = √0.72 ≈ 0.849. Since λ₂ is the complex conjugate of λ₁, it has the same magnitude: |λ₂| ≈ 0.849. Both magnitudes are less than 1, so the system is asymptotically stable. The complex eigenvalues mean the system doesn't simply shrink along a straight line — instead, it spirals inward toward the origin. Each step, the state rotates a bit (due to the imaginary parts) and shrinks a bit (because the magnitude is 0.849, the state gets about 15% closer to zero each step). This spiraling-inward behavior is common in systems with damped oscillations, like a pendulum slowly losing energy to friction.

SUMMARY

Lesson Summary

Summary

The stability of a linear system is determined entirely by its eigenvalues. For a discrete-time system x(n+1) = A × x(n), the system is asymptotically stable when every eigenvalue has magnitude less than 1 (inside the unit circle), unstable when any eigenvalue has magnitude greater than 1, and marginally stable when eigenvalues sit exactly on the boundary.

Visually, stable systems spiral inward toward the origin, unstable systems fly outward without bound, and marginally stable systems loop forever at a constant distance. This eigenvalue-based approach works for any size matrix and extends naturally to continuous-time systems and nonlinear analysis in advanced courses, making it one of the most fundamental tools in mathematics and engineering.