Questions: Stiff Differential Equations and Stability Regions

Question 1

You are integrating a system of ODEs where the solution is smooth and barely changes over the interval [0, 1], yet your explicit Runge-Kutta solver is forced to take over 50,000 steps. What does this indicate, and what is the correct diagnosis?

• A. The solver has a bug — a smooth solution should require very few steps with any correct method
• B. The solution changes rapidly at a microscopic scale not visible in the plot, requiring small steps for accuracy
• C. The system is stiff: large-magnitude eigenvalues force the step size below 2/|λ_max| for stability, even though the solution of interest varies slowly
• D. The step size is limited by the slow component's timescale, which requires proportionally fine resolution

Answer: (C)

This is the hallmark of stiffness. The small step size is forced by the stability region of the explicit method, not by accuracy requirements. An explicit method like RK4 has a bounded stability region: for the test equation y' = λy, the method is only stable when hλ lies within a finite disk in the complex plane. A large negative eigenvalue (say λ = −1000) requires h < 2/1000 = 0.002 for stability, even if the solution component tied to that eigenvalue has already decayed to quasi-steady state and the interesting dynamics are slow. The solver takes many steps to stay stable, not to stay accurate.

Question 2

What is A-stability, and why does it make implicit methods practical for stiff equations while explicit methods remain impractical?

• A. A-stability means the method converges faster than explicit methods, allowing fewer iterations of Newton's method per step
• B. A-stability means the error decays faster in implicit methods, so accuracy is achieved with fewer steps
• C. A-stable methods have stability regions covering the entire left half-plane, so they remain stable for any step size h when Re(λ) < 0 — eliminating the step-size constraint imposed by fast eigenvalues
• D. A-stability guarantees the method is second-order accurate or higher, which is necessary to handle stiff dynamics

Answer: (C)

The stability region is the set of hλ values for which the numerical solution doesn't blow up on the test equation y' = λy. Explicit methods (Euler, RK4) have bounded stability regions — a small disk or finite blob in the complex plane — requiring h < constant/|λ_max|. For a stiff system with |λ_max| = 1000, this forces h < 0.002. An A-stable implicit method (backward Euler, implicit Runge-Kutta) has a stability region covering the entire left half-plane: any h is stable when Re(λ) < 0. This removes the fast-eigenvalue bottleneck entirely. The cost is solving a nonlinear system at each step, which is worthwhile when it allows steps 1000× larger.

Question 3

For a stiff ODE, an explicit solver is forced to take very small steps because the solution changes extremely rapidly.

• T. True
• F. False

Answer: False

This is the central misconception about stiffness. Explicit solvers are forced to take small steps because of stability requirements, not because the solution is changing rapidly. The solution may be smooth, slowly varying, and easy to plot — but the system's eigenvalue structure (with one or more large-magnitude negative eigenvalues) requires h to be smaller than 2/|λ_max| just to prevent the numerical solution from oscillating wildly. The fast-decaying modes may have already settled to near-zero; the solver is still 'afraid' of them because it cannot tell they've settled.

Question 4

The practical test for stiffness is behavioral: if an explicit ODE solver takes far more steps than the smoothness of the solution appears to require, the system is likely stiff and should be passed to an implicit solver.

• T. True
• F. False

Answer: True

True. Since stiffness is defined by the ratio of eigenvalue magnitudes times the integration interval (not by an intrinsic property of f alone), the most reliable diagnostic is to watch what happens when you attempt to integrate with an explicit method. An explicit solver that takes orders of magnitude more steps than the visual smoothness of the solution warrants is hitting the eigenvalue-stability wall. The solution: hand the problem to an implicit solver like backward Euler, Radau, or BDF methods, which are designed for exactly this situation.

Question 5

Explain why implicit methods can take much larger step sizes than explicit methods for stiff equations, and what computational cost they pay in exchange.

Model answer: Implicit methods (such as backward Euler or implicit Runge-Kutta) have A-stable stability regions covering the entire left half-plane, meaning they remain numerically stable for any step size h as long as the underlying ODE is stable (Re(λ) < 0). This removes the constraint h < 2/|λ_max| that cripples explicit methods on stiff problems. The cost is that implicit methods require solving a (possibly nonlinear) algebraic system at each step — typically via Newton's method — whereas explicit methods simply evaluate f. For stiff problems, this extra work per step is repaid many times over by the ability to take steps thousands of times larger.

The trade-off is explicit computation vs. stability range. Explicit methods are cheap per step but trapped by eigenvalues. Implicit methods are expensive per step but free from that trap. For non-stiff problems the extra work is wasteful; for stiff problems it is the only viable approach.