Time-Stepping for ODEs and Stability
ODE, time stepping, Euler method, stability, stiffness
1 Role
This is an optional extension page in the Numerical Methods module.
Its job is to show how the approximation viewpoint from the previous page turns into actual time-marching algorithms for differential equations, and why repeated stepping introduces a new concern:
a method can look locally reasonable and still behave badly over many steps
2 First-Pass Promise
Read this page after Approximation, Differentiation, Integration, and Error Control.
If you stop here, you should still understand:
- why numerical ODE solving means replacing continuous time by discrete steps
- why local truncation error is not the whole story
- how explicit and implicit Euler differ
- why stiffness makes stability dominate step-size choice
3 Why It Matters
Many continuous models are written as ODEs:
- mechanical systems
- circuits and controls
- chemical kinetics
- population models
- gradient-flow and continuous-time optimization viewpoints
To compute with them, we replace the exact trajectory by values on a time grid:
\[ t_0,\ t_1,\ t_2,\dots \]
and build update rules that move from one time level to the next.
That creates a new type of numerical question.
For linear solves or quadrature, we usually focus on the quality of one approximation.
For ODEs, we must also ask:
- what happens after many repeated steps?
- does the method amplify harmless numerical noise?
- does the method respect decay that the true solution should have?
That is why stability is central.
4 Prerequisite Recall
- finite-difference ideas replace derivatives by local formulas
- truncation error measures how much a finite formula misses the exact continuous object
- adaptive error control balances computational cost against approximation quality
5 Intuition
5.1 Time Stepping As Repeated Approximation
For an ODE
\[ y'(t)=f(t,y(t)), \]
a one-step method turns the derivative information into an update from y_n to y_{n+1}.
So the method is not only approximating one derivative. It is composing the same approximation over and over.
5.2 Explicit Euler
The simplest idea uses the slope at the current point:
\[ y_{n+1}=y_n+h f(t_n,y_n). \]
This is cheap and intuitive, but stability can force the step size h to be small.
5.3 Implicit Euler
Another idea uses the slope at the next time level:
\[ y_{n+1}=y_n+h f(t_{n+1},y_{n+1}). \]
Now each step is more expensive because we must solve an equation for y_{n+1}, but the method is often much more stable.
5.4 Stiffness
Some ODEs contain rapidly decaying modes that do not matter much to the long-term shape of the solution, but they still force explicit methods to take tiny steps for stability.
That is the first-pass intuition for stiffness:
the dynamics are easy to describe qualitatively, but hard to step through explicitly without very small h
6 Formal Core
Definition 1 (Definition: One-Step Time-Stepping Method) A one-step method updates the approximate solution by a rule of the form
\[ y_{n+1}=\Phi_h(t_n,y_n). \]
The step size h controls both the cost and the approximation quality.
Definition 2 (Definition: Explicit Euler) For
\[ y'(t)=f(t,y(t)), \]
explicit Euler uses
\[ y_{n+1}=y_n+h f(t_n,y_n). \]
Definition 3 (Definition: Implicit Euler) Implicit Euler uses
\[ y_{n+1}=y_n+h f(t_{n+1},y_{n+1}). \]
This usually requires solving a nonlinear or linear equation at each step.
Definition 4 (Definition: Local Truncation Error) The local truncation error measures the one-step defect obtained by inserting the exact solution into the numerical update rule.
This is the local accuracy story, but repeated stepping means we also need a stability story.
Theorem 1 (Theorem Idea: Stability Is Read Through The Test Equation) A standard first-pass stability lens studies the scalar test equation
\[ y'=\lambda y. \]
Applying a method gives an update
\[ y_{n+1}=R(h\lambda)\,y_n, \]
where R is the method’s stability function.
The method is stable for that step size when repeated multiplication by R(h\lambda) does not create the wrong long-time behavior.
For explicit Euler,
\[ R(z)=1+z. \]
For implicit Euler,
\[ R(z)=\frac{1}{1-z}. \]
7 A Small Worked Example
Consider the decaying ODE
\[ y'=-10y, \qquad y(0)=1. \]
The exact solution is
\[ y(t)=e^{-10t}, \]
so the true behavior is monotone decay.
7.1 Explicit Euler
Explicit Euler gives
\[ y_{n+1}=(1-10h)y_n. \]
To keep the magnitude from growing, we need
\[ |1-10h|<1, \]
which means
\[ 0<h<0.2. \]
So even for a very simple decaying problem, stability already restricts the step size.
7.2 Implicit Euler
Implicit Euler gives
\[ y_{n+1}=\frac{1}{1+10h}y_n. \]
For every h>0, this factor has magnitude less than 1, so the method keeps the decaying behavior for any positive step size.
This example is the first-pass stability lesson:
- explicit Euler is cheap but conditionally stable
- implicit Euler is more expensive per step but much more stable for decaying modes
8 Computation Lens
When you face a time-stepping problem, ask:
- is the main bottleneck accuracy or stability?
- if I halve the step size, do I gain real accuracy or am I only paying a stability tax?
- does the method need to preserve decay, boundedness, or some qualitative structure of the true solution?
- is the equation stiff enough that an implicit method is the more honest computational model?
Those questions usually matter more than memorizing a list of named schemes.
9 Application Lens
9.1 Simulation And Engineering
Physical simulation often depends on whether the method tracks the right long-time behavior, not only whether one step is locally accurate.
9.2 Control And Dynamics
State evolution, feedback models, and discretized continuous-time systems all rely on time stepping and stability language.
9.3 Optimization And Continuous-Time Views
Gradient flows and dynamical-system interpretations of optimization become computational only after choosing a time discretization, and different discretizations behave differently.
10 Stop Here For First Pass
If you can now explain:
- why ODE solvers are repeated local approximations
- why stability is a separate concern from local truncation error
- why explicit and implicit Euler behave differently on decaying dynamics
- why stiffness is the reason some problems force tiny explicit time steps
then this page has done its job.
11 Go Deeper
The strongest adjacent pages are:
12 Optional Deeper Reading After First Pass
The strongest current references connected to this page are:
- MIT 18.335J resource index - official MIT course resource map showing numerical methods for ODEs as a central lecture block. Checked
2026-04-25. - MIT 18.086 lecture 1 difference methods for ODEs - official MIT lecture resource for the basic difference-method and stability viewpoint. Checked
2026-04-25. - MIT 2.086 stiff equations lecture - official MIT lecture page focused on stiff ODEs. Checked
2026-04-25. - Stanford CME108 syllabus pdf - official syllabus showing ODE time stepping as a natural continuation of the approximation toolkit. Checked
2026-04-25. - Stanford CS137 syllabus - official current scientific-computing syllabus linking quadrature and ODE methods in one numerical arc. Checked
2026-04-25. - Stanford CME108 bulletin - official course description placing ODEs inside the broader scientific-computing curriculum. Checked
2026-04-25.
13 Sources and Further Reading
- MIT 18.335J resource index -
First pass- official MIT course map for the ODE-methods block. Checked2026-04-25. - MIT 18.086 lecture 1 difference methods for ODEs -
First pass- official MIT lecture resource for difference methods and stability ideas. Checked2026-04-25. - MIT 2.086 stiff equations lecture -
First pass- official MIT lecture page focused on stiff ODE intuition. Checked2026-04-25. - Stanford CME108 syllabus pdf -
Second pass- official syllabus placing time stepping inside a larger scientific-computing toolkit. Checked2026-04-25. - Stanford CS137 syllabus -
Second pass- official current syllabus linking numerical integration and ODE methods. Checked2026-04-25. - Stanford CME108 bulletin -
Second pass- official course description connecting ODE methods to the broader scientific-computing curriculum. Checked2026-04-25.