Numerical Methods for Ordinary Differential Equations

Numerical Methods for Ordinary Differential Equations develops a practical route from approximation tools to reliable differential-equation solvers. C. Vuik, F. J. Vermolen, M. B. van Gijzen, and M. J. Vuik connect interpolation, differentiation, nonlinear equations, and quadrature with initial-value and boundary-value methods. Eight chapters also include an optional extension to the time-dependent heat equation. This guide follows the linked TU Delft text and emphasizes error sources, convergence evidence, stability, and the distinction between a computed answer and a trustworthy numerical result.

Chapter-wise summaries and study tasks

Chapter 1. Introduction

Numerical mathematics produces approximations whose usefulness depends on both the mathematical problem and the computation. Floating-point representation creates rounding effects, and subtracting nearby quantities can expose cancellation. Error propagation must be separated from the sensitivity of the underlying problem. Landau notation describes asymptotic error behavior, while analysis concepts supply conditions needed for later bounds.

A smaller discretization parameter does not guarantee improvement when arithmetic or data error dominates. Establish a baseline with a known solution, compare absolute and relative errors where meaningful, and state the regime in which an order estimate is expected to apply. Numerical evidence becomes convincing when the observed behavior can be related to a predicted mechanism.

Study task. Compare two algebraically equivalent evaluations that differ through cancellation. Record errors over a range of inputs and explain whether the discrepancy comes from sensitivity or implementation.

Chapter 2. Interpolation

Interpolation constructs an approximating function that matches specified data, with the method and nodes determining behavior between them. Linear and higher-order Lagrange interpolation provide different balances among complexity and accuracy. Taylor and Hermite material incorporates derivative information, while splines assemble local polynomial pieces under continuity conditions. An error formula depends on smoothness and node placement rather than degree alone.

Passing exactly through the supplied points is not sufficient evidence of good approximation elsewhere. Compare the function, interpolant, and error on an independent grid. Local methods and global polynomials can behave differently near boundaries or with many nodes. Interpolation becomes a building block for differentiation and integration, so a representation choice can influence several later algorithms.

Study task. Compare linear interpolation, a higher-degree polynomial, and a spline for one function. Examine errors away from the nodes and explain why exact agreement at data points does not establish overall quality.

Chapter 3. Numerical differentiation

Numerical differentiation approximates derivatives through weighted combinations of sampled values. Forward, backward, and centered differences have different leading truncation errors, while higher-derivative and general formulas can be derived from expansions or interpolation. Reducing spacing can initially improve accuracy but eventually amplify rounding or measurement error. Richardson extrapolation combines results at related spacings to cancel a leading error term when the assumed expansion is valid.

The practical task is to balance error sources rather than choose the smallest possible step. An observed order estimate is informative only in the expected asymptotic regime. Noisy data may require a different strategy from exact evaluations of a smooth analytical function.

Study task. Compute a known derivative across decreasing spacings. Identify truncation- and roundoff-dominated regions, then test Richardson extrapolation where the predicted leading-order behavior is visible.

Chapter 4. Nonlinear equations

Nonlinear equations require an iterative strategy and conditions for success. Bisection uses a sign-changing bracket and continuity for controlled narrowing. Fixed-point iteration depends on the chosen mapping and its local behavior. Newton–Raphson uses derivative information for rapid local convergence under suitable assumptions but can fail from an unsuitable start or near problematic derivatives. Systems extend the approach through Jacobian information and linear solves.

A small step or residual does not establish a small solution error in every problem. Record stopping criteria and consider sensitivity. Safeguards, a bracket when available, and comparison with an independent method make an iteration more dependable than an unmonitored loop.

Study task. Solve one equation by bisection, fixed-point iteration, and Newton’s method from several starts. Compare evaluations, residuals, and actual error, explaining a failure through the method’s conditions.

Chapter 5. Numerical integration

Numerical integration combines sampled values through rules derived from approximation. Rectangle, midpoint, trapezoidal, and Simpson formulas have different error behavior and requirements. Composite rules divide an interval into subintervals, while interpolatory and Gaussian quadrature extend the choice of weights and nodes. Measurement and rounding effects remain relevant as contributions accumulate.

Compare efficiency through function evaluations and achieved accuracy, not only nominal order. Smoothness, oscillation, and endpoint behavior can change the practical result. A refinement comparison can estimate error when the relevant asymptotic model holds, but it should be checked against a known integral or independent calculation during development.

Study task. Compare composite rules at matched evaluation budgets. Estimate their convergence rates and repeat with a less favorable integrand to identify which assumptions explain the change.

Chapter 6. Numerical time integration of initial-value problems

Time integration advances an initial-value problem through successive approximations. The chapter connects existence and uniqueness with single-step methods, local truncation error, global error, and stability. Higher-order methods and Richardson estimates support accuracy assessment. Systems introduce amplification matrices and eigenvalue-based reasoning. Stiffness shows why a physically decaying solution can impose severe stability restrictions on an explicit method; optional multistep material extends the toolkit.

Consistency is a local approximation property; dependable accumulated behavior also needs stability. An implicit method may require a nonlinear solve at every step, making its tolerances and cost part of the analysis. Compare physical dynamics with numerical amplification rather than judging a solver by a smooth-looking curve.

Study task. Integrate nonstiff and stiff test problems with explicit and implicit methods. Vary the step and distinguish instability from an inaccurate but stable approximation.

Chapter 7. The finite-difference method for boundary-value problems

Boundary-value problems specify conditions at separated locations. Finite differences convert derivatives into algebraic systems whose boundary rows must represent the prescribed conditions. Consistency, stability, convergence, and matrix conditioning connect local formulas with the global solution. Neumann conditions, convection–diffusion examples, and nonlinear problems add complications beyond a simple second-derivative stencil. Picard and Newton iterations reuse earlier tools.

Refinement must test the entire scheme, including boundaries. A high-order interior formula does not establish the same order with a poorly treated boundary. Separate algebraic-solver error from discretization error so an unfinished solve is not mistaken for a failure of the underlying approximation.

Study task. Discretize a problem with a known solution and compare meshes. Check boundaries separately, then replace a value condition with a derivative condition and explain the changed matrix.

Chapter 8. The instationary heat equation

The optional final chapter treats the time-dependent heat equation, extending beyond ordinary differential equations. Spatial semi-discretization produces a system of time-dependent unknowns that can be studied with the earlier time-integration analysis. Forward Euler, backward Euler, and the trapezoidal method have different accuracy and stability behavior. Spatial resolution and time-step choice must be considered together.

Refining the grid can make the semi-discrete system increasingly restrictive for an explicit method. A stable step does not remove spatial error, and a fine spatial grid cannot compensate for an unsuitable integrator. The example shows how methods combine, each contributing assumptions and error.

Study task. Refine space and time separately in a heat-equation test. Compare the three time schemes and explain which restriction comes from diffusion stability rather than output frequency.

How to study this book

1. Derive each approximation and its leading error before implementing it; retain a known-solution test.
2. Measure errors across refinements and estimate order where the asymptotic regime is visible.
3. Separate data, roundoff, truncation, iteration, and modeling errors in an experiment log.
4. Compare analytical decay with numerical amplification and justify stability as well as accuracy.
5. When adapting the source’s MATLAB examples to another language, independently check indexing, tolerances, stopping rules, and numerical behavior.

Questions for reflection

• When does a smaller step make an answer worse?
• What distinguishes an unstable algorithm from a sensitive problem?
• How does error in a supporting operation affect a differential-equation solution?

Source edition and coverage

This guide covers all eight chapters of the linked 2023 TU Delft open publication, which identifies earlier editions and a 2016 preface. Starred material is optional relative to the source’s institutional package; the heat-equation chapter explicitly extends beyond ordinary differential equations. These are original study explanations and learning tasks. Read the complete book at the official source.