Numerical Methods in Scientific Computing

Numerical Methods in Scientific Computing by Jos van Kan, Guus Segal, and Fred Vermolen develops numerical treatment of differential-equation models. Thirteen chapters connect conservation, finite differences, finite volumes, variational methods, finite elements, algebraic solvers, diffusion, waves, transport, and moving boundaries. This guide emphasizes derivation, assumptions, verification, and the distinction between accuracy and stability.

Chapter-wise summaries and study tasks

Chapter 1 Review of some basic mathematical concepts

The opening chapter connects mathematical tools with physical modeling. Gradients, directional derivatives, divergence, and the divergence theorem support the translation between local behavior and balances over a region. Conservation laws provide a common structure for many later problems. Minimization offers another route to a model through a quantity whose variation characterizes equilibrium. Linear algebra and analytical results prepare the language used for discretization and solution. The aim is not an isolated formula review: identify which operation represents which physical relationship and the assumptions needed when changing from an integral statement to a differential one.

Study task. Derive a fictional conservation balance over a region, identify the flux and source units, and explain the role of the divergence theorem and any smoothness assumptions.

Chapter 2 A crash course in PDE’s

Partial differential equations require a classification and conditions that make a problem meaningful. Elliptic, parabolic, and hyperbolic forms capture different structures, while boundary and initial conditions supply information not contained in the differential expression alone. Existence and uniqueness are questions about the complete problem, not just the equation's name. Potential flow, convection-diffusion, fluid equations, stress, and biharmonic examples connect the mathematical categories with applications. A numerical solution cannot repair incompatible or insufficient conditions. Before choosing a scheme, state domain, unknown, coefficients, forcing, and the conditions through which the physical setting enters.

Study task. Compare two PDE problems using classification, required data, and expected behavior. Explain how changing a boundary condition can change uniqueness or the meaning of the solution.

Chapter 3 Finite difference methods

Finite differences approximate derivatives on a grid, producing algebraic equations that inherit or distort properties of the continuous problem. Cable and diffusion examples develop discretization, matrix structure, and error. Multidimensional Laplacian problems introduce boundary treatment, nonrectangular geometry, and coordinate choices. A small local truncation error does not alone establish a small global solution error; stability and the inverse problem also matter. Discrete maximum principles and comparison arguments support stronger analysis. Check how boundaries affect order and behavior rather than deriving an interior stencil and assuming the complete implementation has identical properties.

Study task. Derive a simple difference stencil, implement boundary conditions, and compare errors on refined grids against a known solution. Explain what the refinement evidence does and does not prove.

Chapter 4 Finite volume methods

Finite-volume methods begin with balances over control volumes and approximations to fluxes across their boundaries. This structure makes conservation explicit when shared face fluxes are represented consistently. Variable coefficients, boundary cells, general coordinates, and staggered arrangements introduce choices that affect accuracy and physical meaning. The chapter connects scalar diffusion with more involved flow settings, including a Stokes project. A locally conservative computation can still be inaccurate if flux approximation or geometry is unsuitable. Trace signs, normals, areas, volumes, and units through the derivation before relying on a matrix generated by code.

Study task. Construct a small fictional finite-volume diffusion problem and show cancellation of internal face fluxes in the global balance. Test how a boundary approximation affects conservation and accuracy.

Chapter 5 Minimization problems in physics

Minimization principles connect selected physical problems with an energy or functional. Variational reasoning can produce a differential equation and natural boundary conditions under appropriate assumptions. Examples include surfaces, elasticity, and plates, with different derivative and regularity requirements. A stationary point is not automatically a minimum, so mathematical conditions supporting the interpretation matter. The chapter also examines the reverse route from a PDE to a minimization problem where justified. Distinguish essential conditions imposed on admissible functions from conditions arising through the variational statement. These foundations explain why later approximation spaces and assembly rules take their particular form.

Study task. Derive a simple one-dimensional variational problem, identify admissible variations and boundary terms, and state what supports interpreting the stationary solution as a minimum.

Chapter 6 The numerical solution of minimization problems

Ritz's method approximates a minimization problem in a finite-dimensional space. Finite elements make that approach practical through local basis functions, element matrices and vectors, numerical integration, and assembly. One- and two-dimensional examples show how local contributions form a global system. Boundary conditions and periodic constraints must be incorporated consistently, not added after solving. Approximation quality depends on the space, mesh, quadrature, and regularity rather than the number of unknowns alone. Convergence theory explains the relationship between a discrete solution and the intended continuous problem under specified assumptions.

Study task. Assemble a small one-dimensional finite-element system by hand, including a boundary condition. Compare two meshes and separate approximation error, integration choices, and algebraic-solver error.

Chapter 7 The weak formulation and Galerkin’s method

A weak formulation uses test functions and integration to express a problem in a form suitable for approximation. Galerkin methods choose a finite-dimensional test and trial framework; symmetry is useful but not universal. Convection-diffusion examples show why standard choices can struggle when transport dominates diffusion. Petrov-Galerkin and streamline-upwind approaches alter testing to address behavior that an unsuitable discretization may represent poorly. Stabilization introduces its own consistency and accuracy questions. Keep essential and natural boundary conditions distinct and explain why a chosen weak form represents the original problem before evaluating its discrete implementation.

Study task. Derive a weak convection-diffusion formulation and compare standard Galerkin with a stated stabilized alternative. Explain oscillations, parameter choices, and the evidence needed to judge improvement.

Chapter 8 Extension of the FEM

Extensions of the finite-element method change element shape, polynomial order, geometry representation, or the variables being approximated. Quadratic triangles, quadrilaterals, and curved elements require appropriate mapping and integration. Higher order is not automatically better when geometry, regularity, or quadrature is inadequate. Stokes equations, circular symmetry, and fourth-order problems introduce additional structure and compatibility requirements. Mixed approaches can replace a difficult formulation with coupled variables, but their stability must be considered. Evaluate the complete approximation rather than compare element names as if they guaranteed an accuracy ranking.

Study task. Compare linear and higher-order approximations on a suitable fictional problem. Identify geometry mapping, quadrature, boundary, and stability requirements, then explain when a mixed formulation changes the unknowns.

Chapter 9 Solution of large systems of equations

Discretization creates algebraic systems whose efficient solution depends on sparsity, ordering, conditioning, and matrix properties. Direct methods, band or profile approaches, and renumbering address storage and elimination cost. Iterative methods use repeated correction, with convergence determined by the operator and algorithm rather than iteration count alone. Classical schemes, Krylov methods, preconditioning, and multigrid exploit different structures. Conjugate-gradient assumptions should not be ignored for a nonsymmetric system. Nonlinear problems add Picard or Newton iterations and sensitivity to starting values. Distinguish residual reduction from actual error reduction and connect stopping criteria with discretization accuracy.

Study task. Compare two solvers on a small discretized problem, recording residuals, cost, and assumptions. Explain why a small residual may coexist with substantial error and why a preconditioner changes performance.

Chapter 10 The heat- or diffusion equation

The heat or diffusion equation links spatial discretization with time integration. The method of lines produces a system of ordinary differential equations, whose eigenstructure affects stability restrictions. Consistency, stability, and accuracy concern different aspects of the approximation. Von Neumann analysis and matrix-based bounds provide tools under their respective conditions. Explicit and implicit methods trade computational work against stability and error properties; unconditional stability does not mean unlimited accuracy at large time steps. Alternating-direction ideas exploit multidimensional structure. Verify both temporal and spatial refinement rather than allowing one error source to hide the other.

Study task. Run a fictional diffusion example at several mesh sizes and time steps. Test a predicted stability restriction and separate temporal from spatial error using an analytical or manufactured solution.

Chapter 11 The wave equation

Wave problems require attention to propagation, energy behavior, and the quality of phase information. Spatial discretization and time integration can introduce numerical dissipation and dispersion that change amplitude or speed even where a solution remains stable. Direct integration of a second-order system offers different design choices from converting it to first order. The CFL criterion connects information propagation with the numerical grid under specified conditions. A visually smooth trajectory may still have an incorrect arrival time or frequency. Evaluate the feature relevant to the application rather than accepting stability as a complete measure of quality.

Study task. Compare numerical propagation of a fictional pulse using two settings. Measure amplitude and arrival-time error, relate results to dissipation and dispersion, and state the applicable CFL assumptions.

Chapter 12 The transport equation

Transport equations move information along characteristics and can develop behavior that a central approximation handles poorly. Upwind and other schemes make trade-offs among stability, diffusion, accuracy, and treatment of steep gradients. Nonlinear examples such as Burgers and Buckley-Leverett connect conservation with shocks and admissible solutions. A numerical oscillation near a front and excessive smearing are different errors. Flux limiters address selected difficulties but require stated conditions and careful interpretation. Assess conservation and the behavior of discontinuities, not only pointwise agreement where the solution is smooth.

Study task. Simulate a fictional transported profile with central and upwind approaches under suitable settings. Compare front position, smearing, oscillation, and conservation, and explain why a shock needs additional interpretation.

Chapter 13 Moving boundary problems

Moving-boundary problems require determining an interface as part of the solution. The Stefan example couples heat behavior with a condition governing phase-boundary motion. A self-similar solution provides insight and a benchmark under its assumptions. Moving-grid methods follow geometry explicitly, while fixed-domain or level-set approaches represent the interface differently. Each introduces requirements for coupling, initialization, conservation, and resolution. An accurate temperature field does not automatically establish an accurate interface position. Evaluate both quantities and explain what physical balance drives motion rather than prescribing a boundary trajectory that the model is meant to predict.

Study task. Outline a fictional Stefan calculation and compare moving-grid and level-set representations. Identify interface and field error measures and use a source-compatible benchmark to plan verification.

How to study this book

1. Review calculus, vector calculus, differential equations, and linear algebra before attempting the numerical derivations.
2. State the complete continuous problem, including domain, units, coefficients, and boundary or initial conditions.
3. Derive a small discrete example by hand before scaling the implementation.
4. Use analytical or manufactured solutions, conservation checks, and separate spatial and temporal refinement studies.
5. Distinguish modeling, discretization, algebraic, and roundoff errors, and choose stopping tolerances accordingly.

Questions for reflection

• Which property of the continuous problem should the discretization preserve?
• What assumptions support the claimed convergence or stability result?
• Is a discrepancy caused by the mesh, time step, solver, or model?
• Does the reported error measure capture the feature important to the application?

Source edition and coverage

The guide follows the thirteen main chapters of the linked TU Delft source. It includes the final moving-boundary chapter and the finite-element extension chapter, rather than replacing them with a generic numerical-analysis outline. Software implementation and theoretical claims must be checked against the conditions developed in the textbook. These are original study explanations and learning tasks. Read the complete book at the official source.