I'm comfortable with general programming, but I'm relatively new to scientific computing. I've started working on research involving simulations, physical-system models, mathematical equations, and algorithms for analyzing or reproducing those systems. I'd like to become much stronger in numerical methods, computational physics, mathematical modeling, simulation, and scientific programming. What textbooks, courses, libraries, or projects would you recommend for moving beyond programming fundamentals and learning the mathematics, algorithms, and practical techniques behind this work?
5 Answers
Numerical methods are probably the best place to focus first. A book such as Numerical Recipes is an accessible introduction, while Numerical Analysis by Burden and Faires is useful for a more structured treatment. Study Taylor expansions and fixed-point iteration because they appear repeatedly in root finding, interpolation, integration, and differential-equation solvers. It’s also worth learning how IEEE-754 floating-point arithmetic works, especially round-off error, conditioning, and numerical stability.
Your professor may be one of the best resources available. Ask whether you can follow an introductory mathematical-modeling or numerical-analysis course, and try to connect each topic directly to your research. Working through the equations by hand, implementing a small version independently, and then comparing it with established libraries is often more instructive than relying on a library from the start.
A useful progression is linear algebra first: LU and QR factorization, direct and iterative solvers, eigenvalues, singular-value decomposition, matrix norms, conditioning, and eventually preconditioning. Then study analysis topics such as Taylor series, continuity, integration, and convergence. After that, learn finite differences and time-stepping methods, including Euler, upwind schemes, and Runge–Kutta methods. Pay particular attention to consistency, stability, and convergence rather than only memorizing formulas.
Build small projects alongside the theory. Implement a root finder, interpolation and quadrature routines, an ODE solver, and a basic PDE solver for a Poisson equation. Verify the results with manufactured solutions and measure the observed convergence order. Keeping a record of failed approaches, assumptions, and numerical experiments is valuable because it helps you recognize instability and avoid repeating mistakes.
Once the fundamentals are comfortable, finite-element and finite-volume methods are good next steps, especially if your research involves physical systems or fluid dynamics. Learn enough mesh generation to use a tool such as Gmsh, then explore simple Lagrange finite elements before moving on to more advanced schemes. For the programming side, investigate memory layout, cache behavior, multithreading, MPI, GPUs, and parallel linear algebra. High-performance computing is an especially useful specialization for someone coming from a general computer-science background.
You usually don’t need to master every detail of floating-point arithmetic immediately, but you should understand that calculations have limited precision and that equality comparisons and tolerances need care. Avoiding unsafe compiler optimizations and testing tolerances empirically will cover many practical cases at first.