Tobias Weinzierl - Principles of Parallel Scientific Computing: A First Guide to Numerical Concepts and Programming Methods (Undergraduate Topics in Computer Science)

I started to think about this manuscript when I got appointed as Associate Professor at Durham University and had been told to read numerical algorithms for computer scientists plus the parallel programming submodule. In Durham, we teach this first submodule on numerics in the third year of the computer science degree. It spans only ten hours. Ten hours is ridiculously short, but I think it is enough to spark interest in simulation codes and the simulation business. Thinking about it, it is great that our numerics teaching runs parallel to a parallel programming course. Numerical simulation without parallelism today is barely imaginable, so this arrangement gives us the great opportunity to use examples from one area in the other one. We can co-teach the two things: numerics basics and parallel coding.

When we later designed our Master in Scientific Computing and Data Analysis (MISCADA), it became clear that co-teaching the two fields would be a good idea for this course, too. With a very heterogeneous student intakeboth by means of qualification, cultural background, educational system, and discipline interestwe need to run introductory courses that bring all students quickly to a decent level, while numerical principles as well as parallel programming are omnipresent. Obviously, the M.Sc.-level teaching progresses at a different pace and, in Durham, also covers more lectures. Yet, some content overlaps with the third years course.

There seems to be a limited amount of literature that discusses parallel programming and (the basics of) numerical techniques in one go and still remains easily accessible. There are many books that cover scientific computing from a mathematics point of view. Many scientists (including me) insist that a formal approach with a strong mathematical background is imperative for research in scientific computing. However, we have to appreciate that many of our undergrads are enrolled in courses that are lighter on the mathematics side or focus, for example, more on discrete maths. Some maths thus might be a little bit rusty when students enter their third year. Our mathematical textbooks run the risk that we lose out on students if they orbit around the profound maths first. They might get bored before they reach the point where they can program a first numerical code that does something useful.

The other type of teaching literature that is out there combines the first course in numerics with a programming tutorial. Although a lot of the material I have seen is really good, this type of approach is ill-suited for our students. Most of them are decent programmers already.

To make a long story short, I think that there is a need for courses on numerics with a rolling up your sleeves and get the rough idea and make it work in code attitude. Some students like to get their hands dirty and write something that does something before they start to dive into more detailsfor which excellent literature does exist. I am also strongly convinced that it makes sense to teach introductory scientific computing in combination with parallelisation. After all, we want to have future Ph.D. students and colleagues who can help us with the big simulation codes.