my  block

where uzairs roam

Damn. I was reaching for my Polo (’the mint with a…’) just now, and my copy of Numerical Computations slipped off the desk and crashed to the floor. And that got me thinking…

The program I’m doing right now at LSE, the MSc in Applicable Mathematics, is pretty good. There’s clearly lots of optimisation theory and game theory going on, but they’ve crammed some probability/statistics and some computation in as well. And although the upper-limit on the number of courses you can take means it’s a bit of a trick, you can still squeeze in the finance courses you want. (You’re welcome to sit in the lectures unofficially, of course. I don’t think it’s a useful alternative/compromise, and have said as much to the department.) Really, it’s enough to keep you busy, with calculus, functional analysis, learning theory and algorithms showing up at some point or another.

Yup, not much lacking. Except for some hard-core numerical analysis and/or numerical computation. Solving differential equations algorithmically, constructing splines, fast Fourier transforms, etc. The closest thing the MSc offers is the Algorithms & Computation course, which spends one half of the year teaching people Java, and the other half discussing binary trees and Djikstra’s algorithm. Sorry, but there’s a much clearer need for numerical computation in mathematics than binary trees. Oh, and Java isn’t exactly the language of choice for professional mathematicians, is it?

No, I think the department got this badly wrong. What they should have done was make C or C++ the language taught, and form exercises that tackle numerical problems. I do recognise C/C++ is trickier to learn, but perhaps the wiser course would have been to offer parallel courses, one for people who have no programming experience, and the other for people who have programmed before. The only remaining argument, that numerical problems are harder to tackle than binary trees, doesn’t really hold water: Gaussian elimination for solving simultaneous equations literally takes just 50-100 lines to code. Let’s hope this is reconsidered as the program continues to evolve.

The book, BTW, is excellent. I saw it the other day at the library and thought it would be a good way to scrape the rust off (since I haven’t really programmed in several months now), and guess what — it is! Most of the examples are in Fortran, so I’m translating them to C++ for practice :)