Algorithm Leadership
I recently came across an article on re-evaluating classic algorithms in the modern age. The article makes a strong point: the relative cost of computer-vs-programmer time has dramatically dropped in recent years. It isn’t capitally efficient to hand tune certain classes of algorithms anymore. It’s a good read and highly recommended.
On a similar note, I’m having doubts about a PDE solver that I’m writing for my course project. I’m implementing the Pseudo-Spectral Method for the space dynamics. The technique is quite simply a projection onto another basis that requires fewer coefficients in the series expansion. At each time step, I take the forward and backward Fourier transform of the space variables. This might have made sense a few years back when the algorithmic complexity of the FFT beats a space convolution hands down, but the added code complexity is probably not worth it. The only reason I’m doing it now is because I’ve never done anything like this before and it’s pretty cool.