Differences between matlab and comsol script?

Question

I find a lot of use for matlab, but my current employer don't have a matlab license: there is one for Comsol, which comes with comsol Script. On a first glance, comsol script is identical to matlab, but after awhile you find eerie differences. Many graphics graphics' features are not there, and you can't export figures to a vector graphics format. But worse is that the performance seems to be much more poor for heavy computations, there is a lot of crashing on trivial algorithms.

(Of course, the purpose of the comsol package is totally different, but I am only interested in using it like matlab, since we don't own a matlab license.)

Are there differences in the matlab and Comsol script cores?

Do they handle memory and matrix computation differently?

Answer 1

Do they handle memory and matrix computation differently?

Almost certainly, yes (at an implementation level). I haven't heard of comsol before, but it's not surprising that it's slower. Mathworks has been putting a lot of effort into performance optimizations over the last few years. Using a JIT has gotten rid of a lot of the interpreter overhead. They also do cache-friendly tricks like executing

 mybigresult = big1 .* big2 + big1 .* big3

as (C-style pseudo-code)

 for (size_t i=0; i<numelement; i++) {
    mybigresult[i] = big1[i] * big2[i] + big1[i] * big3[i];
 }

instead of creating temporaries and scanning memory multiple times. Finally, in the most recent versions, they've added multithreading support to big swaths of their backend. (Note: semantically, Matlab is still single threaded).

Some free Matlab alternatives: octave (aims to be an open source equivalent), python + numpy (very different language, but with similar basic features).

If Matlab makes you much more productive, you might want to try to get your employer to pay for a license... ($50k/year salary * 10% productivity improvement) = $5k worth of extra work done if Matlab is purchased (just making up numbers here).