Re: C library faster than octave

[Lukas Geyer <lukas@debian.org>]

Halim Boukaram <hwboukaram@yahoo.com> writes:

> I've tested the differential equation solver for
> Ocatve (lsode) vs my own with these results:
> 
> executing an ode solver 1000 times on my P4 1.8:
> 
> 22 seconds for Ocatve (i think euler method)
> 8 seconds for my library's euler method
> 12 seconds for my library's runge kutta order 2
> 22 seconds for my library's runge kutta order 4
> 
> so my estimation is that my library is at least twice
> as fast as octave which makes it at least as fast as
> Matlab (which is also twice as fast as Octave). I'll
> also tell the Octave team maybe they will want to link
> up.

Of course, octave does not use the plain Euler method. Nobody in their
right mind would do that. The octave doc says the ODE solvers are
"based on reliable ODE solvers written in Fortran", so they are
probably both more advanced than even plain Runge-Kutta. You should
test both accuracy and speed. As a test case, I suggest you try the
initial value problem

x' = 1+x^2, x(0)=0

and solve for x(1), x(1.5), x(1.57). Of course the exact solution is
x(t)=tan(t), thus the exact values would be 1.5574, 14.101 and
1255.8.

Best regards,
Lukas