Re: [Help-gsl] question on parallelizing rkf45

[Francesco Abbate]

Hi,

I'm not an expert but it seems to me that your problem does not admit
any obvious parallelization of the code. You are suggesting to
dispatch the integration of each parameter to a different CPU but this
cannot be done trivially because the calculation are not independent.
The reason is that, as you mention, being PD equation you have the
spatial derivatives that introduces cross-dependencies between the ODE
equations.

Making a quick search with Google it seems that there is a good method
to parallelize the code using a technique called "domain
decomposition".

Best regards,
Francesco