fzero system of non linear equations

Hi

I am trying to solve a system of non linear equations, which is embedded in a function run by a ODE solver. At the current point I am doing it by a for loop, however the computation time is really high (around real time of 10+ hour dynamic simulations), so I would like to bring it down. I've found this step to be very time consuming

For the current simulations par.NT = 13, size(ics) = [13,3], size(xn) = [13,2], size(par.Tb) = [1,2] and I do initialize the variables TB and y. yfun is a function which is called (which calls various other functions)

 for n = 1:par.NT     
   TB(n)=fzero(@(T) sum(yfun(ics,T,par,n))-1,sum(xn(n,:).*par.Tb));
   y(n,:) = yfun(ics,TB(n),par,n);
 end

Do you have any tips on making this faster? I have been trying to vectorize the loop in various ways, use arrayfun, fsolve... but I can't seem to increase the speed. I do submit my code to HPC clusters. Using parfor decreases the speed drastically.

Answer 1

[Prashant Arora]

Hi Bastian,

Which part of the for loop is more expensive? Have you tried profiling the yfun function? I believe it takes a structure argument, which can sometimes increase computation time to get fields.