Function 'quadprogIP' runs fast seperately but extremely slow in a for loop simulation

I am currently using the function 'quadprogIP' from https://github.com/xiawei918/quadprogIP

This is a function that solves a nonconvex quadratic optimization, via IBM Cplex function for mixed-integer linear programming. My problem is: if I run the function 'quadprogIP' once, it executes very fast (around 10 minute).

clear
n = 50; 
f = zeros([n,1]); 
A = zeros(n); b = zeros([n,1]); 
Aeq = ones([1,n]); beq = 1; 
LB = zeros([n,1]); UB = ones([n,1]); 
epsilon = 1e-6;

sigma = 2*rand(1,n);
rho = rand(n); 
Cov = sigma.'*sigma.*rho; 
rhoc = ones(n);
Covc = sigma.'*sigma.*rhoc; 

low = 0; high = 1; iter = 1; 
q = (low+high)/2; H = Cov-q*Covc; 
[w,Fq,time,stats] = quadprogIP(H,f,A,b,Aeq,beq,LB,UB,[]); 
TT(iter) = stats.total_time  % This is normally 10 seconds

But if I wanna run a while loop, and the function 'quadprogIP' is called upon in the loop, the average time for each execution is much longer (at least 5 times higher).

while abs(Fq) > epsilon 
     if Fq>0 
        low = q; 
     else
        high = q;
     end
     iter = iter+1; 
     q = (low+high)/2;
     H = Cov-q*Covc;
    [w,Fq,time,stats] = quadprogIP(H,f,A,b,Aeq,beq,LB,UB,[]);  % from Github
    TT(iter) = stats.total_time
end    
mean(TT)  % But this one is a lot higher. 

I don't know what could be the possible reason for this situation, and if there is any way to solve this problem. Thanks!

Answer 1

[Torsten]

Maybe the problems to be solved get larger or more complicated within the for loop so that the average time spent for each problem increases ? But we can only speculate without your code.

Since "quadprogIP" is not an official MATLAB tool, I'd suggest to contact the authors.

[Churui Li]

Thank you for answering! I added the relevant code in the post.

Indeed it is not an official Matlab function, I may turn to the authors if the problem remains after a while.