Have you tried nondimensionalizing your equations, or centering/scaling
your variables to make everything close to unit magnitude? This could
help the solver find a solution with your needed precision. Otherwise
have you tried different algorithms?
Patrick
On 07/21/2013 08:48 PM, Jingchen Feng wrote:
Hi,
I am using gsl to do optimization. The problem is I want the precision to be
high. When I set the stop criteria to be:
status= gsl_multimin_test_gradient(s->gradient,1e-8);
I always get an error code, which tells the value of function can't be improved
any more. After debugging for a whole day, I think the problem is the type of
function return value is 'double', which limits the precision. However, It
seems that gsl does not allow a long double return function to be optimized. Is
there any way to solve this problem?
Thanks!
