[Ipopt] IPOPT CPU Time

[Anil Rao]

 All,

I am using the IPOPT 3.11.8 Matlab mex file from 08-Jun-2014.  Without
getting into too much detail, I am solving an optimal control problem via
direct collocation to an NLP.  The size of the NLP varies due to the fact
that I am performing mesh refinement to obtain a solution that meets a
specified mesh refinement accuracy tolerance.

I am currently performing a study where I am doing computation time
comparisons for different mesh refinement accuracy tolerances and am
finding some strange results. In particular, I am not finding a good
correlation between the CPU time required to solve the problem and the mesh
size or the number of mesh refinement iterations (that is, the number of
times the NLP must be solved).  As a result, it is very difficult for me to
compare different mesh refinement algorithms because a larger size NLP does
not necessarily lead to a larger CPU time.  In fact, in many instances the
CPU time could be much less even though the number of meshes or the size of
the NLP required to meet the accuracy tolerance is much greater.   From
everything I know I am solving a problem where the NLP variables and
constraints are O(1) (because I have scaled the problem appropriately to
make sure that is the case).

I realize that my questions are somewhat vague, but the behavior I am
getting just does not make sense to me.  I am grateful if somebody could
help me figure out how I might arrive at more consistent results by setting
any particular parameters in IPOPT itself.

-- 
Anil V. Rao, PhD
Associate Professor
Department of Mechanical and Aerospace Engineering
University of Florida
Gainesville, FL 32611-6250
Tel:  (352) 672-1529
E-mail:  anilvrao at gmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20150830/756cf521/attachment.html>