Large scale problem fail using ADOL-C for both derivatives and Hessian

[opc.fa...@gmail.com]

Dear all,

I'm using PSOPT 4.01 on Ubuntu 18.04. The main settings I used in my own problem formulation are:

         problem.phases(1).nodes = 5001;

    algorithm.nlp_method                  = "IPOPT";
    algorithm.scaling                     = "automatic";
    algorithm.derivatives                 = "automatic";
    algorithm.hessian                     = "exact";
    algorithm.collocation_method          = "trapezoidal";

In my problem, I have 3 states and 1 control, with fixed terminal time. My intention was to test whether PSOPT can be used for large scale optimal control problems with automatic differentiation on both derivatives and hessian. However, when running my code, on screen after it was shown that ADOLC detects the sparsity of Jacobian, the program failed, showing 'segmentation fail (core dumped)'.

I have seen posts in this user group long ago that ADOLC fail for large scale problems. May I ask is this problem still there in the newest version release? In PSOPT manual, it suggests user to use numerical approach to evaluate Hessian. However, it is my intention to use exact or automatic differentiation for Hessian.

Is there any quick solution to fix this problem in PSOPT and/or ADOLC? Or other automatic differentiation tools have to be used? Is there any tool to generate derivatives and Hessian equations explicitly so that they can be used by PSOPT and IPOPT?

Thank you.