I am interested in the weight sensitivity of the model due to changes in local thickness variables. The analysis with 50 thickness variables (SOL200_correct.bdf) results in gradients that correlate well with finite difference approximations. The same analysis with 2 additional thickness variables (SOL200_incorrect.bdf) results in incorrect gradients (orders of magnitude larger than expected). I used NASTRAN 2017.1 and previous versions - same results.
Is this a numerical problem? and how can I solve it? Thanks.
