Just to test I ran you model using the "Evolutionary" solver engine. No problem getting Solver to run but result changed from 266,145409 to 266,1448396 so hardly any improvement.

I also tested the Non-linear version of OpenSolver (freebie from Univercity of Aukland NZ) it's their OpenSolver 2.9.0 AdvancedWin with a bit better result

OpenSolver_Coin.jpg

Still I don't know why the GRG Nonlinear can't handle this problem.

Alf