pointer robust office

Hello,
I'm currently designing a lna for 10GHz band with Microwave Office. I make the gaasfet stable with a serie resistance in the drain and I match the input to be inside the smallest noise circle and the output is matched to 50R.
According the theory this is an optimal design, but if I use the optimiser it find better noise and gain performance with adaptation networks completely differents.

Can someone explain me what's wrong with my design ?

There are multiple combinations for an optimum NF/Gain design.
You have to trust the optimizer.

There is nothing wrong with your LNA design...unless the optimized parameter values are practically realizable(i.e there are process limitations for implimentation)...

Bye the bye which alogrithm you used for the optimazatopn!

Ok thanks for your advice
For information I use "Pointer - Robust Optimization"

Sometime there are strange things with optimiser: it's giving a stub with 1E-10mm lenght so I delete and I get a complete different result !

Anyway I hope to have the pcb next week and I will see...