I guess feasOpt trumps iis. Just leave feasOpt out of the option file. If your model is "just" quadratic, convince yourself that you can't formulate in a convex way. Moreover, if you have the quadratic terms only in the objective, you can change the "optimalityTarget" (see https://www.gams.com/latest/docs/S_CPLE ... litytarget) and even work with non-convex quadratic objective functions (with Cplex). You might also try a global solver like BARON or GloMIQO. If you need pwl, then the answer to your question is yes. BTW, don't use **2 use sqr(). **2 is real power 2 and very different from sqr().
-Michael