YALMIP results in quadprog reporting "the problem is non-convex"

[Noam]

My constraints are a set of linear inequalities and equalities. My objective is a sum of squares of the variables, i.e.,

objective=sum(v(I).^2);

where I is some subset of indices and v is the array of variables. 

As far as I understand, this objective is definitely positive semidefinite, so *any* quadratic problem with such an objective should be convex. It may be infeasible etc., but not non-convex. Does this imply YALMIP is somehow casting the problem into a non-convex form?

[Johan Löfberg]

Impossible to debug without reproducible code.

Probably numerically ill-conditioned, so that MATLAB fails miserably in performing a decomposition of your quadratic function, which YALMIP does to detect convexity (with various tolerances to account for "normal" numerical problems)

Standard trick is to introduce a new set of decision variables z, add the constraint z == v(i), and use the well-conditioned objective z'*z