Argh, I can't believe I actually made this mistake - this is the umptiest minimum distance estimator I'm writing using NLopt, and I always ensure that a scalar instead of a vector is returned by doing exactly what you're suggesting, but for whatever reason I simply missed it this time around. Thanks for pointing it out!
(And yes, there's a lot more upper and lower bounds, tolerance levels and, indeed, an objective to specify, I just left this out for brevity)
