Embed cannot guarantee finding a global optimum in cases where there are distinct local optima. If you know that your problem is convex (that is, minimizing a convex objective over a convex feasible region), then any local optimum is global, so this problem cannot occur. Otherwise, based on your knowledge of the problem, you must try a variety of starting points to determine the local/global issue. If all starting points yield approximately the same final point, and that point satisfies what you know about the problem, then you can be fairly confident that the point is globally optimal.