The paper shows that (L+1)-level linear programs are as difficult as level L of the polynomial-time hierarchy, even if one only considers problems with unique optimal solutions.
