For 1/(1 - v) < p [less than or equal to] [infinity], the operator [mathematical expression not reproducible] is compact, and since by the assumption (A3), the nullspace of I - T is trivial in C[0,1] implying that the null space of I - [??] is trivial in [??](R), the bounded inverse [mathematical expression not reproducible] exists.

Where [[omega].sub.k] presents the precoding unit-norm beamforming vector for user k is chosen in the direction of the projection of [h.sub.k] on the nullspace of [h.sub.j], j[not equal to]k.

We write N(T) and R(T) for the nullspace and range of an operator T [member of] B(H).

Thus for each eigenvalue pair ([gamma], 1/[gamma]) of [C.sup.T] and -K*, respectively, the coordinates of the eigenvectors of [C.sup.T] associated to [gamma] determine which linear combinations of the coordinate functions of y must satisfy a Fredholm-like orthogonality condition with elements of the nullspace of the resolvent 1/[gamma] + K*.

By [sigma](A), R(A), D(A), and N we denote the spectrum, range, domain, and nullspace of A, respectively.

(ii) from Corollary 12, [D.sup.-1]p is in the nullspace of the Laplacian Lap and [D.sup.-1] d = j;

Observe that the columns of the matrix U associated with null eigenvalues form a basis of the nullspace of A and that means that each individual (each column of A), if expressed in the base formed by the columns of U, will have at most [r.sub.A] nonnull coefficients; that is, we will not find more than [r.sub.A] independent elements in the Pareto front.

Because H and E have the same continuous condition on the tangential components, H will not be compatible with the nullspace of the curl operator if H and E also have the same order.

The first Q columns and the last N - Q columns of V span the row space and nullspace of H, respectively.