Currently, interpolation operators, P, in root-node smoothed aggregation
are formed through an energy minimization process over the columns of P.
Constraining P to exactly interpolate known near null-space candidate(s)
is a row-wise constraint, which together leads to an expensive, global
minimization process. In this talk, a new method to form P is proposed by
minimizing the 
error in interpolation. When combined with near
null-space constraints, the result is a local minimization process for
each row of P. A theoretical framework and two-level results are
discussed, motivating the next steps in achieving convergence rates of
current root-node at a much lower cost.