In this talk, we discuss multigrid methods that employ Hermitian/skew-Hermitian splitting iteration as smoothers. We also introduce modified additive and multiplicative smoothers derived from subspace decomposition. Without relying on the full elliptic regularity assumption, we show that these multigrid methods achieve robust convergence for second-order nonselfadjoint elliptic boundary value problems—provided the coarsest grid is sufficiently fine, though independent of the number of multigrid levels.