but instead, consider optimizing \(\mathbf{u} \cdot \mathbf{w}\) via Lagrange multipliers under constraints.