In the optimization of Support Vector Machines, what set of necessary and sufficient conditions must be met for a solution to be optimal?

KKT conditions

The Karush-Kuhn-Tucker (KKT) conditions generalize the method of Lagrange multipliers to include inequality constraints. For SVMs, they ensure that the dual gap is zero and define the properties of the support vectors.