If the "allowance" A in the support vector classifier is set to exactly zero, then *we use the definition of A as given in the slides (your textbook denotes it as C to increase the confusion). The parameter 'cost' of the svm() function (what we denoted as C) is inversely related to such A. O the classifier would not converge O only a few points will end up within the margin or on the wrong side of the fitted separating hyperplane O we get the maximal margin classifier: no point is allowed to be within the margin of the fitted classifier, let alone on the wrong side of the separating hyperplane O the datapoints can end up within the margin of the fitted classifier, but never on the wrong side of the separating hyperplane