Observe the behaviour of perceptron learning algorithm, for two classes which are (a) linearly separable and (b) linearly inseparable. What is the number of iterations required in each case ?

How does the perceptron behave if the two classes are not linearly separable. Assuming that the two classes are overlapping, do the weights converge, or do they oscillate around an optimum value ? In either case, what is the nature of the separating line/hyperplane.