1: \begin{abstract}

2: We study a semismooth Newton-type method for the 

3: nearest doubly stochastic matrix problem where both differentiability

4: and nonsingularity of the Jacobian can fail.

5: The optimality conditions for this problem are

6: formulated as a system of strongly semismooth functions. 

7: We show that the so-called local error bound condition does not hold for

8: this system. Thus the guaranteed convergence rate of Newton-type methods is at most superlinear. By exploiting the problem structure, we construct a 

9: modified two step semismooth Newton method that guarantees a nonsingular

10: Jacobian matrix at each iteration, and that converges to the nearest doubly stochastic matrix quadratically. To the best of our knowledge, this is 

11: the first Newton-type method which converges $Q$-quadratically in the absence of the local error bound condition.

12: \end{abstract}

13: