1: \begin{abstract}

2: We propose a semismooth Newton algorithm for pathwise

3: optimization (SNAP) for the LASSO and Enet in sparse,

4: high-dimensional linear

5: regression. % $\y = X\bbp + \etaa$.

6: SNAP is derived from a suitable formulation of

7: the KKT conditions based on Newton derivatives.

8: It solves the semismooth KKT equations efficiently

9: by actively and continuously seeking the support of the regression coefficients along the solution path with warm start.

10: At each knot in the path,  SNAP converges  {%\color{red}

11: locally superlinearly for the Enet criterion and achieves an optimal  local convergence rate for the LASSO criterion, i.e., SNAP converges in one step}  at  the  cost  of two matrix-vector multiplication per iteration. Under certain regularity conditions

12: on the design matrix %$X$

13: and the minimum magnitude of the nonzero elements of the target regression coefficients, %$\bbp$,

14: we show that SNAP hits a solution with the same signs as  the regression coefficients %$\bbeta^{\dag}$

15: and achieves a sharp estimation  error bound in finite steps with high probability.

16: The computational complexity of SNAP is shown to be the same as that of LARS and coordinate descent algorithms per iteration. Simulation studies and real data analysis support our theoretical results and demonstrate that SNAP is { faster and accurate  than

17: %the existing methods such as

18: LARS and coordinate descent algorithms}.

19: 

20: %Simulation studies support our theoretical results and

21: %indicate that SNAP is competitive with or outperforms state-of-the-art LASSO solvers  in %terms of efficiency and accuracy. We have implemented SNAP in a Matlab package %\textsl{snap},  which is available at {\color{red} your homepage}

22: \end{abstract}