1: \begin{abstract}

2: A new algorithm for solving large-scale convex optimization problems with a separable objective function is proposed. The basic idea is to combine three

3: techniques: Lagrangian dual decomposition, excessive gap and smoothing.

4: The main advantage of this algorithm is that it dynamically updates the smoothness parameters which leads to numerically robust performance.

5: The convergence of the algorithm is proved under weak conditions imposed on the original problem. The rate of convergence is $O(\frac{1}{k})$, where $k$ is the iteration counter.

6: In the second part of the paper, the algorithm is coupled with a dual scheme to construct a switching variant of the dual decomposition. 

7: We discuss implementation issues and make a theoretical comparison. Numerical examples confirm the theoretical results.

8: 

9: \keywords{Excessive gap \and smoothing technique \and Lagrangian decomposition \and proximal mappings \and large-scale problem \and separable convex optimization \and distributed optimization.}

10: \end{abstract}

11: