1: \begin{abstract}

2:   An efficient Krylov subspace algorithm for computing actions

3:   of the $\varphi$ matrix function for large matrices is proposed.

4:   This matrix function is widely used in exponential time integration,

5:   Markov chains and network analysis and many other applications.

6:   Our algorithm is based on a reliable residual based stopping criterion and

7:   a new efficient restarting procedure.

8:   For matrices with numerical range in the stable complex half plane,

9:   we analyze residual convergence and prove

10:   that the restarted method is guaranteed to converge for

11:   any Krylov subspace dimension.

12:   Numerical tests demonstrate efficiency of our approach for solving

13:   large scale evolution problems resulting from discretized in space

14:   time-dependent PDEs, in particular, diffusion and convection--diffusion

15:   problems.

16: \end{abstract}

17: