1: \begin{abstract}

2: The density functional theory (DFT) in electronic structure calculations can be

3: formulated as either a nonlinear eigenvalue or direct minimization problem. The

4: most widely used approach for solving the former  is the so-called

5: self-consistent field (SCF) iteration. A common observation is that  the

6: convergence of SCF is not clear theoretically while approaches with convergence

7: guarantee for solving the latter are often not competitive to SCF numerically.

8: In this paper, we study gradient type methods for solving the direct

9: minimization problem by constructing new iterations along the gradient on the

10: Stiefel manifold. Global convergence (i.e., convergence to a stationary point from any

11:  initial solution) as well as local convergence rate follows

12: from the standard theory for optimization on manifold directly.  A major computational

13: advantage is that the computation of linear eigenvalue problems is no longer needed. The main costs of our approaches arise from the assembling of the total energy functional and its gradient and the projection onto the manifold. These tasks are  cheaper than eigenvalue computation and they are often  more suitable for parallelization as

14: long as the evaluation of the total

15: energy functional and its gradient is efficient.

16: %\comm{This property is very important in simulation for large and complex system.}

17: %\remove{We implement our methods based on two real space software packages, i.e., Octopus  for Kohn-Sham DFT and RealSPACES for Orbital Free DFT, respectively.}

18: Numerical results show that they can

19: outperform SCF consistently on many practically large

20: systems. %\comm{, especially for some large systems}.  %High parallel scalability can also be achieved in our methods.

21: \end{abstract}

22: