e308d1edf430ad5b.tex
1: \begin{abstract}
2: 	To construct a parallel approach for solving optimization problems with orthogonality constraints is 
3: 	usually regarded as an extremely difficult mission, due to the low scalability of the 
4: 	orthonormalization procedure. 
5: 	However, such demand is particularly huge in some application areas such as materials computation.
6: 	In this paper, we propose a proximal linearized augmented Lagrangian algorithm (PLAM)
7: 	for solving optimization problems with orthogonality constraints. 
8: 	Unlike the classical augmented Lagrangian methods, 
9: 	in our algorithm, the prime variables are updated by minimizing a proximal 
10: 	linearized approximation of the augmented Lagrangian function, 
11: 	meanwhile the 
12: 	dual variables are updated by a closed-form expression which holds at any first-order 
13: 	stationary point.
14: 	The orthonormalization procedure is only invoked once 
15: 	at the last step of the above mentioned algorithm if high-precision feasibility is needed.
16: 	Consequently, the main parts of the proposed algorithm can be parallelized naturally. 
17: 	We establish global subsequence convergence, worst-case complexity 
18: 	and local convergence rate for PLAM under some mild assumptions. 
19: 	To reduce the sensitivity of the penalty parameter, we put forward a modification
20: 	of PLAM, which is called parallelizable column-wise block minimization of PLAM (PCAL).
21: 	Numerical experiments in serial
22: 	illustrate that 
23: 	the novel updating rule for the Lagrangian multipliers significantly accelerates 
24: 	the convergence of PLAM and makes it comparable with the existent feasible solvers for 
25: 	optimization problems with orthogonality constraints, and the performance of PCAL
26: 	does not highly rely on the choice of the penalty parameter.
27: 	Numerical experiments under parallel environment demonstrate that
28: 	PCAL attains good performance and high scalability in solving discretized Kohn-Sham total energy 
29: 	minimization problems.
30: \end{abstract}
31: