| 12:25:53 PM PDT - Wed, Oct 21st 2015 |
|
| I have noticed that when doing a geometry optimization with 'task pspw optimize', near the convergence the i/o time as a percentage increases significantly. For example, When something like driver; trust 0.5; maxiter 10; end is used, there may be only one or two sub-iteration or so for each step, and the movecs file is written and read for each. For large systems this dominates the calculation time. I see that in the dft module there is a 'noio' keyword; is there a similar option, where by perhaps the vectors can be kept in memory between iterations, and only written on the final iteration, or upon convergence?
|