24小时热门版块排行榜

返回列表

【奖励】本帖被评价1次，作者xirainbow增加金币 1 个

当前主题已经存档。

xirainbow

木虫 (正式写手)

[资源] 【分享】SCF is slow or does not converge

   以下内容来自我的博客：http://blog.sina.com.cn/s/blog_5f15ead20100cw4d.html

   Yesterday, I try to optimize graphene ribbon by using PWscf. Unfortunately, at the beginning, I failed. When I went over the .out file, I found the reason: scf does not converge within after 100 iterations. This is the first time I met this kind of failure. In the end, I add nbnd from 8 to 12, then scf can converge within 30 iterations.

I find some useful tip to deal with scf converge failure. I got the following content in "User’s Guide for Quantum-ESPRESSO(version 3.2)"

     "Reduce mixing beta from the default value (0.7) to ∼ 0.3 − 0.1 or smaller. Try the mixing mode value that is more appropriate for your problem. For slab geometries used in surface problems or for elongated cells, mixing mode=’local-TF’ should be the better choice, dampening “charge sloshing”. You may also try to increase mixing ndim to more than 8 (default value). Beware: the larger mixing ndim, the larger the amount of memory you need.
   If the above doesn’t help: verify if your system is metallic or is close to a metallic state, especially if you have few k-points. If the highest occupied and lowest unoccupied state(s) keep exchanging place during self-consistency, forget about reaching convergence. A typical sign of such behavior is that the self-consistency error goes down, down, down, than all of a sudden up again, and so on. Usually one can solve the problem by adding a few empty
bands and a broadening.

      Speciﬁc to US PP: the presence of negative charge density regions due to either the pseudization procedure of the augmentation part or to truncation at ﬁnite cutoff may give convergence problems. Raising the ecutrho cutoff for charge density will usually help, especially in gradient-corrected calculations."
In brief, there are 6 kinds of tips.

1: reduce mixing beta from default value (0.7) to 0.3 to 0.1 to smaller. I think wavefunction(3)=mixing_bata*wavefunction(2)+(1-mixing_bata)wavefunction(1). Wavefunction(1) is the first guess_wavefunction,Wavefunction(2) is the calculated wavefunction,Wavefunction(3) is the next iteration guess_wavefunction. The smaller mixing_beta is, the more similar following guess_wavefunction Wavefunction(3) is with the first guess_wavefunction Wavefunction(1).

   2: change mixing mode. For slab condition, local-TF is first option.

   3: change mixing ndim from default value (8) to bigger value at the cost of larger memory.

   4: adding empty bands and a broadening for matallic or close to metallic state. A typical sign of metalllic state is that the scf error goes down, down, down, then all of a sudden up again. This is just what I met.

   5: raising the ecutrho cutoff for charge desity for US PP condition.

I find the tips about scf convergence in slab calculation. Some of them is universal.

      In a long slab, sometimes it is harder for the runs to converge. There are a number of steps you can take. First, make sure all of the output files from the previous run are deleted (filewfc, etc..) between each run. Second, try lowering the beta parameter. The beta parameter sets the charge mixing between different iterations. Third set the “niter” parameter to a higher number (it defaults to 50). So a line such as “niter=200”. Fourth, check the convergence of the surface energy (do the actual calculation for the full surface energy, don’t just look at energies of slabs). Do you really need that big of a vacuum size or that long of a cell? The bigger your vacuum and slab, the more problems you will have with convergence.

[ Last edited by xirainbow on 2009-7-18 at 09:53 ]

回复此楼