24小时热门版块排行榜    

查看: 759  |  回复: 2

cenwanglai

荣誉版主 (知名作家)

老和山猥琐派九段

优秀版主

[求助] CNEB收敛到0.05后,提高精度不收敛

(1)q-m算法收敛到1eV/A
  (2)  LBFGS算法熟练到0.05
(3)按照(2)的设置,修改EDIFFG=-0.03,很难收敛。

下面一段话是早前Graeme给某个帖子的回复。看他的建议,似乎是 q-m算法收敛到1,然后CG或者LBFGS熟练到0.05,这时EDIFFG=1E-4或者1E-5;如果要达到更高的精度,需要更高的力的精度,要求EDIFFG = 1E-5或者更低。这时可以退回q-m算法。

及CNEB的熟练过程的算法切换是:Q-M (<1 eV/A)->CG/LBFGS (<0.05 eV/A)->Q-M (<0.03 eV/A). 这样的话,如果(有人报道过这样的结果)力精度提高到0.001,EDIFFG岂不是要设置的很小很小?

Quickmin is a first-order optimizer which is relatively stable at any point during optimization. Near the minimum where the potential can be expanded using a harmonic approximation, it is not as efficient as second-order optimizers such as CG or LBFGS.

There are a few caveats. CG and LBFGS are not particularly stable far from the minimum where forces are high. So using quick-min until the forces drop below 1 eV/Ang is a sensible way to start. You can also make sure that the path at this point makes physical sense. Then it will be more efficient to switch to a second order method.

When I talk about forces being accurate enough near the minimum, I'm referring to the errors in the forces which are controlled by the electronic convergence criteria, ediff. When the forces are larger than the errors, curvatures can be determined with sufficient accuracy. Near the minimum, the errors are a higher portion of the force. At this point, you can either go back to quick-min (or fire), or decrease ediff. Your value of ediff=1e-8 is extremely tight. Using a value of 1e-5 will generally speed up your calculations without causing problems unless you are trying to get forces down to ~0.05 eV/Ang.

Also, as a side note, your cutoff is very high and I have never seen a case where you need to use prec=accurate for the calculation of a barrier.

[ Last edited by cenwanglai on 2011-9-27 at 20:45 ]
回复此楼
已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖

04nylxb

木虫 (正式写手)

同求助,我QM收敛到0.2后,再换CG算法,力刷刷又上去了。
他最后一句
Also, as a side note, your cutoff is very high and I have never seen a case where you need to use prec=accurate for the calculation of a barrier.
是否是说,算barrier的时候,prec不用设置成accurate?
集中精力发文章
2楼2012-01-16 11:10:25
已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖

cenwanglai

荣誉版主 (知名作家)

老和山猥琐派九段

优秀版主

引用回帖:
2楼: Originally posted by 04nylxb at 2012-01-16 11:10:25:
同求助,我QM收敛到0.2后,再换CG算法,力刷刷又上去了。
他最后一句
Also, as a side note, your cutoff is very high and I have never seen a case where you need to use prec=accurate for the calculatio ...

CG力开始几步上去是正常的。可以先跑跑看。
3楼2012-01-16 12:46:20
已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖
相关版块跳转 我要订阅楼主 cenwanglai 的主题更新
最具人气热帖推荐 [查看全部] 作者 回/看 最后发表
[基金申请] 欢迎发来filecode的Mz6后的代码验证其规律 +31 医学老男孩 2026-08-13 69/3450 2026-08-16 00:16 by mo_mo_mo_mo
[基金申请] 有时候,自然基金真的不能太认真 (我的申报经验) +9 majunge000 2026-08-11 11/550 2026-08-15 23:20 by botar
[基金申请] filecode +8 cratir 2026-08-14 12/600 2026-08-15 18:08 by zyfgau
[基金申请] 哪位老哥知道今年的国自然具体哪一天放榜? +7 Ldrop2023 2026-08-13 7/350 2026-08-15 15:57 by mzhh2000
[基金申请] 关于Filecode分析方法 +10 majunge000 2026-08-10 13/650 2026-08-15 12:26 by hanpeng972
[基金申请] 各位道友,我要去昆明玩几天,回来见。 +7 Tide man 2026-08-14 8/400 2026-08-15 01:11 by arzu_hma
[基金申请] 是这周出结果还是下周出结果? +4 yuleib84 2026-08-11 4/200 2026-08-14 23:05 by lfy8008
[基金申请] 奇怪,两个人的filecode固定段从头到尾一模一样 +8 布布和一二 2026-08-10 11/550 2026-08-14 14:58 by Equinoxhua
[基金申请] FileCode能看出啥? +10 要乐观耀哥 2026-08-10 32/1600 2026-08-14 09:37 by 要乐观耀哥
[基金申请] 我的国基提前知道中了,可是同事的操作让我实在接受不了,怎么会有这样的人 +10 家与远方 2026-08-10 15/750 2026-08-14 02:08 by 绵羊哥哥
[基金申请] 静等基金结果 +8 gjjjzhong 2026-08-10 21/1050 2026-08-13 17:56 by 且听虎啸
[基金申请] 不应该看fileCode +7 且听虎啸 2026-08-12 9/450 2026-08-13 14:27 by flydreamws
[基金申请] Filecode 又变了,巨变 +3 WH3796 2026-08-12 4/200 2026-08-13 14:13 by 小木虫6752397
[基金申请] 分享一下我之前已中青C的计划书的filecode +4 布布和一二 2026-08-11 5/250 2026-08-13 12:56 by cratir
[基金申请] 结合人工智能,周易传统文化,filecode打分制来了,3分以上希望很大。 +3 Tide man 2026-08-12 4/200 2026-08-13 08:35 by ZJTJZ
[基金申请] 为什么网上很多人说本周 12号出结果 +6 瞬息宇宙 2026-08-10 7/350 2026-08-11 19:25 by Tide man
[基金申请] 什么时候出结果,有咨询渠道??? +3 Tide man 2026-08-11 3/150 2026-08-11 17:54 by kudofaye
[基金申请] 确定了,国自然21号放榜 +6 布布和一二 2026-08-10 7/350 2026-08-10 19:15 by 2000zf36392
[基金申请] 国自然结果 +4 Vierhys 2026-08-10 8/400 2026-08-10 15:06 by Vierhys
[基金申请] 面上项目filecode邪修 +5 西山十月 2026-08-09 7/350 2026-08-10 07:32 by 仁砚薪传
信息提示
请填处理意见