|
|
本帖最后由 liyuanhe211 于 2016-9-1 18:25 编辑
近日 opt+freq 一组分子，结果中以相当高的概率出现不正确的虚频（GS一个，TS俩）。尝试 opt=verytight 和 int=ultrafine之后未能解决这一问题。
故求助 Gaussian support，其回复指出是因为使用了 D3BJ 的色散校正，D3BJ 在 D.01 中有已知的 Bug 导致其 Hessian 不准，从而有这一问题。
大意是（原文见本贴最后）：
- Gaussian 09 D.01 中 D3-BJ 校正的能量和梯度是正确的，但是有一个已知的bug，使得依赖于R^(-8)的一项在色散对Hessian的贡献里没有。这一贡献很小，只对“软”的频率有显著贡献。
- 目前没有办法解决D3-BJ不能用的问题，只能用D3代替。
- 后续版本会修复这个Bug。
总结来说就是要用到频率（比如opt+freq）就不要用D3BJ
----------------------------------------------------------------------------------------------
回件原文如下：
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Hello Yuanhe,
There are two variants of Grimme's D3 dispersion model available in G09 rev. D.01: the original variant, "GD3" (Grimme et al. J. Chem. Phys., 132 (2010) 154104), and the variant that uses Becke-Johnson damping, "GD3BJ" ("GD3BJ"; Grimme et al. J. Comp. Chem. 32 (2011) 1456-65).
Unfortunately, we are aware of a bug in the logic to include one of the contributions to the Hessian from Grimme's D3 model using Becke-Johnson damping ("GD3BJ"; Grimme et al. J. Comp. Chem. 32 (2011) 1456-65). The energy and gradients for the "GD3BJ" model are correct but one of the contributions to the Hessian with this dispersion model is not being included, so the frequencies are not correct. The missing term is the second derivative of the dispersion term that depends on R**(-8), so it is a fairly small term. For "stiffer" frequencies, the inclusion of such term would have an almost negligible effect but its effect would be more noticeable for "softer" frequencies.
We will fix this issue for a future Gaussian release but, unfortunately, there is no workaround for analytic frequency calculations using the "GD3BJ" model in G09 rev. D.01. Probably the best alternative would be to use the "GD3" model, i.e. Grimme's D3 model using the older damping and slightly older scaling parameters (Grimme et al. J. Chem. Phys., 132 (2010) 154104). Energies, gradients and Hessians are all correct for the "GD3" model and they are not affected by the bug discussed above (e.g. "EmpiricalDispersion=GD3"). The older "GD2" model would also be available with correct energies, gradients and Hessians (e.g. "EmpiricalDispersion=GD2").
Sincerely,
Fernando R. Clemente, Ph.D.
Technical Support
Gaussian, Inc.
help@gaussian.com
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
P.S. 按虚频方向移动原子对消虚频并没有什么卵用（Ref. Sob + 自己试验）
另外 Gaussian Support 的回复还是很快的，昨日10 pm 提问， 今天10 am 收到回复，而且十分详细。买了Gaussian版权的机构都有这个服务（不过可能要注意不要买了 A.02 问 D.01 的问题），不妨多加利用。
----------------------------------------------------------------------------------------------
Sob其实之前提到了多次（有的我还赞过（逃。。。）），但印象不深没记住，见：
“GD3BJ优化导致最后频率计算出现虚频是正常的，因为BJ阻尼在高斯里还没有准确的Hessian。”
- 公社群聊记录：2015.11.10 15:52:30
Gaussian中几何优化收敛后Freq时出现NO或虚频的原因和解决方法
http://sobereva.com/278
PS：如果你用的是D3(BJ)，改用零阻尼的D3优化和频率计算
#################
##Update##
刚写完一天，就发现Gaussian 09 E.01 中这个bug被修复了。如果有谁搞到E.01来用，就不用考虑这个问题了。
#################
|
评分 Rate
-
查看全部评分 View all ratings
|
发表于 Post on 2015-12-22 14:04:59
|
只看该作者 Only view this author
|只看大图 View big picture
|倒序浏览 Reverse view
|阅读模式 Reading model
收藏 Add to favorites
楼主 Author
发表于 Post on 2019-11-29 10:03:17