xiaoqiugood的个人博客分享 http://blog.sciencenet.cn/u/xiaoqiugood

博文

赝势平面波方法-摘录

已有 14775 次阅读 2013-5-2 00:39 |个人分类:理论学习|系统分类:科研笔记| 平面

0. 平面波赝势方法II -flash学习网址

http://jpkc.hust.edu.cn/jsjkxyclsjjc/uploadfiles/04.swf


1.摘自

http://abinitio.yo2.cn/articles/%E8%BD%AC%E8%B5%9D%E5%8A%BF%E5%B9%B3%E9%9D%A2%E6%B3%A2%E5%9F%BA%E6%B3%95ppw%E7%AE%80%E4%BB%8B.html

我们通常感兴趣是形成化学键的价电子,而不是那些内壳层的电子.而价电子在原子核附近,为了和低能级的内层电子保持正交,波函数要大幅振荡,从而具有很大的动能.但这动能的很大一部分被原子核附近的势能所抵消,考虑到原子核和内层电子的这种性质,我们可以用一个势场较弱的赝芯来代替核和内层电子.我们同样用较平缓的赝波函数来代替核附近振荡剧烈的价电子波函数。在一般情况下,一个较弱的势和较强的势描述的不是同一个体系,但是,在赝势方法中,目标并不是要描述内壳层的电子行为,而是希望在赝势区外,赝波函数、赝势和真实的波函数、势场相同,从而有相同的电荷密度。赝势近似很好的描述了价电子包括相对论效应在内的一系列性质。

 我们也可以从量子散射的角度来理解构造赝势的可行性。在用分波法计算某个势场的量子散射几率的时候,各分波的相位平移决定了此势场的散射性质。但是我们知道,如果相位平移改变一个π的话,对于势场外的散射解没有任何影响。所以说,给定一个势场外的散射解,可以同时有很多个势场满足此散射解。同样的,对于价电子来说,我们也可以构造一个势,使其对价电子来说有与真实势场同样的散射性质。

 从上面我们可以知道,赝势的作用就是除去波函数在内壳层的剧烈振荡部分,使其在内壳层区域更加平缓。如果给定于某个角动量量子数l,没有相对应的内壳层电子的话,则相应的波函数就没有结点(如铜的3d电子),赝势方法的效果不如有内壳层电子时来的明显。

 对于赝势的基本要求是:
 ·对价电子来说,由赝势决定的散射性质或相移和由核与芯电子产生的散射性质或相移相同
 ·在赝势区域内,赝波函数没有节点

 赝势又可分为局域赝势和非局域赝势。非局域赝势可表示为VNLlm|lm>V_l<lm|$两者的主要差别在于非局域
 赝势对于依赖于角动量,而局域赝势只是到核的距离的函数。

 下面介绍两种最简单的赝势形式:
 ·Ashcroft空芯赝势
  此赝势有如下的形式
  V(r) = -Ze/r , r>rc
  = 0 , r<rc

 ·傅立叶分量参数化赝势
 
  V(r)=ΣK'VK exp(iKR)
 
  这里对K的求和只限于有限的几个K点。这种参数化赝势在平面波基的计算中被广泛采用。

 
 对于用一般方法构造的赝势,在全能量计算中,为了精确地算得交换相关能,要求在赝势区域外,真实波函数与赝波函数不仅对空间依赖相同,而且绝对值相同。
 这就要求在赝势区域内,真实波函数与赝波函数模的平方的积分相同。一般来说,赝势区域的半径越小,赝势的可移植性就越强。

 赝势近似的优点:
 ·需要用到的平面波基比较少,减小了计算量
 ·由于去掉了内层电子,所需要计算的电子波函数比较少
 ·由于势场比较弱,总能量比较小,所以能够更精确的进行计算。应当注意的是在赝势近似下,绝对能量没有意义,只有相对能量才有物理意义

2. 摘自

http://lillian0619.i.sohu.com/blog/view/207401186.htm

1、 对晶体模型进行了参数收敛性测试,现在想要看看优化之后的晶体的总能量,请问应该怎么看呢?

    优化后的结果文件.Castep有 很多 final energy ,选择最后一个,因为计算是一个迭代的过程。

2、在MS castep中在进行计算之前需进行收敛性测试,我只大概知道步骤是先进行截断能的测试,在得到收敛后,再进行k点取值设置,但是我不太明白,这个收敛的具体含义是什么,是从什么地方的什么值看出其收敛的呢?  

ppt的内容没有看懂,先留个空白吧

这个要看你的能量图是不是一条直线,二要看哪个收敛图,它上面不是有4条线吗,每条实线要在各自对应虚线的下方就表明收敛了,哈哈,小小建议,这是自己分析出来的。
3、难道每次计算前都要进行收敛性测试吗?我觉得不同的截断能算出来的能量也差不多,为什么要去寻找能量最低的一个截断能或者其他参数呢?
收敛测试当然是必要的,不然你怎么能确定你的参数是合理的呢。
之所以你会觉得“不同的截断能算出来的能量也差不多” 是因为castep自己会根据你所选择的赝势、晶胞尺寸,来提供预设参数。选择fine精度时,这些参数一般是可以满足能量收敛性的。所以在这些参数附近测试,能量值自然变化不大,因为已经收敛了。
但有两个缺点
1.默认参数可能精度不够;尤其是涉及应力的计算,比如力学性质或晶格优化,其需要对应力进行收敛测试,通常其对参数的要求要高于能量收敛测试。
2.默认参数可能精度过高;对于小体系并没有什么影响,但大体系则很明显。因为截断能和k点数目的不同是可以成倍增加或减少计算量的,在大体系中可能会造成几天的计算差别,所以花个几小时找到适合的参数其实是很划算的。

我说的收敛就是“收敛性测试”里的收敛,并非某一次计算的时候scf的收敛;这是两个概念,“收敛性测试”和几何优化是否成功是没有关系的。举个例子,比如说你用截断能100eV几何优化某个晶胞,得到晶格参数为3.0A,使用200eV优化为3.1A,使用300eV为3.13A,使用400eV为3.14A......
随着截断能不断增加,得到的晶格参数逐收敛于某一数值,这个过程就是收敛测试。你要根据你的需要,比如你想让精度达到0.01A,那就要使用400eV以上的截断能,如果是精度0.1A就够,用200eV的截断能就可以了。
但上述这些优化过程都是成功的,但并不代表其参数就是收敛的

所谓基态就是能量最低态。这是因为自洽场迭代以及变分原理导致的。
其原则是: 基态是能量最低态,只要你找到一个能量最低态,就是基态。

3、几何优化的时候会不会选择高对称性,我每次都选择了 可以最后优化后的晶格常数变化很大,角度也变了,我就是想计算之前选择NO就不会出项这样的结果了,还是有一种说法把晶格固定住,你对于这个问题是怎么看的?

你可以学习下这个选项到底是怎么回事,帮助文件里应该有详细的解释。
大致上是说如果选择使用对称性,则优化的时候体系保持初始对称性不变;如果不选择,则不考虑对称性的限制进行更彻底的优化。
是否选择对称性与体系有关,如果你算的是简单的晶体或高对称性的分子,使用高对称无可厚非,而且可以大幅度降低计算量。如果你算的是复杂体系,比如结构变化复杂,涉及到对称性的变化,这个时候选择约束对称性可能无法找到最稳定的构型.

//////////////////////////////

请问k-point到底是怎么回事?有没有相关资料?我不知道如何设置,在例子中经常设置一个中,低或者高,但是文献中经常设置Morkhorst-pack grid, 好像如果表面计算,必须设置这种,但是我不明白是什么意思.

在各种周期性边界条件的第一原理计算方法中,需要涉及到在布里渊区的积分问题。如果采用普通的在布里渊区内均匀选取k点的方法,那么为了得到精确的结果k点的密度必须很大,从而导致非常大的计算量。这使得计算的效率非常低下。因此,需要寻找一种高效的积分方法,可以通过较少的k点运算取得较高的精度。

因此引入K点,k是一个指标标记平移群平变换的不可约表示。黄昆在其固体物理书中说K是空间平移操作本征值的量子数,而空间平移群的本征值是复空间单位圆。在固体物理里我们还会看到,k还有更多的含义。

在量子化学里面,能量与分子所属点群的关系,点群的不可约表示数与能级数相等。
在固体物理里面,由于固体属于空间群,因此空间群的不可约数也应与能量本征值对应;然而空间群的不可约表示数是由平移群诱导K点波矢群的可允表示,再由可允表示诱导空间群的不可约表示,因此空间群的不可约表示与平移群的不可约表示联系紧密,而平移群的不可约表示是由布里渊区的K点来表示的,因而固体本征能量也是与K点对应的。

为了作图方便,一般在能带计算中选取一个K面或K线来进行计算,这实际上也就是我们通常所看到的能带图的形式。
由于晶体对称操作有很多,k 矢量的取值有很多,但对一般的k点,没有解析解,只有在布里渊区中某些高对称点,如r, x或I点才有解析解。所以一个可行的办法就是让k 的取值沿着一定的路径走(通常就是选择一些较高的对称点,来确定k的取值路径),最后回到起点。这里虽然是确定了几个点,其实是通过这几个点形成波矢。

至于K-point设置,一般来说,k点越密越多,计算精度也越高,当然计算成本也越高。 对于k点的需求,金属>>半导体,绝缘体,不过呢,很多时候主要还是受硬件限制简约化可以使k点的数目大大下降。对于原子数较多的体系的计算,就需要谨慎的尝试k点数目,尽量减少k点数目。


Morkhorst-pack grid 是K-point的一种设置方法,目前被大多数计算软件采用

3. 华中科技大学:江建军课件

http://jpkc.hust.edu.cn/jsjkxyclsjjc/course/read.asp?fid=3&cid=33&id=38

4. 杂化泛函计算FAQ--HSE计算能带结构

http://emuch.net/bbs/viewthread.php?tid=4016318


5.摘自

http://www.tcm.phy.cam.ac.uk/~mds21/thesis/node15.html


The Choice of a Basis Set - Plane Waves

The Kohn-Sham orbitals, tex2html_wrap_inline2527 in Equation gif, may be represented in terms of any complete basis set. Many choices are possible including atomic orbitals, Gaussians, LAPWgif and plane waves, the basis set we use in practice. The use of a plane wave (PW) basis set offers a number of advantages, including the simplicity of the basis functions, which make no preconceptions regarding the form of the solution, the absence of basis set superposition error, and the ability to efficiently calculate the forces on atoms (See Section gif).

In general, the representation of an arbitrary orbital in terms of a PW basis set would require a continuous, and hence infinite, basis set. However, the imposition of periodic boundary conditions allows the use of Bloch's Theorem [20] whereby the tex2html_wrap_inline2527 may be written

equation260

where the sum is over reciprocal lattice vectors tex2html_wrap_inline2531 and tex2html_wrap_inline2533 is a symmetry label which lies within the first Brillouin zone. Thus, the basis set for a given tex2html_wrap_inline2533 will be discrete, although in principle it will still be infinite. In practice, the set of plane waves is restricted to a sphere in reciprocal space most conveniently represented in terms of a cut-off energy, tex2html_wrap_inline2537 , such that for all values of tex2html_wrap_inline2531 used in the expansion

equation266

Thus, the convergence of the calculation with respect to basis set may be ensured by variation of a single parameter, tex2html_wrap_inline2537 . This is a significant advantage over many other basis set choices, with which calculated properties often show extreme sensitivity to small changes in basis set and no systematic scheme for convergence is available.

The choice of periodic boundary conditions is natural in the case of bulk solids which exhibit perfect translational symmetry. If we wish to model an isolated molecule we must artificially introduce periodic boundary conditions by construction of a supercell (See Figure gif). In this scheme the calculations are performed on a periodic array of molecules, separated by large vacuum regions. In the limit of large separation between the periodic images, the results will be those for an isolated molecule. Therefore, care must be taken to converge the results with respect to supercell dimensions.


figure273
Figure: A schematic illustration of a supercell geometry for a molecule. The boundaries of the supercell are depicted by dashed lines.


The electron density tex2html_wrap_inline2469 and energy are given by averaging the results for all values of tex2html_wrap_inline2533 in the first Brillouin zone,

equation280

where tex2html_wrap_inline2547 , and

equation287

In an extended system, these integrals are replaced by weighted sums over a discrete set of tex2html_wrap_inline2533 -points which must be carefully selected to ensure convergence of the results [21]. An isolated molecule will exhibit no dispersion, ie. there will be no variation of E and tex2html_wrap_inline2469 with tex2html_wrap_inline2533 . Therefore, these properties need only be calculated at a single tex2html_wrap_inline2533 -point. There has been significant discussion regarding the optimal choice of tex2html_wrap_inline2533 -point for performing calculations on isolated systems [22]. For the molecular calculations presented in this thesis, the tex2html_wrap_inline2561 point, the origin in tex2html_wrap_inline2533 -space, was chosen. This choice confers significant savings in storage and computation, as the coefficients tex2html_wrap_inline2565 may be represented as real numbers, whereas in general they are complex.

The principle disadvantage of the use of a PW basis set is the number of basis functions required to accurately represent the Kohn-Sham orbitals. This problem may be reduced by the use of pseudopotentials as described in the next section, but several hundred basis functions per atom must still be used, compared with a few tens of basis function with the use of some atom-centred basis sets.


/////////////////////

The Exchange-Correlation Term

The Kohn-Sham DFT approach to the solution of the many-body Schrödinger equation has not required any approximations thus far. However, the exchange-correlation energy, tex2html_wrap_inline2477 in Equation gif, is defined as the difference between the true functional tex2html_wrap_inline2463 and the remaining terms. As the true form of F is unknown, we must use an approximation for tex2html_wrap_inline2505 .

A number of possible approximations may be made. The simplest, known as the Local Density Approximation (LDA), defines tex2html_wrap_inline2505 as

equation236

where tex2html_wrap_inline2509 is the exchange-correlation energy per unit volume of a homogeneous electron gas of density tex2html_wrap_inline2469 . The values of tex2html_wrap_inline2513 were calculated by Ceperly and Alder using Quantum Monte Carlo techniques [13] and parameterised by Perdew and Zunger [14]. Although a gross approximation, LDA has been found to give good results in a wide range of solid state systems [15]. Generalised Gradient Approximations (GGAs) add a term in the gradient of the electron density to the parameterisation of tex2html_wrap_inline2505 . Although GGAs do not offer a consistent improvement over LDA in all types of system, they have been shown to improve on the LDA for calculations of molecular structures [16] and in representing weak inter-molecular bonds [17]. For this reason the GGA due to Perdew and Wang [18] has been used in this thesis. In cases where the external potential is spin dependent, an approximation must be made to tex2html_wrap_inline2505 which depends on both the total electronic density tex2html_wrap_inline2519 and the polarisation tex2html_wrap_inline2521 , where tex2html_wrap_inline2523 and tex2html_wrap_inline2525 are the densities of spin up and spin down electrons respectively. We have used the spin dependent GGA also due to Perdew and Wang [18] for the spin-dependent calculations presented in this thesis.

The reasons for the success of these approximations are not well understood, although this may be partially attributed to the fact that both obey the sum rule for the exchange-correlation hole in the electron density [19]. Certainly, LDA and GGAs give rise to a systematic overestimation of the electronic binding energy. However, differences in energies may be accurately computed and it is these which are important for the estimation of physical and chemical properties.




https://blog.sciencenet.cn/blog-567091-685808.html

上一篇:关注的问题-物理+化学+计算
下一篇:VASP总能计算参数设置-摘录
收藏 IP: 128.84.126.*| 热度|

0

该博文允许注册用户评论 请点击登录 评论 (0 个评论)

数据加载中...
扫一扫,分享此博文

Archiver|手机版|科学网 ( 京ICP备07017567号-12 )

GMT+8, 2024-11-24 13:28

Powered by ScienceNet.cn

Copyright © 2007- 中国科学报社

返回顶部