关于能带不收敛(转贴)

greencalm

问uring my calculation of bandstructure,the output always tells
>>that\"Warning:band structure not converged\",what causes this
>>problem.How to solve it? In this condition,is the result I get such
>>as DOS reasonable? Thanks!
>
答:This problem usually occurs when there are degenerate states in the
>band structure, unfortunately this is very common! The problem is
>being looked at, but for the moment you can get the bands to
>converge by increasing the number of conjugate gradient steps it
>uses per iteration (bs_max_cg_steps) and/or the total number of
>iterations it does (bs_max_iter). Both of these are in the .param
>file, and the default values are:
>
>bs_max_cg_steps : 4
>bs_max_iter    : 60
>
>As for how these affect subsequent analyses, it really depends on
>how far the bands are from convergence and how many bands are
>affected.
>
>I hope that helps,
>
>hil Hasnip
>
>--------------------------------
>Dr P. Hasnip (pjh1003@cam.ac.uk)
>Atomistic Simulation Group
>Materials Science & Metallurgy
>Cambridge CB2 3QZ
>Tel +44 (0)1223 334496

greencalm

答2:>In general it's never safe to ignore a convergence failure report
>for
>any part of a CASTEP calculation.   In this case you can repeat
>just one of the k-points and compare the eigenvalues to determine
>the level of convergence actually achieved.
>
>For a DOS calculation you only need eigenvalues as accurate as the
>divisions used to calculate the DOS, which is actually quite coarse.
>I don't know the value but I can guess that it's probably greater
>than
>50meV.  While the convergence tolerance set is (I guess) somewhere
>around
>1e-5eV.  So you will propbably find that the actual convergence
>achieved is
>adequate.  But I'll emphasise again the need to check.
>
>There is a known problem with bandstructure convergence in CASTEP.
>(This is
>also a big ussue for linear response calculations  which perform an
>initial
>bandstructure calculation and need rather higher accuracy than a DOS
>or a
>bandstructure plot).  During the investigation of this we have
>recently
>found that increasing the parameter BS_MAX_CG_STEPS to up the number
>of
>consecutive conjugate-gradient iterations from the default 4 to
>25-50
>helps increase the rate of convergence.  For future runs I recommend
>this setting.
>
>Keith Refson
>
>
>--
>Dr Keith Refson,
>Building R3
>Rutherford Appleton Laboratory
>Chilton
>Didcot
>Oxfordshire OX11 0QX
>T: 01235 778023      K.Refson@
>F: 01235 445720      @rl.ac.uk

greencalm

在那个邮件列表那边拷贝过来的 呵呵

lichenliang

不错，很好啊

hanjiangxue

真不错

lichenliang

我想问一下：文中提到的BS_MAX_CG_STEPS 和bs_max_iter  这两个参数，在castep计算框中没有直接找到啊。是不是要通过修改别的参数从而间接修改这两个参数啊？是这样吗？