24小时热门版块排行榜

返回列表

【奖励】本帖被评价5次，作者TIGERYZZ增加金币 5 个

当前主题已经存档。

TIGERYZZ

金虫 (小有名气)

应助: 0 (幼儿园)
贵宾: 0.009
金币: 867.9
帖子: 249
在线: 38.6小时
虫号: 488084

[资源] 【分享】ATK_adaptive_current

这是个计算电流的函数，应该还算是beta版的。貌似可以设置精度==
具体的功能我也不太清楚，安装和使用方法在下面。
希望虫子们用了以后一起来说一下这个程序~~~

This is not just a simple NanoLanguage script, it's actually quite
complex. So, there are binary and script files involved. I have attached
the Windows and 32-bit Linux versions in a zip file. 64-bit Linux
version is not available right now, but we can do it later.

To use it:

1) Extract the zip file in some convenient directory; it will create a
subdirectory "ATK_adaptive_current"

2) In this directory there is a subdirectory "win32" and inside it
another subdirectory called "quadpack". Copy this directory to the "lib"
directory in your ATK installation on Windows (so that "quadpack"
becomes a subdirectory of "lib"

. For Linux, same thing, except that
"quadpack" should become a subdirectory of lib/python2.4/site-packages.

3) To calculate the current using the new routine, edit the script
"calc_current_adaptive.py" to set the checkpoint file name (with path if
necessary), k-points (or, wait, you have hydrogen-terminated edges? if
so, 1x1 is fine), and level of accuracy.

About the accuracy:

* The syntax to the function calculateCurrentAdaptive closely resembles
that of calculateCurrent, with an import difference. Instead of setting
the number of integration points, you set the desired accuracy (and a
verbosity).
* You can specify the max absolute error you want in the result. The
routine will use as many points, with more points in areas of quickly
varying T(E), as needed to obtain this accuracy, up to a maximum of
interval subdivisions (so that it doesn't run forever).
* There is also a "relative accuracy", and the routine is not converged
until both errors are below the threshold, so if your current is 1e-8
Ampere, it will not converge until the error is less than 1e-11 Ampere,
even if epsabs=1e-6 Ampere.
* So, in some sense there is no need to use epsabs, since usually it's
more important to have a good relative accuracy. Thus you can actually
leave epsabs at 1e-6 Ampere (default).
* I would suggest using 1e-2 for epsrel to begin with, and see how many
evaluations that requires, before trying 1e-3 (default).
* If you set verbosity>0, the routine will print the current, the error,
and the number of iterations used. This will be very useful to see that
you have a good accuracy!
* If you set verbosity>1, it will also print the points E where T(E) is
evaluated, and T(E). You can pipe this output to a file, bring it into
Excel, sort it, and use this to plot T(E), and at the same see how the
routine distributes the points irregularly; more points will be used
where T(E) varies quickly.

纳米盘地址：
http://www.namipan.com/d/ATK_ada ... 456600e82d04fa20200
zdhlover说下不下来，额我传了个到小木虫~~

[ Last edited by TIGERYZZ on 2009-5-11 at 13:42 ]

回复此楼