我的步长选择也是在飞秒量级。
刚才在maillist上有人给我发来这样一个解释:
Limitation in the time step
The limitation of the time step to a femtosecond can be understood as follows: Numerical schemes to solve Eqn.2.1 use a Taylor expansion.
Different schemes truncate at different orders of the time step which determines the accuracy of the position calculation and also the time step to be used. In brief, for numerical stability it is required that the change in potential with a change in position of the coordinate of a particle in time must not be large (note that the second term of the Taylor expansion in Eqn.2.2 is equal to ). A simple rule of thumb frequently used is that the particle can move at most the distance between two neighboring atoms [1] (lecture 2). Typical interatomic distances in a graphene plane is Å and the typical velocity of an interstitial hydrogen atom which has equilibrated with the graphene plane at say 900 K is Å/femtosecond. Therefore the time step has to be femtoseconds in this case and one would have to loop over iterations to simulate one second!!
http://www.ipr.res.in/~manoj/CompPhysII/node5.html
就是说每一步长的移动的距离在原子间距的1/20为宜。一般情况下,原子间距为1.4埃,在温度为900K时原子速度为0.047A/fs,这么一算得到步长为1.5fs。