Hello,
could you explain my the secret of choosing to time step for md
integrator? I want to perform a 10-20 ns
simulation in water in order
to, for ex., calculate some interactions or to see if the
structurewould relax to a stable state.
If I use vsites, heavyh and LINCS I can run with 6 fs step. This would
be quite quick.
I just wonder if this is `by book`.
I appreciate your help.
SDA
-------------------------------------------------------------------------------------------------
Turn off any thermostats you have on (perform an NVE run) and then try
out short runs (say, 5000 steps) with different time steps. Plot out the
total
energy graphs for the different runs.
If your time step is too big you'll end up with total energy either
increasing or decreasing. With a very big time step you might end up
with an exploding system..
If you use constraints you are able to use larger time steps.
With hydrogen bond length constraints the time step is about 2 fs, if
the molecules are flexible the time step needs to be a lot smaller,
since hydrogen atoms oscillate very fast because of their small mass.
补充一个从陈正隆那本书里看来的，可以作为一开始的时间步长的估计。
积分步长（intergration timestep）的选取：
通常的原则是：积分步长应小于系统中最快运动周期的1/10。
有一种12-6势能估计原子系统的积分步长的方法：
U’‘(r
min)=k //势能函数最低点的两次微分相当于力常数k
k=4π^2*v^2*u //由简谐振子的关系式可得。u is the reduced mass，v is the frequency。
各种模拟系统所建议采取的积分步长：
系统
|
运动型态
|
积分步长（s）
|
简单原子体系
|
移动
|
10^-14
|
刚性分子
|
移动与转动
|
5X10^-15
|
软性分子，限制键长
|
移动，转动，扭动
|
2X10^-15
|
软性分子，无键长限制
|
移动，转动，扭动，振动
|
10^-15或5X10^-16
|
