Why larger time step is not allowed in Molecular Dynamics Simulation?

Hi, 

I am trying to understand why MD simulation crashes if larger time step (~100 fs) is applied to integrate the equation of motion. I know that there are two major things play role in solving F = ma and obtaining equilibrium solution. One is the accuracy, and the other is numerical stability of the integration algorithm.

For example, I should not use any time steps (dt) that is larger than the time period of high frequency thermal oscillation. If I use, I will have to compromise accuracy, and I may not find the real equilibrium. On the other hand, the stability of numerical scheme, e.g. verlet algorithm used in MD, depends on local error propgation and is related to type of problem, system frequency and simulation dt. 

I tried to use larger dt even at very low temperature (~1 K), still found that MD simulation is crashing immediately. I thought use of larger dt would produce erroneous result instead of terminating the simulation itself.

I believe the use of larger dt brings some pair of atoms very close to each other (or overlap) in the next step, causes accumulation of exceedingly huge repulsive forces and finally crashes down the simulation.

I would like to know is there any way I can handle larger dt in MD simulation? In other words, what am I missing in understanding MD simulation? Please suggest.

-Ashfaq Adnan, PhD Student, School of AAE, Purdue University.