Differences

This shows you the differences between two versions of the page.

--- howto:md [2014/02/09 15:18] – oschuett
+++ howto:md [2015/05/11 18:39] – [MD Convergence] 130.60.136.225
@@ Line 23: / Line 23: @@
 If the MD has to be trusted then one should be sure that the trajectory can be trusted. Actually a simple, and very sensitive test that there are no big technical errors is to perform an NVE trajectory and look at the energy conservation. The energy conservation has normally two features, short time oscillations (that are larger when the system is not yet equilibrated) and a long range drift. If you express these in //K/text//, then you can compare them with the temperature that you are simulating at. Another (equivalent) possibility is to express them as fraction of the kinetic energy. The oscillation and drift (normally per //ps//, but it also depends on how many picoseconds you want to simulate, and if you want an NVE trajectory) should be small with respect to the kinetic energy (1% or less is a good value, but obviously it depends on the accuracy that you want to achieve, more might be acceptable, or less needed).
-To improve the energy conservation one can either improve the forces with EPS_SCF and EPS_DEFAULT, improve $\tilde n$ increasing the cutoff, and reduce the timestep. For the timestep a good value is normally around 1 fs, but if you are simulating high temperature or special system you might need to reduce it. Normally having good forces and larger timestep is advantageous. Obviously the timestep cannot be smaller than the frequency of the highest oscillations of the system (typically H atoms, that is the reason sometime D atoms are used).
+To improve the energy conservation one can either improve the forces with EPS_SCF and EPS_DEFAULT, improve $\tilde n$ increasing the cutoff, and reduce the timestep. For the timestep a good value is normally around 1 fs, but if you are simulating high temperature or special system you might need to reduce it. Normally having good forces and larger timestep is advantageous. Obviously the timestep cannot be larger than the period of the highest oscillation of the system (typically H atoms, that is the reason why sometimes D atoms are used).
 Conserving the total energy well is not enough if the system is very heterogeneous and the interesting part is small. In that case there is the risk that even a large error on it might pass unnoticed. If this is to be excepted (for example the atom with the sharpest gaussian basis set is in the interesting part) then checking that part of the system (oscillations, kinetic energy,...) is advisable.