Difference between revisions of "Selecting search parameters for GMIN"

From CUC3
Jump to navigation Jump to search
import>Mp466
import>Mp466
 
Line 3: Line 3:
 
1. The number of '''STEPS''' should be long enough to at least see one conformational change without restarting.
 
1. The number of '''STEPS''' should be long enough to at least see one conformational change without restarting.
   
2. The step size '''STEP''' should be roughly 2/3 of an interaction distance. Longer distances maybe required for the system.
+
2. The step size '''STEP''' should be roughly 2/3 of an interaction distance for atomic clusters. Longer distances maybe required for the system.
   
 
3. The number of My-LBFGS minisation steps '''MAXIT''' should be large enough so 99% of the steps converge. This promotes more
 
3. The number of My-LBFGS minisation steps '''MAXIT''' should be large enough so 99% of the steps converge. This promotes more
Line 9: Line 9:
   
 
4. The final quenches should be ordered with respect to their energies. If they are not, it means that lower
 
4. The final quenches should be ordered with respect to their energies. If they are not, it means that lower
energy structure have been found upon tight convergence and the '''SLOPPYCONV''' needs to be tightened.
+
energy structure have been found upon tight convergence and the '''SLOPPYCONV''' needs to be tightened. Also if
  +
the same minimum is being found with the tight convergence it suggests '''SLOPPYCONV''' is not differentiating between
  +
minima well, and therefore should be tightened as well.
   
 
5. The '''TEMPERATURE''' should be high enough to over come barrier heights., but not so large that all barriers are
 
5. The '''TEMPERATURE''' should be high enough to over come barrier heights., but not so large that all barriers are

Latest revision as of 17:27, 3 August 2009

There are a few general tips for selecting parameters for geometry optimisation.

1. The number of STEPS should be long enough to at least see one conformational change without restarting.

2. The step size STEP should be roughly 2/3 of an interaction distance for atomic clusters. Longer distances maybe required for the system.

3. The number of My-LBFGS minisation steps MAXIT should be large enough so 99% of the steps converge. This promotes more efficient sampling.

4. The final quenches should be ordered with respect to their energies. If they are not, it means that lower energy structure have been found upon tight convergence and the SLOPPYCONV needs to be tightened. Also if the same minimum is being found with the tight convergence it suggests SLOPPYCONV is not differentiating between minima well, and therefore should be tightened as well.

5. The TEMPERATURE should be high enough to over come barrier heights., but not so large that all barriers are overcome thereby causing the rescaling of the move step to result in very large steps which may take a long time to converge. This is system dependent and requires testing to ensure efficient sampling.