Thanks for your quick response. It now seems there may be a computer “glitch” on our side. We did the tests you suggested and also tried to reproduce the strange results previously mentioned and now the runs are more “normal”. Since you imply that a simulation cell with a fixed number of atoms and periodic boundary conditions should be insensitive to the placement of a vacuum layer, we will check that the potential we are using is well converged wrt to cutoff or perhaps there really was bad processor. Thanks for your help. Natalie
Thanks for your quick response. It now seems there may be a computer
"glitch" on our side. We did the tests you suggested and also tried to
reproduce the strange results previously mentioned and now the runs are more
"normal". Since you imply that a simulation cell with a fixed number of
atoms and periodic boundary conditions should be insensitive to the
placement of a vacuum layer, we will check that the potential we are using
is well converged wrt to cutoff or perhaps there really was bad processor.
strictly speaking, the position of the origin *does* have an impact,
but it should not be as drastic as what you described.
the problem i am referring to originates in the variable "number
density" of floating point numbers. due to the way how floating point
numbers are encoded on a computer, you have the situation, that you
can only represent a limited number of numbers in a given interval and
this amount depends on the absolute value.
for example. between 2.0 and 3.0 you can represent only half as many
numbers as between 1.0 and 2.0. if you go to the interval of 4.0 and
5.0 it is halved again and so on. consequently, the systematic error
on computing the force between two atoms depends on what their
absolute position is. close to 1.0 it will be much smaller than at
1000.0. for a homogeneous bulk system, those errors will cancel a lot,
but in a slab configuration less so. how large this error is, depends
a lot on the overall size of the system, and you can reduce it by
placing the system so that the origin is in the middle of the system.
axel.