High order of the polynomial in the fitsvsl

Alloy Theoretic Automated Toolkit (ATAT)
1

Dear Axel,

I want to know whether it is make sense to use the 3- or 4 order of the polynomial to fit the stiffness vs length.
For the 4 order polynomial (command "fitsvsl -f -op=4"), the miscibility gap decreases and I have good result. In the case of linear fitting, the miscibility gap donot decrease but increase (compare to the one without vibrations effect).
It looks like the stiffness and length obey the linear relation within the range of nearest neighbor (about 2.6 angstrom), but tend to be a 4 order to a large distance (3.1 ansgstrom). I cannot find the use of higher order fitting in the literatures, except the use of second order fitting in the Cu-Li system.So I am not sure the 4 order fitting make sense in the Length Dependent Transferable method. And Is there a command in the "fitsvsl" which can be used to cut the fitting distance just as the one (-fr) in the "fitfc". Thanks.

Regards,
wz chen

2

It looks like you really have two consecutive linear trends at different distances
(probably for 1nn and then for 2nn neighbors). Your curves fit the points, but I would double check that no structures have bonds to fall in the interpolated region in between which may be nonphysical.

Usually, only 1nn force constant are transferable. In some structures (e.g. BCC), the geometry is such that 2nn are considered 1nn by default in the fitsvsl code. You can override that by specifying a cutoff -msl=?? to eliminate the 2nn. You would then use a linear trend only. Maybe that’s a more physical option.

3

Got it, really appreciate for your help.
Regards,
wz chen