Singular matrix in lu_decomposition

Alloy Theoretic Automated Toolkit (ATAT)
1

Happy new year ATAT alloy theorists!

I am trying to use fitsvsl: I have run fitsvsl -ns=3 -er=8 for an AB2 system. pollmach runstruct_vasp completed.
When fitsvsl -f -ns=3 -er=8 was run as the next step, it prints:

  1. Reading 0/vol_0.5/p+0.2_8_0
    …
    …
    …
  2. Reading 203/vol_1/p+0.2_8_9
    Singular matrix in lu_decomposition
    Segmentation fault.

Any advice to solve this problem will be helpful.

With regards,
Ravi
IGCAR/Kalpakkam

2

Possible solutions:

  1. Decrease -fr=… (reducing the number of force constant makes it more likely that the fitting matrix is nonsingular)
  2. use the -cs option. This will set to zero the force constant that cannot be determined.
  3. If no reasonable value of -fr makes it work, then you may need to re-generate perturbations with a bigger supercell, i.e. -er=larger number.
3

-fr and -cs are available for fitfc. They are not available for fitsvsl.

Of course my original post is incomplete: We have done CE for an AB2 system on modified AlB2-type lattice. That is, Al sites are occupied fully by B atoms. B sites are occupied randomly by both A and B at equal concentrations. Our aim is to compute the ground state configuration and its free energy as a function of temperature and composition (not just for the ground state composition).

Ground state is determined. EMC2 can give its free energy as function of composition and temperature.

We would like to include phonon contributions as well via fitsvsl.

But, like I have posted, it stops with the singular matrix error.

Any advice would be helpful.

4

Is it possible that for some types of bonds you only have one bond length available?
What does gnuplot fitsvsl.gnu
look like?