Fitfc error: Segmentation fault, -er within range

Alloy Theoretic Automated Toolkit (ATAT)
1

I am doing harmonic phonon calculations for self diffusion in FCC metals. Using fitfc I get the following error:

Reading…
vol_0
p-0.1_5_0
p+0.1_5_0
p+0.1_5_1
p-0.1_5_2
p+0.1_5_2
p+0.1_5_3
p+0.1_5_4
Generating Fitting matrices
1/651
94/651
187/651
280/651
373/651
466/651
559/651
Performing least square fit…
Problem size= 651 x 66
Singular matrix in lu_decomposition
Segmentation fault (core dumped)

I found another user on here who had the same issue with the solution being that -er was not large enough. However I set -er= 5 with a nearest neighbor spacing of about 2.45 angstroms, so -er=5 is within range. Theses are the other parameters that I used for the fitfc commands:

-er=5.0 -ms=0.04 -ns=1 -dr=0.1 -fr=4.0 -T0=0 -T1=1600 -dT=10

I would be happy to tar my files and send them as well if a more detailed look is required.
Thanks in advance!!

2

SOLVED!!

I emailed Dr. van de Walle and received the following helpful information:

The reason for the error is that you are asking for too many force constants to fit (too large -fr ) compared to the size of the supercell you chose (-er=… option during structure generation). Typically fr<=er/2
Possible solutions:

  1. reduce -fr (-fr=3 works with your data, but I don’t know if it’s accurate enough)

  2. re-generate supercells with a larger -er

  3. The option -cs can help (it sets to zero any force constant it cannot fit). But this is only a fix if only a few cannot be determined.

I only checked solutions 1 and 3 but those two worked for me!