# Atomic forces not decreasing for structural optimization

**URL:** <https://discourse.abinit.org/t/atomic-forces-not-decreasing-for-structural-optimization/3529>\
**Category:** Ground state\
**Created:** [June 24, 2024, 10:04pm UTC](https://discourse.abinit.org/t/atomic-forces-not-decreasing-for-structural-optimization/3529 "2024-06-24T22:04:51Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![iavas](https://discourse.abinit.org/letter_avatar_proxy/v4/letter/i/6de8d8/32.png) [@iavas](https://discourse.abinit.org/u/iavas)\
**Post date:** [June 24, 2024, 10:04pm UTC](https://discourse.abinit.org/t/atomic-forces-not-decreasing-for-structural-optimization/3529/1 "2024-06-24T22:04:51Z")

</div>

Hello all,

I’m trying to perform structure optimization calculations for vacancy defects in a batch of iron-based alloys and would like to test my workflow on pure BCC-Fe first. I used a 4x4x4 BCC supercell with 128 Fe atoms and removed the middle-most atom to create the vacancy. Only the atom coordinates were allowed to relax. The input file and the pseudopotential (acquired from Pseudo Dojo) are below:  
[Fe-bulk-v.abi](https://discourse.abinit.org/uploads/short-url/bBGKCKA1LxkRJB0eFxx4uJIjB9u.abi) (6.9 KB)  
[Fe.psp8.txt](https://discourse.abinit.org/uploads/short-url/m0aj3qqN8L511BTDDHQIvXin8Sh.txt) (273.5 KB)

The problem I am having is that after trying to perform more than twenty steps of relaxation, the maximum atomic force is still the same as it was at the beginning, hovering around 1e-2 Ha/bohr, two orders of magnitude above the convergence criterion I was hoping for, with no signs of decreasing. The results file is below (the job has been manually killed):  
[Fe-bulk-v.abo](https://discourse.abinit.org/uploads/short-url/89h8ZpTp7YkCRlXemiG7wIVodxj.abo) (1.3 MB)

I have very little experience with structural optimization of such large supercells and am not sure if this is normal. Also, a single SCF step (not an iteration, just one SCF step) on this supercell takes roughly 6 to 7 CPU hours (more than 2 minutes at ~200 CPU parallelism). Is such a computation time normal?