psm3 |
- Date: 2025/09/05 07:51
- Name: Kazume NISHIDATE
<nisidate@iwate-u.ac.jp>
- If you are using OpenMX compiled under Intel oneAPI and struggling to achieve a converged result, try the following option in your job script.
export FI_PROVIDER=psm3
Changing the environment variable FI_PROVIDER to psm3 instructs the system to use theetraditionalfmemory shearing style.
I usually use the LX system at Tohoku University, but the computer had been halted for 10 days due to scheduled maintenance. During this period, I transferred the data from the LX to the Ohtaka system at the University of Tokyo.
I had been struggling with the DFT+U/SOI calculation on the Ohtaka system to reproduce the reported band structures based on the similar input data file I posted in this forum before. I tried numerous computational options in the input file, but the calculations still did not converge, even after 300 SCF cycles. Yesterday, the LX system was resumed, and I changed my platform computer to the LX, and found a strange thing; the calculation, which did not converge in 300 SCF cycles at Ohtaka, converged in 92 SCF cycles.
I had encountered a similar phenomenon when I was calculating the DFT using QuantumESPRESSO about two and a half years ago on the Ohtaka system. At that time, the same calculation, which previously converged in a few hours, did not converge in 24 hours.
I asked the vendor of the Ohtaka about this trouble. It took a few weeks to identify the cause of the problem. They installed a new Intel oneAPI compiler, and I had been using the recompiled QuantumESPRESSO, which caused the problem. The Intel oneAPI compiler changed the internal memory sharing mechanism, and in a rare case, this results in a failed convergence.
I remembered that incident and added the line in my JOB script, and now I can obtain the converged result on the Ohtaka too.
Ifm sharing this information, hoping that it may help someone who is encountering a similar problem.
¼ŠÙ”‰è Kazume NISHIDATE Ph.D
Department of Systems Innovation Engineering, Graduate School of Science and Engineering, Iwate University 4-3-5 Ueda, Morioka, Iwate, 020-8551 JAPAN Phone:+81-19-621-6391 kazume.nishidate@gmail.com, nisidate@iwate-u.ac.jp https://sites.google.com/site/nisidatelab/

| |