Top Page > Browsing

OpenMX hangs on Band calc and DOS Date: 2005/03/18 04:57 Name: Ken Lux   <lux@engr.wisc.edu> Hi, I've been trying to get openmx2.3 to do a calculation of the band structure and DOS of CePt2. I have run into a couple problems with the program hanging when it goes into Band_DFT_kpath and also with DOS calculations, but first more details: System - Linux(Slackware 10.0), P4 3.2GHz, 1GB RAM, openmx2.3 I used adpack1.8 to make fully relativistic pseudo atomic orbitals and pseudopotentials (although I couldn't get good convergence on the Ce, but I decided to soldier on ahead). I'm doing a non-colinear calculation with spin-orbit coupling (since I'm dealing with 4f electrons). I did just the SCF calculation 1st and then am using the restart file generated from that and turning on the band and DOS calcs (if I tried to do the band or DOS calcs first, the system would hang and not output restart files or an .out file). When I try to run openmx with the above input file I get: Band_DFT_kpath start kpath 1 (0.000000 0.000000 0.000000)->(0.000000 0.500000 0.500000) 2 (0.000000 0.500000 0.500000)->(0.250000 0.500000 0.750000) 3 (0.250000 0.500000 0.750000)->(0.500000 0.500000 0.500000) 4 (0.500000 0.500000 0.500000)->(0.000000 0.000000 0.000000) And the computer keeps running at 100% CPU utilization for ever (at least 48 hrs), but never seems to oputput the final kpath (if I tell openmx to use 30 points per kpath, it hangs after the 1st kpath). I noticed that the cept2.Band file is 100 lines long so it has 90 lines of band structure data. Also, I noticed in the directory listing that the cept2.DOS.val file is 0 bytes long. Any ideas as to why it hangs up and/or a possible fix? Thanks, Ken

Page: [1]

Re: OpenMX hangs on Band calc and DOS ( No.1 ) Date: 2005/03/18 18:07 Name: Ken Lux Additional info: I get the same problem with the GaAs example - hanging on the 4th kpath. Re: OpenMX hangs on Band calc and DOS ( No.2 ) Date: 2005/03/19 00:51 Name: T.Ozaki Hi, I guess that your problem comes from a lapack routine, since I could not reproduce the stop problem even for the GaAs case using OpenMX2.3 in my computational environment. In OpenMX2.3, "dstegr", which is a lapack routine, is used as default to diagonalize the tridiagonalized matrix. However, the routine seems to hang up often as you met, although it is perhaps the fastest one. In addition, the hanging behaviour depends on computational platform. Such a case would be resolved by using "dstevx" instead of "dstegr". Then, the following keyword is available. scf.lapack.dste dstevx # dstegr|dstedc|dstevx, default=dstegr Would you try again your calculation using this option ? Also, I have prepared fully relativistic pseudo potentials and basis functions of cerium, which are now downloadable on the website. Although you seemed not to get convergence, I was able to generate them with perfect SCF convergence (it is recommended to compile ADPACK by using gcc rather than icc for numerical stability). In the dimer calculation on the website, Ce9.0 is used as basis set. However, the use of basis function with a smaller cutoff radius would be better in bulk systems in order to avoid the overcompleteness. After the release of OpenMX2.3, we have fixed several bugs. If you would prefer the latest version to OpenMX2.3, the source code is available from http://staff.aist.go.jp/t-ozaki/source2.72.tar.gz (the default of "dsteXX" has been also changed from "dstegr" to "dstevx" because of the above reason). But this is a tentative release, since bugs we have recognized still remain. Regards, T.Ozaki Re: OpenMX hangs on Band calc and DOS ( No.3 ) Date: 2005/03/19 03:30 Name: Ken Lux OK, I tried the GaAs with v.2.3 and with the keyword changed and it works fine! I'll give the CePt2 a try and post how it turns out. Thanks, Ken Re: OpenMX hangs on Band calc and DOS ( No.4 ) Date: 2005/03/22 10:01 Name: Ken Lux  <lux@engr.wisc.edu> It looks like it works with the CePt2 system now as well. Although it doesn't seem to make much of a difference if I use a restart file or not. Thanks! how to use *.RST.tar file ( No.5 ) Date: 2005/03/26 12:42 Name: s. j. luo  <sjluo@phys.ntu.edu.tw> Hi, The SCF iteration is finished when the SCF criterion is not satisfied,I want to continue the SCf calculations by using *.RST.tar in order to save time.Could you tell me how to *.RST.tar! Thank you in advance! luo

Page: [1]