Re: [SIESTA-L] SCF convergence issue
Also, maybe increasing the number of pulay steps (8-10) would help to speed up convergence - not sure if it would solve this problem, though. I've always had better convergence when I set the number of pulay steps to 8 or so. Marcos Have a look at the distances your nuclei are moving per CG step. Try reducing your MD.MaxCGDispl if many of them are moving that far. It'll increase the number of CG steps you'll need, but it might stabilise your system. Edan. Cherry Y. Yates wrote: snipped -- Dr. Marcos Verissimo Alves Post-Doctoral Fellow Condensed Matter and Statistical Physics Sector International Centre for Theoretical Physics Trieste, Italy I have become so addicted to vi that I try to exit OpenOffice by typing :wq!
Re: [SIESTA-L] SCF convergence issue
Have a look at the distances your nuclei are moving per CG step. Try reducing your MD.MaxCGDispl if many of them are moving that far. It'll increase the number of CG steps you'll need, but it might stabilise your system. Edan. Cherry Y. Yates wrote: snipped
[SIESTA-L] SCF convergence issue
Dear SIESTA guys, I am trying to relax a big structure using CG and diag. At first few steps, it looks pretty good, something like this: siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax Ef(eV) siesta:1 -258035.2072 -258036.9370 -258036.9370 0.0275 -4.3651 siesta:2 -258036.2208 -258035.0256 -258035.0256 0.0112 -4.5080 siesta:3 -258035.1812 -258036.5612 -258036.5612 0.0218 -4.4097 siesta:4 -258035.1795 -258035.8834 -258035.8834 0.0112 -4.4638 siesta:5 -258035.1793 -258035.4838 -258035.4838 0.0049 -4.4687 siesta:6 -258035.1789 -258035.2657 -258035.2657 0.0015 -4.4613 siesta:7 -258035.1787 -258035.1951 -258035.1951 0.0003 -4.4550 siesta:8 -258035.1786 -258035.1827 -258035.1827 0.0001 -4.4527 however after several CG steps, the dDMax of the second step starts shooting up: siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax Ef(eV) siesta:1 -258041.6771 -258049.1974 -258049.1974 0.0445 -4.3312 siesta:2 -414986.5091 -257706.6909 -257706.6909 7.6328 -4.9342 siesta:3 -258044.3070 -258045.0301 -258045.0301 0.0782 -3.7154 siesta:4 -258037.8846 -258045.8548 -258045.8548 0.0263 -4.5521 siesta:5 -258037.6817 -258045.1416 -258045.1416 0.0244 -4.5881 siesta:6 -258037.5376 -258043.5246 -258043.5246 0.0203 -4.6387 siesta:7 -258037.4613 -258039.8974 -258039.8974 0.0113 -4.5536 siesta:8 -258037.4461 -258039.1308 -258039.1308 0.0090 -4.5065 siesta:9 -258037.4415 -258038.3142 -258038.3142 0.0050 -4.4399 siesta: 10 -258037.4422 -258037.8985 -258037.8985 0.0029 -4.4292 siesta: 11 -258037.4395 -258037.4958 -258037.4958 0.0010 -4.4607 siesta: 12 -258037.4387 -258037.5226 -258037.5226 0.0006 -4.4739 siesta: 13 -258037.4384 -258037.5255 -258037.5255 0.0004 -4.4780 siesta: 14 -258037.4383 -258037.5025 -258037.5025 0.0003 -4.4779 siesta: 15 -258037.4383 -258037.4524 -258037.4524 0.0001 -4.4761 and eventually dDMax of the second step becomes so big that it can never converge: siesta:1 -258060.4975 -258056.7123 -258056.7123 0.1295 -4.4799 siesta:2 -1304652.4483 -252914.3188 -252914.3188 20.6318-40.9257 siesta:3 -270662.4603 -258022.4392 -258022.4392 1.3946 -2.5632 siesta:4 -262389.1412 -258028.2383 -258028.2383 1.0408 -2.9631 siesta:5 -261346.4355 -258032.8542 -258032.8542 0.9443 -3.0476 siesta:6 -261008.6664 -258033.7068 -258033.7068 0.9124 -3.0832 siesta:7 -1415808.5682 -248461.9222 -248461.9222 8.0613 siesta:8 -1329228.9170 -249101.5407 -249101.5407 7.8758 siesta:9 -1389456.5404 -248138.9810 -248138.9810 7.9741 siesta: 10 -1490261.9252 -246579.3815 -246579.3815 8.0538 siesta: 11 -1720098.0877 -242387.2899 -242387.2899 8.0561 siesta: 12 -6743529.762738225.975738225.9757 21.0844 siesta: 13 -3691005.7503 -74168.3252 -74168.3252 16.3687 siesta: 14 -2226132.5093 -141461.1037 -141461.1037 12.2798 siesta: 15 -1669147.7537 -173101.0090 -173101.0090 7.9487-79.2650 siesta: 16 -1938495.5805 -183836.1974 -183836.1974 18.7913 I wonder if anyone knows why this could happen, and how can you get around it. I use a Mixing parameter 0.08, 3 Pulay steps. I could set the mixing parameter to a very small number, but it will take many steps to get SCF converged... Thanks, Cherry Yahoo! oneSearch: Finally, mobile search that gives answers, not web links. http://mobile.yahoo.com/mobileweb/onesearch?refer=1ONXIC
Re: [SIESTA-L] Error in Cholesky factorisation in cdiag
First I compiled it with mkl and I got a lot of NAN in the output files. Then I changed to acml, and everything is fine. Cherry --- Sergey Lisenkov [EMAIL PROTECTED] wrote: Use acml library (you can download it from AMD website) instead of mkl. I found mkl library doesn't work at all for SIESTA 2.00. really? I cannot recall that I had any problems with mkl. sergey GOOD LUCK, Cherry --- bipul rakshit [EMAIL PROTECTED] wrote: hello siesta, I am doing calculation on beta Sn... but i get the Error in Cholesky factorisation in cdiag (see below) In the input file sn.fdf i attached. Also i already see the previous mails regarding this i use all those options in sn.fdf... but nothing works. also i use ifort complier and mkl libraries and running in serial mode and using FFLAGS= -static -static-libcxa -w -mp1 -tpp6 -O3 plz help me in this regard siesta: == Begin CG move = 0 == superc: Internal auxiliary supercell: 4 x 4 x 6 = 96 superc: Number of atoms, orbitals, and projectors:192 3072 3072 outcell: Unit cell vectors (Ang): 5.8197000.000.00 0.005.8197000.00 2.9098502.9098501.587614 outcell: Cell vector modules (Ang) :5.8197005.8197004.410779 outcell: Cell angles (23,13,12) (deg): 48.7220 48.7220 90. outcell: Cell volume (Ang**3): 53.7708 xijorb: WARNING: orbital pair 1 2739 is multiply connected InitMesh: MESH =24 x24 x18 = 10368 InitMesh: Mesh cutoff (required, used) =80.00086.477 Ry * Maximum dynamic memory allocated =18 MB Error in Cholesky factorisation in cdiag Stopping Program from Node:0 forrtl: severe (174): SIGSEGV, segmentation fault occurred Image PCRoutineLineSource siesta 081B4A41 Unknown Unknown Unknown siesta 08139EF8 Unknown Unknown Unknown siesta 080DD29B Unknown Unknown Unknown siesta 08074FB9 Unknown Unknown Unknown siesta 08068E75 Unknown Unknown Unknown siesta 080FAB89 Unknown Unknown Unknown siesta 0804A9A6 Unknown Unknown Unknown libc.so.6 4961EF70 Unknown Unknown Unknown siesta 0804A8B1 Unknown Unknown Unknown - Did you know? You can CHAT without downloading messenger. Click here Finding fabulous fares is fun. Let Yahoo! FareChase search your favorite travel sites to find flight and hotel bargains. http://farechase.yahoo.com/promo-generic-14795097 Get the free Yahoo! toolbar and rest assured with the added security of spyware protection. http://new.toolbar.yahoo.com/toolbar/features/norton/index.php
