Hi, Sometimes if you compile on some architechures with the fast and optimization flags these bizzare problems can pop up. I haven't seen this problem but similar errors of this type have occured when those flags are set. If this has already been mentioned, I apologize for repeating it.
Sincerely, -Kirk Bevan Quoting Demetra Psiachos <[EMAIL PROTECTED]>: > Hi, > > I had the same problem...I posted it in April 2006 and it still is not > resolved. > Perhaps I'll post it again. n.b. I used the same architecture and > compiler as Hai Ping below but I didn't try version 1.3. > > The option Diag.DivideAndConquer .false. > did not work in my case (nor did any of the other things I tried). > > Here's part of my old message to the group. > > > =================================================== > Hi, I am having a bizarre problem: Upon restarting an MD calculation, > specifically Harris Functional in my case, Siesta does about 5-10 MD steps > and then dies with the error: > > siesta: > Begin MD step = 12 > outcell: Unit cell vectors (Ang): > 20.121426 0.000000 0.000000 > 10.060713 17.425666 0.000000 > 0.000000 0.000000 18.521195 > outcell: Cell vector modules (Ang) : 20.121426 20.121426 18.521195 > outcell: Cell angles (23,13,12) (deg): 90.0000 90.0000 60.0000 > outcell: Cell volume (Ang**3) : 6494.0729 > > Error in Cholesky factorisation in rdiag > Stopping Program from Node: 0 > Error in Cholesky factorisation in rdiag > Stopping Program from Node: 1 > Error in Cholesky factorisation in rdiag > > > There is no problem when the run > is the initial one (i.e. UseSaveData .false.) but when UseSaveData is set > to .true. I get this problem. I believe it has to do with scalapack, as > the specific routine that fails is pdpotrf which is in scalapack). I tried > changing the number of processors and the blocksize but the problem > remained. What does not make any sense at all is why this problem would > appear ONLY upon restarting a calculation. > > Does anyone have an idea that I can try that might help? > Many Thanks, > > Demetra Psiachos > =========================================================================== > > > > > > > > > On Mon, 26 Jun 2006, Javier Junquera wrote: > > > Dear lan: > > > > In Siesta 2.0, try to switch the Divide And Conquer solver off: > > > > Diag.DivideAndConquer .false. > > > > Hope this helps, > > > > Javier > > > >> Dear developers : > >> I have come to errors in performing Pt-bulk CG-relaxation > >> calculation both in siesta-2.0 and 1.3 . > >> Would you please give me some hints ? > >> The error occured in first CG step , and the code gave this > >> information : > >> * Error in Cholesky factorisation in cdiag > >> Stopping Program from Node: 0 > >> Aborted > >> * The siesta-2.0 code was compiled serially in AMD-operon workstation > >> with PGI6.0 compilor , and the arch.make was set up just with *configure > >> *script . > >> The same system also performed in siesta-1.3 , and errors the code > > >> giving : > >> *siesta: iscf Eharris(eV) E_KS(eV) FreeEng(eV) dDmax > Ef(eV) > >> siesta: 1 -3087.9064 -3080.0184 -3080.0184 0.8488 12.8295 > >> timer: Routine,Calls,Time,% = IterSCF 1 15.810 91.39 > >> siesta: 2 -3085.8326 -3075.6683 -3075.6683 0.6424 11.3072 > >> siesta: 3 -3082.6427 -3077.6962 -3077.6962 0.3645 11.6410 > >> siesta: 4 -3081.6698 -3078.7760 -3078.7760 0.2053 11.8195 > >> siesta: 5 -3081.3761 -3079.4520 -3079.4520 0.1157 11.9159 > >> siesta: 6 -3081.2881 -3079.9189 -3079.9189 0.0654 11.9684 > >> siesta: 7 -3081.2616 -3080.2569 -3080.2569 0.0373 11.9971 > >> siesta: 8 -3081.2537 -3080.5068 -3080.5068 0.0216 12.0127 > >> Terminating due to failed diagonalisation > >> Stopping Program* > >> siesta-1.3 was compiled parallelly with PGI 6.0 compilor and acml-3.0 > >> mathlib , and the compiling flag is > >> FFLAGS= -O3 -Munroll -Mnoframe -tp k8-64 -Mcache_align -Mnolre . > >> Would you please help me ? Thank you in advance > >> Regards, > >> Hai.Ping > >> > >> > > > > > > > > >
