From audrius.alkauskas at unibas.ch Mon Jul 5 15:24:31 2004 From: audrius.alkauskas at unibas.ch (Audrius Alkauskas) Date: Mon, 05 Jul 2004 15:24:31 +0200 Subject: [CPMD-list] Line minimization? Message-ID: <40E9568F.4040205@unibas.ch> Hello There is something I don't get about geometry optimization. In conjugate-gradients - does CPMD perform line minimization after calculating new search direction or just does a step in that direction, proportional to the new gradient? Also, in quasi-newton methods? I have sometimes troubles in geometry optimization, and I have to switch from one method to another depending on the stage, but I could not find an answer to my question in the CPMD manual. Audrius From ZRLSRB at ch.ibm.com Mon Jul 5 16:39:55 2004 From: ZRLSRB at ch.ibm.com (Salomon Billeter) Date: Mon, 5 Jul 2004 16:39:55 +0200 Subject: [CPMD-list] Line minimization? Message-ID: Hi, The conjugate gradient and steepest descent optimizers (and, if specified to do so, the L-BFGS optimizer) perform a line search for locating the minimizer along the current search direction. The implementations of the DIIS/BFGS and BFGS methods don't do this. Instead, they use the predicted step as long as it does not exceed some criteria. This saves the extra energy/gradient evaluations involved in the line search but sometimes causes these optimizers to get caught in some higher-energy region of the configurational space, requiring the user to restart and/or to switch to another algorithm. The default of the implementation of the L-BFGS method tries to predict a trust region based on the previous progress and the predicted step length. Based on the success (energy decrease, Wolfe conditions) of the attempted step, it either rejects it (and reduces the trust radius) or accepts it (and keeps or increases the maximum trust radius for the next step). This avoids the extra energy/gradient evaluations involved in the line search but ensures comparable step sizes and therefore stability. The parameters in this implementation have been chosen to optimize the tradeoff between the number of rejected steps and as large a step as possible. Given the anharmonicity of the potential energy surface, unavoidable numerical errors, and the error due to the exchange-correlation functional evaluated on a finite real-space mesh, and since in the Newton-Raphson step, written in the basis of Hessian eigenmodes, the softest modes often dominate the error, the Newton-Raphson step length alone can be dramatically wrong, particularly when using an extrapolated inverse Hessian (e.g. using the BFGS formula). Therefore, the adapted trust radius or the line search will make the optimization much more robust. Best regards, Salomon From csimon at ccr.jussieu.fr Tue Jul 6 15:07:50 2004 From: csimon at ccr.jussieu.fr (Christian Simon) Date: Tue, 6 Jul 2004 15:07:50 +0200 Subject: [CPMD-list] MacOSX-G5: solutions for serial compilation Message-ID: <7BBE7E4B-CF4D-11D8-A9B1-000393B3D8A0@ccr.jussieu.fr> Dear friends and colleagues, I have now a quite clean procedure to compile cpmd-v3.9.1 on Macs. This includes tricks with the filesystem, the configure file, and some modifications of the source. For the moment, this works fine with one node (bi-pro). Before solving the parallel issues, or addressing the compilation of the contributions, I would like to test the current executable obtained, and check accuracy and reliability. Are there some typical tests you can suggest ? (among Mac owner any volunteer to help ?) I also would like to have some feedback from the team about the consistency of the directives I wrote, and suggestions if there is a more elegant way to code... Unfortunately I have no web/ftp server were I could store all the things I have done... Many thanks to all people how helped me ! Sincerely, -- Dr. Christian SIMON, Maitre de Conferences Laboratoire LI2C-UMR7612 Bat. F74, piece 757 Universite Pierre et Marie Curie Tel:+33.1.44.27.32.65 Case 51 Fax:+33.1.44.27.32.28 4 Place Jussieu 75252 Paris Cedex 05 FRANCE/Europe From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Tue Jul 6 15:24:15 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel Kohlmeyer) Date: Tue, 06 Jul 2004 15:24:15 +0200 Subject: [CPMD-list] MacOSX-G5: solutions for serial compilation In-Reply-To: Your message of "Tue, 06 Jul 2004 15:07:50 +0200." <7BBE7E4B-CF4D-11D8-A9B1-000393B3D8A0@ccr.jussieu.fr> Message-ID: <200407061324.i66DOF621051@yello.theochem.ruhr-uni-bochum.de> >>> "CS" == Christian Simon writes: CS> Dear friends and colleagues, dear christian, CS> I have now a quite clean procedure to compile cpmd-v3.9.1 on Macs. CS> This includes tricks with the filesystem, the configure file, and some CS> modifications of the source. For the moment, this works fine with one CS> node (bi-pro). CS> Before solving the parallel issues, or addressing the compilation of the CS> contributions, I would like to test the current executable obtained, and CS> check accuracy and reliability. CS> Are there some typical tests you can suggest ? CS> (among Mac owner any volunteer to help ?) well a good starting point would be the cpmd-test distribution. some of the examples need quite a bit of cpu power/memory or a parallel machine, though. the advantage is, that you have reference outputs. please note, that the absolute value for the total energy has changed between version 3.7 and 3.9, so you can only compare geometries or relative energies. CS> I also would like to have some feedback from the team about CS> the consistency of the directives I wrote, and suggestions if there is CS> a more elegant way to code... CS> Unfortunately I have no web/ftp server were I could store CS> all the things I have done... there should several ways to deal with that. i'll contact you personally on that one. regards, axel. CS> Many thanks to all people how helped me ! CS> Sincerely, CS> -- CS> Dr. Christian SIMON, Maitre de Conferences CS> Laboratoire LI2C-UMR7612 Bat. F74, piece 757 CS> Universite Pierre et Marie Curie Tel:+33.1.44.27.32.65 CS> Case 51 Fax:+33.1.44.27.32.28 CS> 4 Place Jussieu CS> 75252 Paris Cedex 05 CS> FRANCE/Europe CS> _______________________________________________ CS> CPMD-list mailing list CS> CPMD-list at cpmd.org CS> http://cpmd.org/mailman/listinfo/cpmd-list -- ======================================================================= Axel Kohlmeyer e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673 Ruhr-Universitaet Bochum - NC 03/53 Fax: ++49 (0)234/32-14045 D-44780 Bochum http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/ ======================================================================= If you make something idiot-proof, the universe creates a better idiot. From t.orban at csuohio.edu Tue Jul 6 15:50:22 2004 From: t.orban at csuohio.edu (Tivadar Orban) Date: Tue, 06 Jul 2004 09:50:22 -0400 Subject: [CPMD-list] MacOSX-G5: solutions for serial compilation In-Reply-To: <7BBE7E4B-CF4D-11D8-A9B1-000393B3D8A0@ccr.jussieu.fr> Message-ID: On 7/6/04 9:07 AM, "Christian Simon" wrote: > Dear friends and colleagues, > > I have now a quite clean procedure to compile cpmd-v3.9.1 on Macs. > This includes tricks with the filesystem, the configure file, and some > modifications of the source. For the moment, this works fine with one > node (bi-pro). > > Before solving the parallel issues, or addressing the compilation of the > contributions, I would like to test the current executable obtained, and > check accuracy and reliability. > > Are there some typical tests you can suggest ? > (among Mac owner any volunteer to help ?) This is great news!!! I am willing to help on testing or anything else I could help you with I have a compiled 3.5 version that has run time errors (as earlier described by Defosse Gregory). The same approach does not work in my case for version 3.7.2 though Regards, Tivadar Orban > I also would like to have some feedback from the team about > the consistency of the directives I wrote, and suggestions if there is > a more elegant way to code... > > Unfortunately I have no web/ftp server were I could store > all the things I have done... > Many thanks to all people how helped me ! > Sincerely, > -- > Dr. Christian SIMON, Maitre de Conferences > Laboratoire LI2C-UMR7612 Bat. F74, piece 757 > Universite Pierre et Marie Curie Tel:+33.1.44.27.32.65 > Case 51 Fax:+33.1.44.27.32.28 > 4 Place Jussieu > 75252 Paris Cedex 05 > FRANCE/Europe > > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list From proffess at yandex.ru Tue Jul 6 18:50:21 2004 From: proffess at yandex.ru (Sergei Lisenkov) Date: Tue, 6 Jul 2004 20:50:21 +0400 (MSD) Subject: [CPMD-list] MPI or OpenMP In-Reply-To: References: Message-ID: <40EAD84D.000001.04238@ariel.yandex.ru> Dear CPMD users/authors, I do my calculations on 64 CPU HP SuperDome machine using MPI version. Because that machine is SMP, I am wondering: what is the best - to use mpi version or use OpenMP versions or use mixed MPI/OpenMP ? Thank you very much, Best wishes, Sergey From hutter at pci.unizh.ch Thu Jul 8 10:55:37 2004 From: hutter at pci.unizh.ch (Juerg Hutter) Date: Thu, 8 Jul 2004 10:55:37 +0200 (MEST) Subject: [CPMD-list] MPI or OpenMP In-Reply-To: <40EAD84D.000001.04238@ariel.yandex.ru> References: <40EAD84D.000001.04238@ariel.yandex.ru> Message-ID: Hi > Dear CPMD users/authors, > > I do my calculations on 64 CPU HP SuperDome machine using MPI version. Because that machine is SMP, I am wondering: what is the best - to use mpi version or use OpenMP versions or use mixed MPI/OpenMP ? The optimal parallel setup will depend on the system you want to calculate. However, there are some general rules that apply. SMP performance levels off between 4 and 8 CPUs, on some machines even between 2 and 4. MPI performance is superior usually up to N/2, where N is the number of grid points in x direction. For most systems and cutoffs N will be between 64 and 256. Therefore, in your case the MPI version will be most likely most efficient. Juerg Hutter > > Thank you very much, > Best wishes, > Sergey > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list > From eunggun.kim at chemistry.gatech.edu Fri Jul 9 23:54:35 2004 From: eunggun.kim at chemistry.gatech.edu (Eung-Gun Kim) Date: Fri, 9 Jul 2004 17:54:35 -0400 Subject: [CPMD-list] Error with UPF Message-ID: <1089410075.40ef141bab19b@webmail.mail.gatech.edu> Dear List Subscribers, I am trying to use a pseudo in UPF (universal pseudopotential format) with CPMD 3.9.1. I obtained this UPF file from 'Siesta' and this pseudo file seems to work fine with another plane-wave code 'PWSCF.' But when I use it with CPMD, an error occurs with "UNKNOWN TYPE OF NLCC." The same error occurs with or without the NLCC keyword in the input. A short description of my input and output files follows: 1) Header of the UPF pseudo -------------------------------------------------- ATM3 no_date Troullier-Martins 0 Version Number Al Element NC Norm - Conserving pseudopotential T Nonlinear Core Correction SLA PZ NOGX NOGC PZ Exchange-Correlation functional 3 Z valence 0 Total energy 0.0000000 0.0000000 Suggested cutoff for wfc and rho 3 Max angular momentum component 1068 Number of points in mesh 4 4 Number of Wavefunctions, Number of Projectors Wavefunctions nl l occ 3S 0 2 3P 1 1 3D 2 0 4F 3 0 2) A portion of the input ---------------------------------------------------- &DFT FUNCTIONAL LDA LDA CORRELATION PZ &END . . . &ATOMS *al_pz.upf UPF NLCC Kleinman-Bylander RAGGIO=1.0 LMAX=3 . . 3) A portion of the output --------------------------------------------------- . . ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ al_pz.upf ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ATM3 no_date Troullier-Martins ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ **************************************************************** * ATOM MASS RAGGIO NLCC PSEUDOPOTENTIAL * * Al 26.9815 1.0000 YES UPF NC 4 PROJECTORS * **************************************************************** . . . UNKNOWN TYPE OF NLCC PROGRAM STOPS IN SUBROUTINE NLCCSET| [PROC= 0] p0_22710: p4_error: : 999 --------------------------------------------------------------------------- Thanks very much for reading. Best regards, E. Kim From proffess at yandex.ru Sat Jul 10 08:02:44 2004 From: proffess at yandex.ru (Sergei Lisenkov) Date: Sat, 10 Jul 2004 10:02:44 +0400 (MSD) Subject: [CPMD-list] Slow geometry optimization In-Reply-To: References: Message-ID: <40EF8684.000010.28239@colgate.yandex.ru> Dear CPMD users/authors, I am perfoming a geometry optimization of a cluster containing 96 atoms on parallel machine using 44 processors. I use a Conjugate gradient method and Goedecker pseudopotential. I ran this task two weeks ago and now I cannot reach the convergence. With another DFT code (PWSCF) it tooks a 2 days for full optimization. Could anybody give me some descriptions? Thank you very much, Best wishes, Sergey From dogbe at unr.nevada.edu Sat Jul 10 11:06:33 2004 From: dogbe at unr.nevada.edu (John Kofi Dogbe) Date: Sat, 10 Jul 2004 02:06:33 -0700 (PDT) Subject: [CPMD-list] WARNING! XC FUNCTIONALS ... Message-ID: Dear CPMD Users, Please can anyone tell me what the source of the following warning might be? And do I have to worry about it? At first, I thought it is a straight forward warning, but it appears not to be. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! WARNING! XC FUNCTIONALS INCONSISTENT FOR Si_MT_revPBE.psp !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! I suspected that it was coming from the following section of my input file: &DFT FUNCTIONAL REVPBE GRADIENT CORRECTION REVPBEX REVPBEC NEWCODE &END However, I can't get rid of it with using any other functional. In fact I thought the most appropriate functional for the revised PBE pseudopontentials should be what I was using. I guess I'm wrong given the warning. Any suggestions? Thank you. John -- **The first is not necessarily the Leader** "If I have spoken evil, bear witness of the evil: but if well, why smitest thou me?" -- Jesus Christ (John, 18:23) From hutter at pci.unizh.ch Sat Jul 10 16:30:37 2004 From: hutter at pci.unizh.ch (Juerg Hutter) Date: Sat, 10 Jul 2004 16:30:37 +0200 (MEST) Subject: [CPMD-list] Error with UPF In-Reply-To: <1089410075.40ef141bab19b@webmail.mail.gatech.edu> References: <1089410075.40ef141bab19b@webmail.mail.gatech.edu> Message-ID: Hi > > I am trying to use a pseudo in UPF (universal pseudopotential format) with CPMD > 3.9.1. I obtained this UPF file from 'Siesta' and this pseudo file seems to > work fine with another plane-wave code 'PWSCF.' But when I use it with CPMD, > an > error occurs with "UNKNOWN TYPE OF NLCC." The same error occurs with or > without > the NLCC keyword in the input. A short description of my input and output > files NLCC is set by the UPF pseudopotential file, it overrides any settings from the input file. Therefore you can't switch off NLCC in this case. Unfortunately, the output "UNKNOWN TYPE OF NLCC" means exactly what it says. This type of NLCC (meaning UPF) is currently not supported in CPMD. The only UPF PP supported are norm-conserving without NLCC. If you want to add UPF NLCC support to CPMD, you have to provide the Fourier transform code in nlccset.F for TYPE 3. reagrds Juerg Hutter > follows: > > 1) Header of the UPF pseudo > -------------------------------------------------- > > > ATM3 no_date Troullier-Martins > > > 0 Version Number > Al Element > NC Norm - Conserving pseudopotential > T Nonlinear Core Correction > SLA PZ NOGX NOGC PZ Exchange-Correlation functional > 3 Z valence > 0 Total energy > 0.0000000 0.0000000 Suggested cutoff for wfc and rho > 3 Max angular momentum component > 1068 Number of points in mesh > 4 4 Number of Wavefunctions, Number of Projectors > Wavefunctions nl l occ > 3S 0 2 > 3P 1 1 > 3D 2 0 > 4F 3 0 > > > 2) A portion of the input > ---------------------------------------------------- > > &DFT > FUNCTIONAL LDA > LDA CORRELATION PZ > &END > . > . > . > &ATOMS > *al_pz.upf UPF NLCC Kleinman-Bylander RAGGIO=1.0 > LMAX=3 > . > . > > 3) A portion of the output > --------------------------------------------------- > > . > . > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ > al_pz.upf > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ > ATM3 no_date Troullier-Martins > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ > > > **************************************************************** > * ATOM MASS RAGGIO NLCC PSEUDOPOTENTIAL * > * Al 26.9815 1.0000 YES UPF NC 4 PROJECTORS * > **************************************************************** > . > . > . > > UNKNOWN TYPE OF NLCC > > > > > PROGRAM STOPS IN SUBROUTINE NLCCSET| [PROC= 0] > p0_22710: p4_error: : 999 > > > --------------------------------------------------------------------------- > > Thanks very much for reading. > > Best regards, > > E. Kim > > > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list > From hutter at pci.unizh.ch Sat Jul 10 16:32:41 2004 From: hutter at pci.unizh.ch (Juerg Hutter) Date: Sat, 10 Jul 2004 16:32:41 +0200 (MEST) Subject: [CPMD-list] Slow geometry optimization In-Reply-To: <40EF8684.000010.28239@colgate.yandex.ru> References: <40EF8684.000010.28239@colgate.yandex.ru> Message-ID: Hi from your mail it's difficult to guess what causes the problems in optimization. However, geometry optimization has been discussed extensively on this list, and Salomon Billeter has given many hints how to improve convergence. Have a look at some of the older posts. regards Juerg Hutter ---------------------------------------------------------- Juerg Hutter Phone : ++41 1 635 4491 Physical Chemistry Institute FAX : ++41 1 635 6838 University of Zurich E-mail: hutter at pci.unizh.ch Winterthurerstrasse 190 CH-8057 Zurich, Switzerland ---------------------------------------------------------- On Sat, 10 Jul 2004, Sergei Lisenkov wrote: > Dear CPMD users/authors, > > I am perfoming a geometry optimization of a cluster containing 96 atoms on parallel machine using 44 processors. I use a Conjugate gradient method and Goedecker pseudopotential. I ran this task two weeks ago and now I cannot reach the convergence. With another DFT code (PWSCF) it tooks a 2 days for full optimization. Could anybody give me some descriptions? > > Thank you very much, > Best wishes, > Sergey > > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list > From hutter at pci.unizh.ch Sat Jul 10 16:35:40 2004 From: hutter at pci.unizh.ch (Juerg Hutter) Date: Sat, 10 Jul 2004 16:35:40 +0200 (MEST) Subject: [CPMD-list] WARNING! XC FUNCTIONALS ... In-Reply-To: References: Message-ID: Hi > Dear CPMD Users, > > Please can anyone tell me what the source of the following warning might > be? And do I have to worry about it? At first, I thought it is a straight > forward warning, but it appears not to be. > > !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! > WARNING! XC FUNCTIONALS INCONSISTENT FOR Si_MT_revPBE.psp > !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! CPMD checks the functionals used to generate the PP with the actual functional used in the calculation. The functional used in CPMD for the PBE family is not common as it uses the PADE form for the LDA correlation. This caused the warning but has no effect on the results. If you want to get rid of the warning replace the XC string in the PP file &ATOM section by "0944". regards Juerg Hutter > > I suspected that it was coming from the following section of my input > file: > > &DFT > FUNCTIONAL REVPBE > GRADIENT CORRECTION REVPBEX REVPBEC > NEWCODE > &END > > However, I can't get rid of it with using any other functional. In fact > I thought the most appropriate functional for the revised PBE > pseudopontentials should be what I was using. I guess I'm wrong given the > warning. Any suggestions? > > Thank you. > John > -- > **The first is not necessarily the Leader** > > "If I have spoken evil, bear witness of the evil: > but if well, why smitest thou me?" > -- Jesus Christ (John, 18:23) > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list > From Ari.P.Seitsonen at iki.fi Sat Jul 10 16:55:29 2004 From: Ari.P.Seitsonen at iki.fi (Ari P Seitsonen) Date: Sat, 10 Jul 2004 16:55:29 +0200 (CEST) Subject: [CPMD-list] Slow geometry optimization In-Reply-To: <40EF8684.000010.28239@colgate.yandex.ru> References: <40EF8684.000010.28239@colgate.yandex.ru> Message-ID: Dear Sergey, Do you mean that the i) electronic or ii) ionic structure does not converge? I'd use the "LFBGS" optimiser, it works best for my systems. If also this doesn't help, one can use ANNEALING (i.e. mocular dynamics). If these don't help you can send us the input file and we can check if there's something non-optimal. Greetings from Cagliari/Sardinia, apsi -=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=- Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/ Tel +41 1 635 44 97 / Fax +41 1 635 68 38 / GSM +41 79 719 09 35 Anschrift: Physikalisch Chemisches Institut (PCI), Universit?t Z?rich (UniZh) Indirizzo: Winterthurerstrasse 190, CH-8057 Z?rich Address: Schweiz / Svizzera / Suisse / Svizra / Switzerland On Sat, 10 Jul 2004, Sergei Lisenkov wrote: > Dear CPMD users/authors, > > I am perfoming a geometry optimization of a cluster containing 96 atoms on parallel machine using 44 processors. I use a Conjugate gradient method and Goedecker pseudopotential. I ran this task two weeks ago and now I cannot reach the convergence. With another DFT code (PWSCF) it tooks a 2 days for full optimization. Could anybody give me some descriptions? > > Thank you very much, > Best wishes, > Sergey > > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list > From audrius.alkauskas at unibas.ch Sun Jul 11 02:58:52 2004 From: audrius.alkauskas at unibas.ch (Audrius Alkauskas) Date: Sun, 11 Jul 2004 02:58:52 +0200 Subject: [CPMD-list] Slow geometry optimization Message-ID: <40F090CC.6000405@unibas.ch> Hello I thank /Salomon Billeter/ for nice explanation. I just recently tried LBFGS and found a Comp. Mat. Sci. publication on that. It looks promising - no unexpected jumps, everything's runs smoothly. However, I think it's always good (but that I understood too late) to save all the geometry files in the process of optimization. It's especially important when one tries to beat the third decimal digit in the total energy (as in my case). Audrius From eunggun.kim at chemistry.gatech.edu Sun Jul 11 05:20:57 2004 From: eunggun.kim at chemistry.gatech.edu (Eung-Gun Kim) Date: Sat, 10 Jul 2004 23:20:57 -0400 Subject: [CPMD-list] Error with UPF In-Reply-To: References: <1089410075.40ef141bab19b@webmail.mail.gatech.edu> Message-ID: <1089516057.40f0b219498e0@webmail.mail.gatech.edu> Dear Prof. Hutter, Thanks very much for your kind reply. Best regards, E. Kim Quoting Juerg Hutter : > Hi > > > > > I am trying to use a pseudo in UPF (universal pseudopotential format) with > CPMD > > 3.9.1. I obtained this UPF file from 'Siesta' and this pseudo file seems > to > > work fine with another plane-wave code 'PWSCF.' But when I use it with > CPMD, > > an > > error occurs with "UNKNOWN TYPE OF NLCC." The same error occurs with or > > without > > the NLCC keyword in the input. A short description of my input and output > > files > > NLCC is set by the UPF pseudopotential file, it overrides any settings > from the input file. Therefore you can't switch off NLCC in this case. > Unfortunately, the output "UNKNOWN TYPE OF NLCC" means exactly what it > says. This type of NLCC (meaning UPF) is currently not supported in CPMD. > The only UPF PP supported are norm-conserving without NLCC. > > If you want to add UPF NLCC support to CPMD, you have to provide > the Fourier transform code in nlccset.F for TYPE 3. > > reagrds > > Juerg Hutter > > > > > > follows: > > > > 1) Header of the UPF pseudo > > -------------------------------------------------- > > > > > > ATM3 no_date Troullier-Martins > > > > > > 0 Version Number > > Al Element > > NC Norm - Conserving pseudopotential > > T Nonlinear Core Correction > > SLA PZ NOGX NOGC PZ Exchange-Correlation functional > > 3 Z valence > > 0 Total energy > > 0.0000000 0.0000000 Suggested cutoff for wfc and rho > > 3 Max angular momentum component > > 1068 Number of points in mesh > > 4 4 Number of Wavefunctions, Number of Projectors > > Wavefunctions nl l occ > > 3S 0 2 > > 3P 1 1 > > 3D 2 0 > > 4F 3 0 > > > > > > 2) A portion of the input > > ---------------------------------------------------- > > > > &DFT > > FUNCTIONAL LDA > > LDA CORRELATION PZ > > &END > > . > > . > > . > > &ATOMS > > *al_pz.upf UPF NLCC Kleinman-Bylander RAGGIO=1.0 > > LMAX=3 > > . > > . > > > > 3) A portion of the output > > --------------------------------------------------- > > > > . > > . > > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ > > al_pz.upf > > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ > > ATM3 no_date Troullier-Martins > > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ > > > > > > **************************************************************** > > * ATOM MASS RAGGIO NLCC PSEUDOPOTENTIAL * > > * Al 26.9815 1.0000 YES UPF NC 4 PROJECTORS * > > **************************************************************** > > . > > . > > . > > > > UNKNOWN TYPE OF NLCC > > > > > > > > > > PROGRAM STOPS IN SUBROUTINE NLCCSET| [PROC= 0] > > p0_22710: p4_error: : 999 > > > > > > > --------------------------------------------------------------------------- > > > > Thanks very much for reading. > > > > Best regards, > > > > E. Kim > > > > > > _______________________________________________ > > CPMD-list mailing list > > CPMD-list at cpmd.org > > http://cpmd.org/mailman/listinfo/cpmd-list > > > From apu at veccal.ernet.in Sun Jul 11 12:25:29 2004 From: apu at veccal.ernet.in (apu at veccal.ernet.in) Date: Sun, 11 Jul 2004 15:25:29 +0500 Subject: [CPMD-list] CPMD-3.7.2 hurdle Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040711/5d3ee3a8/attachment.html From apu at veccal.ernet.in Sun Jul 11 12:27:39 2004 From: apu at veccal.ernet.in (apu at veccal.ernet.in) Date: Sun, 11 Jul 2004 15:27:39 +0500 Subject: [CPMD-list] CPMD-3.7.2 hurdle Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040711/148f919c/attachment.html From bala at jncasr.ac.in Mon Jul 12 07:54:58 2004 From: bala at jncasr.ac.in (Dr. Balasubramanian Sundaram) Date: Mon, 12 Jul 2004 11:24:58 +0530 Subject: [CPMD-list] CPMD-3.7.2 hurdle References: Message-ID: <40F227B2.F0647B34@jncasr.ac.in> Hi Apu, Here is a Makefile (serial version) that works fine on our Compaq (DEC Alpha) (HP now). Is your machine an Alpha or the Compaq SC80? I have a suspicion that it could be the former. Anyway, here is our config: Machine : Compaq Alpha ES40. Operating System : OSF1 V5.1 Compiler : Compaq Fortran V5.4A-1472 Library : DXML v3.4 ---------------Makefile for cpmd v. 3.7.2 ------------------------------- SRC = . DEST = . BIN = . #QMMM_FLAGS = -D__QMECHCOUPL #QMMM_LIBS = -L. -lmm FFLAGS = -align dcommons -convert big_endian LFLAGS = -ldxml $(QMMM_LIBS) CFLAGS = -D__alpha CPP = cpp -P CPPFLAGS = -C -D__alpha -DPOINTER8 -D__DERF -DLAPACK -DFFT_DEFAULT NOOPT_FLAG = CC = cc -c -O FC = f90 -O -c LD = f90 -O AR = /usr/bin/ar -r --------------------------------------------------------------------------- 1) By the way, it may be a good idea to move to CPMD v.3.9.1, unless you have a requirement to stick to v.3.7.2. 2) In case of compilation issues, it may be a good idea to send the first few lines of your Makefile also. Best Wishes, Bala http://www.jncasr.ac.in/bala apu at veccal.ernet.in wrote: > > > > Dear All, > > I am trying to configure CPMD-3.7.2 in a COMPAQ unix platform.After > using the following path: > > ./Configure > ./Configure COMPAQ-SC80> Makefile > vi Makefile > make > I am getting comment at the end: > > ld: > > Can't locate file for: > -lelan > > f90:Severe:Failed while > trying to Link. > > *** Exit 1 > > Stop. > > > > Please suggest what I should do now. > > Thank in advance. > > Regards, > > Apu > > > > **************************************************** > Apu Sarkar > Research Fellow > Variable Energy Cyclotron Centre > Kolkata 700 064 > phone: 91-33-2337-1230 (extn. 3190) > Fax: 91-33-2334-6871 > INDIA > **************************************************** > > --------------------------------------------------------------- > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list From proffess at yandex.ru Tue Jul 13 15:52:07 2004 From: proffess at yandex.ru (Sergei Lisenkov) Date: Tue, 13 Jul 2004 17:52:07 +0400 (MSD) Subject: [CPMD-list] Slow geometry optimization In-Reply-To: References: Message-ID: <40F3E907.00000B.14966@colgate.yandex.ru> Dear CPMD users/authors, I use a LBFGS method for optimizing geometry of my cluster. In output file I see a follow warnings: ..... ODIIS| Insufficient progress; reset! 59 2.839E-05 1.553E-07 -828.863287 -3.418E-07 51.08 60 5.096E-05 1.365E-07 -828.863287 -9.460E-08 51.73 61 1.236E-05 1.360E-07 -828.863287 -3.734E-08 52.48 62 1.236E-05 1.279E-07 -828.863287 -4.681E-08 53.13 63 1.235E-05 1.251E-07 -828.863287 -2.473E-08 52.91 64 1.261E-05 1.321E-07 -828.863287 -9.653E-08 53.28 65 1.241E-05 1.299E-07 -828.863287 1.575E-07 53.62 66 1.235E-05 1.262E-07 -828.863287 -3.323E-08 54.42 67 1.231E-05 1.526E-07 -828.863287 -1.783E-07 54.49 68 1.237E-05 1.279E-07 -828.863287 1.996E-07 54.47 69 1.235E-05 1.246E-07 -828.863287 -5.295E-08 54.94 70 1.232E-05 1.315E-07 -828.863287 6.441E-08 55.15 ODIIS| Insufficient progress; reset! 71 1.235E-05 1.218E-07 -828.863287 -1.055E-07 51.15 72 1.844E-05 1.194E-07 -828.863287 -8.023E-08 51.76 73 1.218E-05 1.190E-07 -828.863287 -3.826E-08 52.80 74 1.215E-05 1.175E-07 -828.863287 -4.495E-08 52.60 75 1.203E-05 1.178E-07 -828.863287 -7.415E-08 52.93 76 1.318E-05 1.600E-07 -828.863287 5.527E-07 53.22 77 1.232E-05 1.212E-07 -828.863287 -3.498E-07 54.15 78 1.232E-05 1.211E-07 -828.863287 2.053E-09 54.05 79 1.229E-05 1.219E-07 -828.863287 -3.314E-08 55.21 80 1.226E-05 1.218E-07 -828.863287 -1.651E-08 55.93 81 1.344E-04 1.696E-06 -828.863281 6.555E-06 55.08 ODIIS| Insufficient progress; reset! 82 1.232E-05 1.232E-07 -828.863287 -6.519E-06 51.19 83 1.222E-05 1.157E-07 -828.863287 -8.527E-08 51.94 84 1.218E-05 1.154E-07 -828.863287 -5.031E-08 52.18 85 1.210E-05 1.159E-07 -828.863287 -8.005E-08 53.32 86 1.197E-05 1.162E-07 -828.863287 -1.219E-07 52.82 87 1.179E-05 1.159E-07 -828.863288 -1.588E-07 54.26 88 1.182E-05 1.155E-07 -828.863288 -1.892E-07 53.96 89 1.153E-05 1.155E-07 -828.863288 -2.704E-07 55.06 90 1.100E-05 1.163E-07 -828.863288 -4.140E-07 55.13 91 1.033E-05 1.244E-07 -828.863289 -6.197E-07 55.22 92 9.193E-06 1.462E-07 -828.863290 -9.055E-07 55.58 RESTART INFORMATION WRITTEN ON FILE ./RESTART.1 ...... Is it normal? Or should I do anything? Thank you for help, Best, Sergey From Ari.P.Seitsonen at iki.fi Tue Jul 13 16:31:05 2004 From: Ari.P.Seitsonen at iki.fi (Ari P Seitsonen) Date: Tue, 13 Jul 2004 16:31:05 +0200 (CEST) Subject: [CPMD-list] Slow geometry optimization In-Reply-To: <40F3E907.00000B.14966@colgate.yandex.ru> References: <40F3E907.00000B.14966@colgate.yandex.ru> Message-ID: Dear Sergey, Please try 'PCG MINIMIZE' for the optimisation of the wave functions. Or release a bit the convergence criterion for the wave functions (default 1.0E-5, your calculation has reached 1.3E-5, so I'd try e.g. 2.0E-5 and then increase it, or use the keyword 'CONVERGENCE ADAPT'). Greetings, apsi -=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=- Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/ Tel +41 1 635 44 97 / Fax +41 1 635 68 38 / GSM +41 79 719 09 35 Anschrift: Physikalisch Chemisches Institut (PCI), Universit?t Z?rich (UniZh) Indirizzo: Winterthurerstrasse 190, CH-8057 Z?rich Address: Schweiz / Svizzera / Suisse / Svizra / Switzerland On Tue, 13 Jul 2004, Sergei Lisenkov wrote: > Dear CPMD users/authors, > > I use a LBFGS method for optimizing geometry of my cluster. In output file I see a follow warnings: > > ..... > ODIIS| Insufficient progress; reset! > 59 2.839E-05 1.553E-07 -828.863287 -3.418E-07 51.08 > 60 5.096E-05 1.365E-07 -828.863287 -9.460E-08 51.73 > 61 1.236E-05 1.360E-07 -828.863287 -3.734E-08 52.48 > 62 1.236E-05 1.279E-07 -828.863287 -4.681E-08 53.13 > 63 1.235E-05 1.251E-07 -828.863287 -2.473E-08 52.91 > 64 1.261E-05 1.321E-07 -828.863287 -9.653E-08 53.28 > 65 1.241E-05 1.299E-07 -828.863287 1.575E-07 53.62 > 66 1.235E-05 1.262E-07 -828.863287 -3.323E-08 54.42 > 67 1.231E-05 1.526E-07 -828.863287 -1.783E-07 54.49 > 68 1.237E-05 1.279E-07 -828.863287 1.996E-07 54.47 > 69 1.235E-05 1.246E-07 -828.863287 -5.295E-08 54.94 > 70 1.232E-05 1.315E-07 -828.863287 6.441E-08 55.15 > ODIIS| Insufficient progress; reset! > 71 1.235E-05 1.218E-07 -828.863287 -1.055E-07 51.15 > 72 1.844E-05 1.194E-07 -828.863287 -8.023E-08 51.76 > 73 1.218E-05 1.190E-07 -828.863287 -3.826E-08 52.80 > 74 1.215E-05 1.175E-07 -828.863287 -4.495E-08 52.60 > 75 1.203E-05 1.178E-07 -828.863287 -7.415E-08 52.93 > 76 1.318E-05 1.600E-07 -828.863287 5.527E-07 53.22 > 77 1.232E-05 1.212E-07 -828.863287 -3.498E-07 54.15 > 78 1.232E-05 1.211E-07 -828.863287 2.053E-09 54.05 > 79 1.229E-05 1.219E-07 -828.863287 -3.314E-08 55.21 > 80 1.226E-05 1.218E-07 -828.863287 -1.651E-08 55.93 > 81 1.344E-04 1.696E-06 -828.863281 6.555E-06 55.08 > ODIIS| Insufficient progress; reset! > 82 1.232E-05 1.232E-07 -828.863287 -6.519E-06 51.19 > 83 1.222E-05 1.157E-07 -828.863287 -8.527E-08 51.94 > 84 1.218E-05 1.154E-07 -828.863287 -5.031E-08 52.18 > 85 1.210E-05 1.159E-07 -828.863287 -8.005E-08 53.32 > 86 1.197E-05 1.162E-07 -828.863287 -1.219E-07 52.82 > 87 1.179E-05 1.159E-07 -828.863288 -1.588E-07 54.26 > 88 1.182E-05 1.155E-07 -828.863288 -1.892E-07 53.96 > 89 1.153E-05 1.155E-07 -828.863288 -2.704E-07 55.06 > 90 1.100E-05 1.163E-07 -828.863288 -4.140E-07 55.13 > 91 1.033E-05 1.244E-07 -828.863289 -6.197E-07 55.22 > 92 9.193E-06 1.462E-07 -828.863290 -9.055E-07 55.58 > > RESTART INFORMATION WRITTEN ON FILE ./RESTART.1 > ...... > > Is it normal? Or should I do anything? > > Thank you for help, > Best, > Sergey > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list > From knba713 at hotmail.com Tue Jul 13 17:09:52 2004 From: knba713 at hotmail.com (Xudong Tang) Date: Tue, 13 Jul 2004 11:09:52 -0400 Subject: [CPMD-list] Total energy problem of 2 Si atoms system Message-ID: hi, CPMD guys, I used CPMD on a 2 Si atoms system. Total energy variation has been calculated by changing the distance between 2 Si atoms. In most area, the curve looks smooth and nice, but there is a weird peak happening where their distance is around 2.1 nm. Please check the attachment. Anybody here has an idea about what is the wrong with my calculation? Xudong my input file: &CPMD OPTIMIZE WAVEFUNCTION MAXSTEP 100000 ODIIS 15 &END &SYSTEM SYMMETRY 8 CELL 20.0 1.5 1.0 0.0 0.0 0.0 CUTOFF 30 ANGSTROMS DUAL 3 TESR 2 2 2 &END &ATOMS *SI_SGS KLEINMAN-BYLANDER RAGGIO=1.0 LMAX=P LOC=P 2 0.0000 0.0000 0.0000 0.0000 2.0951 0.0000 CONSTRAINTS FIX ATOMS 2 1 2 END CONSTRAINTS &END &DFT FUNCTIONAL LDA &END _________________________________________________________________ FREE pop-up blocking with the new MSN Toolbar ? get it now! http://toolbar.msn.click-url.com/go/onm00200415ave/direct/01/ -------------- next part -------------- A non-text attachment was scrubbed... Name: Potential function.xls Type: application/vnd.ms-excel Size: 33280 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040713/b959c081/attachment.xlb From ct at chemie.hu-berlin.de Tue Jul 13 17:34:35 2004 From: ct at chemie.hu-berlin.de (Christian Tuma) Date: Tue, 13 Jul 2004 17:34:35 +0200 Subject: [CPMD-list] Slow geometry optimization In-Reply-To: <40F3E907.00000B.14966@colgate.yandex.ru> References: <40F3E907.00000B.14966@colgate.yandex.ru> Message-ID: <20040713153435.GG451677@uranus.chemie.hu-berlin.de> Hello Sergei, this has nothing to do with geometry optimisations. It shows the progress of the wavefunction optimisation using the ODIIS scheme (electronic problem). Such resets may happen using this scheme. You can try other schemes and compare the progress, results, timings, etc.. This can be useful, because sometimes ODIIS gets stuck in a wrong electronic state. You can also try to change some options related to ODIIS, please read the manual and previous postings in this mailing list. good luck, Christian. On Tue, Jul 13, 2004 at 05:52:07PM +0400, Sergei Lisenkov wrote: > Dear CPMD users/authors, > > I use a LBFGS method for optimizing geometry of my cluster. In output file I see a follow warnings: > > ..... > ODIIS| Insufficient progress; reset! > 59 2.839E-05 1.553E-07 -828.863287 -3.418E-07 51.08 > 60 5.096E-05 1.365E-07 -828.863287 -9.460E-08 51.73 > 61 1.236E-05 1.360E-07 -828.863287 -3.734E-08 52.48 > 62 1.236E-05 1.279E-07 -828.863287 -4.681E-08 53.13 > 63 1.235E-05 1.251E-07 -828.863287 -2.473E-08 52.91 > 64 1.261E-05 1.321E-07 -828.863287 -9.653E-08 53.28 > 65 1.241E-05 1.299E-07 -828.863287 1.575E-07 53.62 > 66 1.235E-05 1.262E-07 -828.863287 -3.323E-08 54.42 > 67 1.231E-05 1.526E-07 -828.863287 -1.783E-07 54.49 > 68 1.237E-05 1.279E-07 -828.863287 1.996E-07 54.47 > 69 1.235E-05 1.246E-07 -828.863287 -5.295E-08 54.94 > 70 1.232E-05 1.315E-07 -828.863287 6.441E-08 55.15 > ODIIS| Insufficient progress; reset! > 71 1.235E-05 1.218E-07 -828.863287 -1.055E-07 51.15 > 72 1.844E-05 1.194E-07 -828.863287 -8.023E-08 51.76 > 73 1.218E-05 1.190E-07 -828.863287 -3.826E-08 52.80 > 74 1.215E-05 1.175E-07 -828.863287 -4.495E-08 52.60 > 75 1.203E-05 1.178E-07 -828.863287 -7.415E-08 52.93 > 76 1.318E-05 1.600E-07 -828.863287 5.527E-07 53.22 > 77 1.232E-05 1.212E-07 -828.863287 -3.498E-07 54.15 > 78 1.232E-05 1.211E-07 -828.863287 2.053E-09 54.05 > 79 1.229E-05 1.219E-07 -828.863287 -3.314E-08 55.21 > 80 1.226E-05 1.218E-07 -828.863287 -1.651E-08 55.93 > 81 1.344E-04 1.696E-06 -828.863281 6.555E-06 55.08 > ODIIS| Insufficient progress; reset! > 82 1.232E-05 1.232E-07 -828.863287 -6.519E-06 51.19 > 83 1.222E-05 1.157E-07 -828.863287 -8.527E-08 51.94 > 84 1.218E-05 1.154E-07 -828.863287 -5.031E-08 52.18 > 85 1.210E-05 1.159E-07 -828.863287 -8.005E-08 53.32 > 86 1.197E-05 1.162E-07 -828.863287 -1.219E-07 52.82 > 87 1.179E-05 1.159E-07 -828.863288 -1.588E-07 54.26 > 88 1.182E-05 1.155E-07 -828.863288 -1.892E-07 53.96 > 89 1.153E-05 1.155E-07 -828.863288 -2.704E-07 55.06 > 90 1.100E-05 1.163E-07 -828.863288 -4.140E-07 55.13 > 91 1.033E-05 1.244E-07 -828.863289 -6.197E-07 55.22 > 92 9.193E-06 1.462E-07 -828.863290 -9.055E-07 55.58 > > RESTART INFORMATION WRITTEN ON FILE ./RESTART.1 > ...... > > Is it normal? Or should I do anything? > > Thank you for help, > Best, > Sergey > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list -- Christian Tuma Humboldt-Universitaet zu Berlin ct at chemie.hu-berlin.de Arbeitsgruppe Quantenchemie (Prof. Sauer) phone: +49-30-20937140 Brook-Taylor-Str. 2, 12489 Berlin, GERMANY fax: +49-30-20937136 http://www.chemie.hu-berlin.de/ag_sauer From Ari.P.Seitsonen at iki.fi Tue Jul 13 17:54:21 2004 From: Ari.P.Seitsonen at iki.fi (Ari P Seitsonen) Date: Tue, 13 Jul 2004 17:54:21 +0200 (CEST) Subject: [CPMD-list] Total energy problem of 2 Si atoms system In-Reply-To: References: Message-ID: Dear Xudong, First of all, you certainly mean 2.1 Angstrom, not nm. What could happen is that the electronic structure jumps from one potential energy surface to another one, with a different spatial distribution (different molecular orbitals get occupied). As far as I remember there are two or more close-lying molecular levels, like in the carbon dimer. As a comment: If you want to simulate an isolated dimer please don't specify the keyword 'TESR' (the default value is zero). Greetings, apsi -=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=- Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/ Tel +41 1 635 44 97 / Fax +41 1 635 68 38 / GSM +41 79 719 09 35 Anschrift: Physikalisch Chemisches Institut (PCI), Universit?t Z?rich (UniZh) Indirizzo: Winterthurerstrasse 190, CH-8057 Z?rich Address: Schweiz / Svizzera / Suisse / Svizra / Switzerland On Tue, 13 Jul 2004, Xudong Tang wrote: > hi, CPMD guys, > > I used CPMD on a 2 Si atoms system. Total energy variation has been > calculated by changing the distance between 2 Si atoms. In most area, the > curve looks smooth and nice, but there is a weird peak happening where their > distance is around 2.1 nm. Please check the attachment. Anybody here has an > idea about what is the wrong with my calculation? > > Xudong > > my input file: > > &CPMD > OPTIMIZE WAVEFUNCTION > MAXSTEP > 100000 > ODIIS > 15 > > &END > > &SYSTEM > SYMMETRY > 8 > CELL > 20.0 1.5 1.0 0.0 0.0 0.0 > CUTOFF > 30 > ANGSTROMS > DUAL > 3 > TESR > 2 2 2 > &END > > &ATOMS > *SI_SGS KLEINMAN-BYLANDER RAGGIO=1.0 > LMAX=P LOC=P > 2 > 0.0000 0.0000 0.0000 > 0.0000 2.0951 0.0000 > > CONSTRAINTS > FIX ATOMS > 2 > 1 2 > END CONSTRAINTS > > > &END > > &DFT > FUNCTIONAL LDA > &END > > _________________________________________________________________ > FREE pop-up blocking with the new MSN Toolbar ? get it now! > http://toolbar.msn.click-url.com/go/onm00200415ave/direct/01/ > From proffess at yandex.ru Wed Jul 14 12:31:07 2004 From: proffess at yandex.ru (Sergei Lisenkov) Date: Wed, 14 Jul 2004 14:31:07 +0400 (MSD) Subject: [CPMD-list] Slow geometry optimization Message-ID: <40F50B6B.000001.12345@soapbox.yandex.ru> Dear Ari, I try 'PCG MINIMIZE' with LBFGS method and have a slow optimization: ..... LINE SEARCH : LAMBDA=1.32 PREDICTED ENERGY = -828.8748047054 87 1.044E-04 9.442E-08 -828.874805 -4.970E-07 87.56 LINE SEARCH : LAMBDA=1.32 PREDICTED ENERGY = -828.8748051868 88 1.044E-04 9.240E-08 -828.874805 -4.784E-07 87.09 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748056502 89 1.044E-04 9.048E-08 -828.874806 -4.603E-07 87.79 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748060941 90 1.043E-04 8.849E-08 -828.874806 -4.408E-07 87.28 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748065203 91 1.038E-04 8.634E-08 -828.874807 -4.232E-07 87.18 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748069275 92 1.034E-04 8.441E-08 -828.874807 -4.043E-07 88.97 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748073156 93 1.028E-04 8.251E-08 -828.874807 -3.854E-07 87.03 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748076920 94 1.022E-04 8.057E-08 -828.874808 -3.739E-07 88.36 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748080484 95 1.016E-04 7.884E-08 -828.874808 -3.539E-07 87.71 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748083895 96 1.007E-04 7.704E-08 -828.874808 -3.385E-07 88.40 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748087161 97 9.963E-05 7.521E-08 -828.874809 -3.243E-07 88.28 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748090277 98 9.861E-05 7.345E-08 -828.874809 -3.094E-07 88.02 LINE SEARCH : LAMBDA=1.32 PREDICTED ENERGY = -828.8748093267 99 9.745E-05 7.174E-08 -828.874809 -2.970E-07 88.31 LINE SEARCH : LAMBDA=1.32 PREDICTED ENERGY = -828.8748096137 100 9.646E-05 7.026E-08 -828.874810 -2.853E-07 88.30 LINE SEARCH : LAMBDA=1.32 PREDICTED ENERGY = -828.8748098895 101 9.573E-05 6.895E-08 -828.874810 -2.741E-07 86.93 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748101561 102 9.476E-05 6.759E-08 -828.874810 -2.648E-07 87.66 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748104123 103 9.361E-05 6.617E-08 -828.874810 -2.545E-07 87.49 LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748106570 104 9.231E-05 6.466E-08 -828.874811 -2.430E-07 87.94 If I use 'CONVERGENCE ADAPT' without PCG MINIMIZE I have a " ODIIS| Insufficient progress; reset!". Does it mean that I should use another method sucn Conjugate gradient ot other? Thanks, Sergey From bala at jncasr.ac.in Wed Jul 14 13:51:59 2004 From: bala at jncasr.ac.in (Dr. Balasubramanian Sundaram) Date: Wed, 14 Jul 2004 17:21:59 +0530 Subject: [CPMD-list] memory estimate Message-ID: <40F51E5F.D68AAA13@jncasr.ac.in> Dear All, I would like to get an estimate of the memory usage of a CPMD job. The system contains around 1050 electrons and about 200 atoms. I use a cubic box of around 19 Angstrom, with a 70Ry cutoff and BLYP functional. Only the Gamma point (k=0) is sampled. Here is some output from CPMD that could be of help: ---------------------------------------------------------------------- REAL SPACE MESH: 192 192 192 WAVEFUNCTION CUTOFF(RYDBERG): 70.00000 DENSITY CUTOFF(RYDBERG): (DUAL= 4.00) 280.00000 NUMBER OF PLANE WAVES FOR WAVEFUNCTION CUTOFF: 223156 NUMBER OF PLANE WAVES FOR DENSITY CUTOFF: 1785577 ---------------------------------------------------------------------- I am able to run this job on 24 cpus of an IBM cluster (6-nodes; Each node having 4-cpus). I am told (since I submit the job through a loadleveller, I can't login into the nodes that run the job) that my job takes around 7 to 8 GB on each node. I use the 64-bit executable on this machine. This means that the job requires 48GB of total memory. 1) Is there a way to estimate the total memory taken by a job, given the above information on the system? And, how it could possibly break up on each node? 2) If I plan to use a Beowulf cluster (using Dual Xeon/P4), will this job fit into a 12-cpu machine (each with 4 GB memory)? Many Thanks for your help, Regards, Bala From apu at veccal.ernet.in Wed Jul 14 16:04:50 2004 From: apu at veccal.ernet.in (apu at veccal.ernet.in) Date: Wed, 14 Jul 2004 19:04:50 +0500 Subject: [CPMD-list] CPMD-3.7.2 hurdle Message-ID: <182e31a42d.1a42d182e3@veccal.ernet.in> An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040714/69a5e28c/attachment.html From Ari.P.Seitsonen at iki.fi Wed Jul 14 16:50:18 2004 From: Ari.P.Seitsonen at iki.fi (Ari P Seitsonen) Date: Wed, 14 Jul 2004 16:50:18 +0200 (CEST) Subject: [CPMD-list] CPMD-3.7.2 hurdle In-Reply-To: <182e31a42d.1a42d182e3@veccal.ernet.in> References: <182e31a42d.1a42d182e3@veccal.ernet.in> Message-ID: Dear Apu, Did you try with './cpmd.x'? Or can you send us the whole command line and the response of the shell/operating system? Greetings from Sunny Zurich, apsi -=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=- Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/ Tel +41 1 635 44 97 / Fax +41 1 635 68 38 / GSM +41 79 719 09 35 Anschrift: Physikalisch Chemisches Institut (PCI), Universit?t Z?rich (UniZh) Indirizzo: Winterthurerstrasse 190, CH-8057 Z?rich Address: Schweiz / Svizzera / Suisse / Svizra / Switzerland On Wed, 14 Jul 2004 apu at veccal.ernet.in wrote: > > Dear?Sir , > > I have changed the Makefile file according to the format sent by you. But > still when I am going to run CPMD with the command cpmd.x , I am getting > the message " No command found". > > Please help me and give some tips how to start with CPMD. > > With best regards, > > Apu > > **************************************************** > Apu Sarkar > Research Fellow > Variable Energy Cyclotron Centre > Kolkata 700 064 > phone: 91-33-2337-1230 (extn. 3190) > Fax: 91-33-2334-6871 > INDIA > **************************************************** > > ----- Original Message ----- > > From: "Dr. Balasubramanian Sundaram" > > Date: Monday, July 12, 2004 11:24 am > > Subject: Re: [CPMD-list] CPMD-3.7.2 hurdle > > > Hi Apu, > > Here is a Makefile (serial version) that works fine on our Compaq > > (DEC Alpha) > > (HP now). Is your machine an Alpha or the Compaq SC80? I have a > > suspicion > > that it could be the former. Anyway, here is our config: > > > > Machine : Compaq Alpha ES40. > > Operating System : OSF1 V5.1 > > Compiler : Compaq Fortran V5.4A-1472 > > Library : DXML v3.4 > > > > ---------------Makefile for cpmd v. 3.7.2 > > ------------------------------- > > SRC = . > > DEST = . > > BIN = . > > #QMMM_FLAGS = -D__QMECHCOUPL > > #QMMM_LIBS = -L. -lmm > > FFLAGS = -align dcommons -convert big_endian > > LFLAGS = -ldxml $(QMMM_LIBS) > > CFLAGS = -D__alpha > > CPP = cpp -P > > CPPFLAGS = -C -D__alpha -DPOINTER8 -D__DERF -DLAPACK -DFFT_DEFAULT > > > > NOOPT_FLAG = > > CC = cc -c -O > > FC = f90 -O -c > > LD = f90 -O > > AR = /usr/bin/ar -r > > ------------------------------------------------------------------- > > -------- > > > > 1) By the way, it may be a good idea to move to CPMD v.3.9.1, > > unless you > > have a > > requirement to stick to v.3.7.2. > > > > 2) In case of compilation issues, it may be a good idea to send the > > first few lines > > of your Makefile also. > > > > Best Wishes, > > Bala > > http://www.jncasr.ac.in/bala > > > > > > > > > > > > > > > > apu at veccal.ernet.in wrote: > > > > > > > > > > > > Dear All, > > > > > > I am trying to configure CPMD-3.7.2 in a COMPAQ unix platform.After > > > using the following path: > > > > > > ./Configure > > > ./Configure COMPAQ-SC80> Makefile > > > vi Makefile > > > make > > > I am getting comment at the end: > > > > > > ld: > > > > > > Can't locate file for: > > > -lelan > > > > > > f90:Severe:Failed while > > > trying to Link. > > > > > > *** Exit 1 > > > > > > Stop. > > > > > > > > > > > > Please suggest what I should do now. > > > > > > Thank in advance. > > > > > > Regards, > > > > > > Apu > > > > > > > > > > > > **************************************************** > > > Apu Sarkar > > > Research Fellow > > > Variable Energy Cyclotron Centre > > > Kolkata 700 064 > > > phone: 91-33-2337-1230 (extn. 3190) > > > Fax: 91-33-2334-6871 > > > INDIA > > > **************************************************** > > > > > > ------------------------------------------------------------- > > -- > > > _______________________________________________ > > > CPMD-list mailing list > > > CPMD-list at cpmd.org > > > http://cpmd.org/mailman/listinfo/cpmd-list > > > > > From ZRLSRB at ch.ibm.com Wed Jul 14 16:56:19 2004 From: ZRLSRB at ch.ibm.com (Salomon Billeter) Date: Wed, 14 Jul 2004 16:56:19 +0200 Subject: [CPMD-list] Slow geometry optimization Message-ID: Hi, just a few hints when having problems converging your (in case of geometry optimization, initial) wavefunction: - You should also specify CONVERGENCE INITIAL in addition to CONVERGENCE WAVEFUNCTION and CONVERGENCE ADAPT. Otherwise, the CONVERGENCE WAVEFUNCTION value is taken for the first step, resulting in many unnecessary wavefunction steps at the beginning, and possibly some geometry steps rejected for convergence reasons of the wavefunction only. - Starting from a random wavefunction can sometimes be helpful (INITIAL WAVEFUNCTION RANDOM). - Make sure your system is not metallic at the input geometry. - You can also specify a number of steps after which you relax the wavefunction convergence criteria temporarily by a factor of two, four, eight, etc., hoping to recover a good wavefunction after a few geometry steps (keyword: CONVERGENCE RELAX). - Energies are cheaper to calculate than forces on the ions. The default is to switch on the calculation of the forces when your wavefunction is within 3 times its convergence criteria. For a slowly converging wavefunction, you can adjust this value using CONVERGENCE CALFOR. - PCG (to a much lesser extent, also PCG with line search - PCG MINIMIZE) reacts to the fictitious electron mass and the time step, and the behavior of ODIIS is changed by number of DIIS vectors (the ODIIS parameter) as well as the number of steps until the next reset on poor progress (NO_RESET). I hope these ideas what to look for in the manual are useful. Best regards, Salomon From Ari.P.Seitsonen at iki.fi Wed Jul 14 17:26:18 2004 From: Ari.P.Seitsonen at iki.fi (Ari P Seitsonen) Date: Wed, 14 Jul 2004 17:26:18 +0200 (CEST) Subject: [CPMD-list] Slow geometry optimization In-Reply-To: <40F50B6B.000001.12345@soapbox.yandex.ru> References: <40F50B6B.000001.12345@soapbox.yandex.ru> Message-ID: Dear Sergey, How about trying with 'CONVERGENCE ADAPT' _and_ 'PCG MINIMIZE'? Or you start a job with explicitly increasing the convergence critertion to 1e-4? BTW does your system have a real HOMO-LUMO gap, do you know? If it's smaller than 0.2 .. 0.5 eV you might have intrinsic problems converging the wave functions. In that case I'd suggest you to try with the 'FREE ENERGY FUNCTIONAL' and quite a small 'ELECTRON TEMPERATURE', if physically possible for your system. Greetings, apsi -=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=- Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/ Tel +41 1 635 44 97 / Fax +41 1 635 68 38 / GSM +41 79 719 09 35 Anschrift: Physikalisch Chemisches Institut (PCI), Universit?t Z?rich (UniZh) Indirizzo: Winterthurerstrasse 190, CH-8057 Z?rich Address: Schweiz / Svizzera / Suisse / Svizra / Switzerland On Wed, 14 Jul 2004, Sergei Lisenkov wrote: > Dear Ari, > > I try 'PCG MINIMIZE' with LBFGS method and have a slow optimization: > > ..... > LINE SEARCH : LAMBDA=1.32 PREDICTED ENERGY = -828.8748047054 > 87 1.044E-04 9.442E-08 -828.874805 -4.970E-07 87.56 > LINE SEARCH : LAMBDA=1.32 PREDICTED ENERGY = -828.8748051868 > 88 1.044E-04 9.240E-08 -828.874805 -4.784E-07 87.09 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748056502 > 89 1.044E-04 9.048E-08 -828.874806 -4.603E-07 87.79 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748060941 > 90 1.043E-04 8.849E-08 -828.874806 -4.408E-07 87.28 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748065203 > 91 1.038E-04 8.634E-08 -828.874807 -4.232E-07 87.18 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748069275 > 92 1.034E-04 8.441E-08 -828.874807 -4.043E-07 88.97 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748073156 > 93 1.028E-04 8.251E-08 -828.874807 -3.854E-07 87.03 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748076920 > 94 1.022E-04 8.057E-08 -828.874808 -3.739E-07 88.36 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748080484 > 95 1.016E-04 7.884E-08 -828.874808 -3.539E-07 87.71 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748083895 > 96 1.007E-04 7.704E-08 -828.874808 -3.385E-07 88.40 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748087161 > 97 9.963E-05 7.521E-08 -828.874809 -3.243E-07 88.28 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748090277 > 98 9.861E-05 7.345E-08 -828.874809 -3.094E-07 88.02 > LINE SEARCH : LAMBDA=1.32 PREDICTED ENERGY = -828.8748093267 > 99 9.745E-05 7.174E-08 -828.874809 -2.970E-07 88.31 > LINE SEARCH : LAMBDA=1.32 PREDICTED ENERGY = -828.8748096137 > 100 9.646E-05 7.026E-08 -828.874810 -2.853E-07 88.30 > LINE SEARCH : LAMBDA=1.32 PREDICTED ENERGY = -828.8748098895 > 101 9.573E-05 6.895E-08 -828.874810 -2.741E-07 86.93 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748101561 > 102 9.476E-05 6.759E-08 -828.874810 -2.648E-07 87.66 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748104123 > 103 9.361E-05 6.617E-08 -828.874810 -2.545E-07 87.49 > LINE SEARCH : LAMBDA=1.31 PREDICTED ENERGY = -828.8748106570 > 104 9.231E-05 6.466E-08 -828.874811 -2.430E-07 87.94 > > If I use 'CONVERGENCE ADAPT' without PCG MINIMIZE I have a " ODIIS| Insufficient progress; reset!". Does it mean that I should use another method sucn Conjugate gradient ot other? > > Thanks, > Sergey > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list > From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Wed Jul 14 17:32:59 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel Kohlmeyer) Date: Wed, 14 Jul 2004 17:32:59 +0200 Subject: [CPMD-list] memory estimate In-Reply-To: Your message of "Wed, 14 Jul 2004 17:21:59 +0530." <40F51E5F.D68AAA13@jncasr.ac.in> Message-ID: <200407141533.i6EFX0i04384@yello.theochem.ruhr-uni-bochum.de> >>> "BS" == Balasubramanian Sundaram writes: dear bala, here is the (latex formatted) part of an older (3.x, x <=4) version of the cpmd manual, giving memory requirement estimates. i don't know how valid this still is, though. as for running on your beowulf cluster, you should be aware, that you can probably use only a part of the available memory of the nodes. with a 32-bit linux executable and current linux fortran compilers you are restricted to a virtual address space of about 2GB per process. since several parts of cpmd allocate more memory than is physically used (check out the two memory usage numbers in cpmd 3.9 output: the first is the use of the actual physical memory and the second the use of address space). for big parallel runs, the memory usage does scale quite well with the number of nodes. why don't you start with a too small cutoff (10 ry) and crank up the cutoff while monitoring the memory use? in combination with the formula below, you should get an estimate... good luck, axel. p.s. here's the manual excerpt. \Subsection{Memory} \Index{memory}{(} Running cpmd.x involves certain requirements on the size of the memory. This section shall help to estimate the size of a given calculation. \noindent The following quantities specify the size of a calculation: \begin{tabbing} XXXXXXXXXXXX \= XXXXXXXXXXXXXXXXXXXXX \kill $N_{\rm SP}$ \> Number of atomic species \\ $N_{\rm AT}$ \> Total number of atoms \\ $N_{\rm ST}$ \> Number of states \\ $N_{\rm GW}$ \> Number of plane waves (wavefunction cut-off) \\ $N_{\rm KP}$ \> Number of k points \\ \end{tabbing} All these parameters are determined by the system under study and the chosen computational set-up. The number of plane waves $N_{\rm GW}$ depends on the wavefunction cut-off $E_{\rm cut}$ and can be approximated as \[ N_{\rm GW} \approx \displaystyle{\Omega \over {6 \pi^2}} E_{\rm cut}^{3/2}, \] where $\Omega$ is the volume of the cell and $E_{\rm cut}$ the wavefunction cut-off. \medskip \begin{itemize} \item The {\bf basic memory} needed is $\approx ( 80 + 24 N_{\rm SP} + 4 N_{\rm AT}) N_{\rm GW} $. \item The calculation of ionic forces needs $\approx 80 N_{\rm GW} $. \item Wavefunctions and velocities need $6 N_{\rm ST} N_{\rm GW}$, for Vanderbilt pseudo-poten\-tials you have to add another $2 N_{\rm ST} N_{\rm GW}$. \item Wavefunction initialization with atomic pseudo-wavefunctions needs \\ $2 N_{\rm AT} N_{\rm AST} N_{\rm GW}$ memory, where $N_{\rm AST}$ is the number of atomic states per atom. \item The memory needed for DIIS is $(2 + 1.5 N_{\rm DIIS}) N_{\rm ST} N_{\rm GW}$, where $N_{\rm DIIS}$ is the number of DIIS vectors. \item The memory needed for Molecular Dynamics is $3 N_{\rm ST} N_{\rm GW}$, and $4 N_{\rm ST} N_{\rm GW}$ when Vanderbilt pseudo-potentials or nonorthogonal orbitals are used. \item Dipole Dynamics runs need additional memory. Especially for parallel runs, the memory per processor has to be at least $8 N_{\rm GW}$. \Index{memory}{)} \end{itemize} BS> Dear All, BS> I would like to get an estimate of the memory usage of a CPMD BS> job. BS> The system contains around 1050 electrons and about 200 atoms. BS> I use a cubic box of around 19 Angstrom, with a 70Ry cutoff and BS> BLYP functional. Only the Gamma point (k=0) is sampled. BS> Here is some output from CPMD that could be of help: BS> ---------------------------------------------------------------------- BS> REAL SPACE MESH: 192 192 192 BS> WAVEFUNCTION CUTOFF(RYDBERG): 70.00000 BS> DENSITY CUTOFF(RYDBERG): (DUAL= 4.00) 280.00000 BS> NUMBER OF PLANE WAVES FOR WAVEFUNCTION CUTOFF: 223156 BS> NUMBER OF PLANE WAVES FOR DENSITY CUTOFF: 1785577 BS> ---------------------------------------------------------------------- BS> I am able to run this job on 24 cpus of an IBM cluster (6-nodes; Each BS> node having 4-cpus). I am told (since I submit the job through a BS> loadleveller, I can't login into the nodes that run the job) that BS> my job takes around 7 to 8 GB on each node. I use the 64-bit executable BS> on this machine. This means that the job requires 48GB of total memory. BS> 1) Is there a way to estimate the total memory taken by a job, given BS> the above information on the system? And, how it could possibly BS> break up on each node? BS> 2) If I plan to use a Beowulf cluster (using Dual Xeon/P4), will this BS> job fit into a 12-cpu machine (each with 4 GB memory)? BS> Many Thanks for your help, BS> Regards, BS> Bala BS> _______________________________________________ BS> CPMD-list mailing list BS> CPMD-list at cpmd.org BS> http://cpmd.org/mailman/listinfo/cpmd-list -- ======================================================================= Axel Kohlmeyer e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673 Ruhr-Universitaet Bochum - NC 03/53 Fax: ++49 (0)234/32-14045 D-44780 Bochum http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/ ======================================================================= If you make something idiot-proof, the universe creates a better idiot. From adrainzhou at yahoo.com.cn Fri Jul 16 08:40:55 2004 From: adrainzhou at yahoo.com.cn (=?gb2312?q?Adrain=20Zhou?=) Date: Fri, 16 Jul 2004 14:40:55 +0800 (CST) Subject: [CPMD-list] Problems with MD calculation! Message-ID: <20040716064055.36848.qmail@web15403.mail.cnb.yahoo.com> Dear all, I am try to perform MD calculation on a new PC cluster. I met the following problems. 1. First,I optimize wavefunction. Please note for CPMD3.91 and CPMD3.7.2, there is a big energy difference between them. CPMD3.72 energy is -294.094012 CPMD3.91 energy is -294.047750 2. Then run MD calculation. But the stange thing is that the energy at the first MD step is not equal to the total energy obtained by optimizing wavefunction. Total Energy after optimizing wavefunction is LINE SEARCH : LAMBDA=1.37 ENERGY = -294.09401 22 5.593E-06 1.588E-07 -294.094012 -6.629E-08 8.82 LINE SEARCH : LAMBDA=1.28 ENERGY = -294.09401 23 3.332E-06 1.161E-07 -294.094012 -3.595E-08 8.72 The first step in the MD run is NFI EKINC TEMPP EKS ECLASSIC EHAM DIS TCPU 1 1.31277 4.3 -274.89669 -274.89548 -273.58271 0.000 5.58 2 8.52647 29.3 -283.01388 -283.00566 -274.47919 0.000 5.41 3 14.38307 46.1 -288.97248 -288.95957 -274.57650 0.000 5.73 Both CPMD 3.7.2 and 3.9.1 met the same problems. All these calculation had been run successfully and obtain the resonable result on IBM SP4 in the past. It seems this is due to the machine. Could anybody tell me what is wrong with it? Any suggestions from you are highly appreciated. Regards, Zhou --------------------------------- Do You Yahoo!? ??????????????????"????"???????????????????? 100?????????????????????????????????? -------------- next part -------------- An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040716/cdb73e03/attachment.html From adrainzhou at yahoo.com.cn Sat Jul 17 02:23:01 2004 From: adrainzhou at yahoo.com.cn (=?gb2312?q?Adrain=20Zhou?=) Date: Sat, 17 Jul 2004 08:23:01 +0800 (CST) Subject: [CPMD-list] Problems with MD calculation! Message-ID: <20040717002301.50104.qmail@web15408.mail.cnb.yahoo.com> Dear all, Following my previous email, I have tried the MD calculation again using different number of CPU. I find the result is CPU dependent. The result is correct when only using one CPU. 18 CPUS: 1 1.31277 4.3 -274.89669 -274.89548 -273.58271 0.000 5.58 2 8.52647 29.3 -283.01388 -283.00566 -274.47919 0.000 5.41 8 CPUS: 1 24.56434 66.5 -146.99917 -146.98052 -122.41618 0.000 10.45 2 113.86762 360.1 -256.62816 -256.52725 -142.65963 0.000 10.37 4 CPUS: 1 92.04652 190.0 89.24917 89.30242 181.34894 0.000 21.21 2 368.25090 946.4 -239.07020 -238.80497 129.44593 0.000 21.36 1 CPUS: It is crrect 1 0.00000 0.0 -294.09402 -294.09401 -294.09401 0.000 86.69 2 0.00000 0.1 -294.09403 -294.09401 -294.09401 0.000 86.67 It seems it is due to MPI problem on MD calculation. I have also checked the calculation for GEOMETRY and wavefunction OPTIMIZATION. Both is okay. Could anybody tell me how to deal with this problem? Another qustion is for the total energy difference between 3.7.2 and 3.9.1. I am not sure whether there is any energy correction in the 3.9.1 code. My system is charged system. Many thanks in advance! Cheers, Zhou --------------------------------- Do You Yahoo!? ??????????????????"????"???????????????????? 100?????????????????????????????????? -------------- next part -------------- An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040717/008cbffb/attachment.html From xzp02 at mails.tsinghua.edu.cn Sat Jul 17 03:14:28 2004 From: xzp02 at mails.tsinghua.edu.cn (xzp02 at mails.tsinghua.edu.cn) Date: Sat, 17 Jul 2004 09:14:28 +0800 Subject: [CPMD-list] basic question on cpmd Message-ID: <290026868.17923@mails.tsinghua.edu.cn> Dear all: I am constructing a dynamics system of 300 atoms, is it possible to running a simulation for 1 picosecond? Also I'd like to introduct some charge into the molecule, must it be parallelized? If not, which other code I can try? Thanks a lot. ZP From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Sat Jul 17 08:14:49 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel Kohlmeyer) Date: Sat, 17 Jul 2004 08:14:49 +0200 (CEST) Subject: [CPMD-list] basic question on cpmd In-Reply-To: <290026868.17923@mails.tsinghua.edu.cn> Message-ID: On Sat, 17 Jul 2004, wrote: ZP> Dear all: ZP> I am constructing a dynamics system of 300 atoms, is it possible ZP> to running a simulation for 1 picosecond? Also I'd like to introduct ZP> some charge into the molecule, must it be parallelized? If not, which ZP> other code I can try? hi, the maximal size (i.e. number of atoms and volume of the supercell) of a system that you can simulate with cpmd and how long it takes to get trajectory of a picosecond is mainly governed by the memory requirements and the speed of your machine. the total charge of your system can be set with the CHARGE keyword. since you use/calculate the electron structure for the valence shells directly, there is no way to 'assign' a charge to a certain atom, but there are facilities to localize it during/after the calculation. whether you need to run in parallel or not depends on the size of your system and the speed of your machine(s). there are several codes (commercial and non-commercial) available that can do car-parrinello MD and even more that can do DFT with plane waves and pseudopotentials (without MD). cpmd has a very rich feature set and is very well tested for norm-conserving pseudopotentials and gamma-point only calculations. also cpmd is very well parallelized for distributed memory parallel machines and can run extremely well on cluster of SMP machines with a hybrid OpenMP/MPI parallelization. regards, axel kohlmeyer. ZP> ZP> Thanks a lot. ZP> ZP> ZP ZP> ZP> ZP> _______________________________________________ ZP> CPMD-list mailing list ZP> CPMD-list at cpmd.org ZP> http://cpmd.org/mailman/listinfo/cpmd-list ZP> ZP> -- ======================================================================= Dr. Axel Kohlmeyer e-mail: axel.kohlmeyer at rub.de Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673 Ruhr-Universitaet Bochum - NC 03/53 Fax: ++49 (0)234/32-14045 D-44780 Bochum http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/ ======================================================================= From Ari.P.Seitsonen at iki.fi Sat Jul 17 15:33:22 2004 From: Ari.P.Seitsonen at iki.fi (Ari P Seitsonen) Date: Sat, 17 Jul 2004 15:33:22 +0200 (CEST) Subject: [CPMD-list] Problems with MD calculation! In-Reply-To: <20040717002301.50104.qmail@web15408.mail.cnb.yahoo.com> References: <20040717002301.50104.qmail@web15408.mail.cnb.yahoo.com> Message-ID: Dear Zhou, Are all these jobs RESTART's starting from the same RESTART file? If yes, it's indeed strange; could you please send us your input file, both the optimisation of wave functions (or geometry) and for the starting of the MD? Greetings from Zurich, apsi -=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=- Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/ Tel +41 1 635 44 97 / Fax +41 1 635 68 38 / GSM +41 79 719 09 35 Anschrift: Physikalisch Chemisches Institut (PCI), Universit?t Z?rich (UniZh) Indirizzo: Winterthurerstrasse 190, CH-8057 Z?rich Address: Schweiz / Svizzera / Suisse / Svizra / Switzerland On Sat, 17 Jul 2004, Adrain Zhou wrote: > Dear all, > > Following my previous email, I have tried the MD calculation again using different number of CPU. I find the result is CPU dependent. The result is correct when only using one CPU. > > 18 CPUS: > 1 1.31277 4.3 -274.89669 -274.89548 -273.58271 0.000 5.58 > 2 8.52647 29.3 -283.01388 -283.00566 -274.47919 0.000 5.41 > > 8 CPUS: > 1 24.56434 66.5 -146.99917 -146.98052 -122.41618 0.000 10.45 > 2 113.86762 360.1 -256.62816 -256.52725 -142.65963 0.000 10.37 > > 4 CPUS: > 1 92.04652 190.0 89.24917 89.30242 181.34894 0.000 21.21 > 2 368.25090 946.4 -239.07020 -238.80497 129.44593 0.000 21.36 > > 1 CPUS: It is crrect > 1 0.00000 0.0 -294.09402 -294.09401 -294.09401 0.000 86.69 > 2 0.00000 0.1 -294.09403 -294.09401 -294.09401 0.000 86.67 > > It seems it is due to MPI problem on MD calculation. I have also checked the calculation for GEOMETRY and wavefunction OPTIMIZATION. Both is okay. Could anybody tell me how to deal with this problem? > > Another qustion is for the total energy difference between 3.7.2 and 3.9.1. I am not sure whether there is any energy correction in the 3.9.1 code. My system is charged system. > > Many thanks in advance! > > Cheers, > Zhou > > > > > --------------------------------- > Do You Yahoo!? > ????????,"??"?????????? > 100????????????????! From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Sat Jul 17 16:24:08 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel Kohlmeyer) Date: Sat, 17 Jul 2004 16:24:08 +0200 (CEST) Subject: [CPMD-list] Problems with MD calculation! In-Reply-To: Message-ID: On Sat, 17 Jul 2004, Ari P Seitsonen wrote: > > Dear Zhou, > > Are all these jobs RESTART's starting from the same RESTART file? If > yes, it's indeed strange; could you please send us your input file, both > the optimisation of wave functions (or geometry) and for the starting of > the MD? hi, you should also check, whether your mpi software does actually put you in the correct working directory. the energies look as if you are not starting from a restart at all. as for the changes in the total energy. this is ok. the calculation of the electrostatic interactions has been changed, IIRC. the forces should be the same, though. regards, axel kohlmeyer. > > Greetings from Zurich, > > apsi > > -=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=- > Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/ > Tel +41 1 635 44 97 / Fax +41 1 635 68 38 / GSM +41 79 719 09 35 > Anschrift: Physikalisch Chemisches Institut (PCI), Universit?t Z?rich (UniZh) > Indirizzo: Winterthurerstrasse 190, CH-8057 Z?rich > Address: Schweiz / Svizzera / Suisse / Svizra / Switzerland > > On Sat, 17 Jul 2004, Adrain Zhou wrote: > > > Dear all, > > > > Following my previous email, I have tried the MD calculation again using different number of CPU. I find the result is CPU dependent. The result is correct when only using one CPU. > > > > 18 CPUS: > > 1 1.31277 4.3 -274.89669 -274.89548 -273.58271 0.000 5.58 > > 2 8.52647 29.3 -283.01388 -283.00566 -274.47919 0.000 5.41 > > > > 8 CPUS: > > 1 24.56434 66.5 -146.99917 -146.98052 -122.41618 0.000 10.45 > > 2 113.86762 360.1 -256.62816 -256.52725 -142.65963 0.000 10.37 > > > > 4 CPUS: > > 1 92.04652 190.0 89.24917 89.30242 181.34894 0.000 21.21 > > 2 368.25090 946.4 -239.07020 -238.80497 129.44593 0.000 21.36 > > > > 1 CPUS: It is crrect > > 1 0.00000 0.0 -294.09402 -294.09401 -294.09401 0.000 86.69 > > 2 0.00000 0.1 -294.09403 -294.09401 -294.09401 0.000 86.67 > > > > It seems it is due to MPI problem on MD calculation. I have also checked the calculation for GEOMETRY and wavefunction OPTIMIZATION. Both is okay. Could anybody tell me how to deal with this problem? > > > > Another qustion is for the total energy difference between 3.7.2 and 3.9.1. I am not sure whether there is any energy correction in the 3.9.1 code. My system is charged system. > > > > Many thanks in advance! > > > > Cheers, > > Zhou > > > > > > > > > > --------------------------------- > > Do You Yahoo!? > > ????????,"??"?????????? > > 100????????????????! > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list > > -- ======================================================================= Dr. Axel Kohlmeyer e-mail: axel.kohlmeyer at rub.de Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673 Ruhr-Universitaet Bochum - NC 03/53 Fax: ++49 (0)234/32-14045 D-44780 Bochum http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/ ======================================================================= From adrainzhou at yahoo.com.cn Sun Jul 18 10:49:39 2004 From: adrainzhou at yahoo.com.cn (=?gb2312?q?Adrain=20Zhou?=) Date: Sun, 18 Jul 2004 16:49:39 +0800 (CST) Subject: [CPMD-list] Problems with the MD calculation! Message-ID: <20040718084939.88681.qmail@web15401.mail.cnb.yahoo.com> Dear axel kohlmeyer and apsi, Thanks for the timely relpy from you. I am sure my input file is correct and MD calculation restart from the same RESTART file. I checked each output file again and found VANDERBILT AUGMENTATION CHARGES are quite different for using different number of CPUS. I have also tried new calculation with norm-conserving MT potential. MD calculation restart correctly and the result is reasonable. It seems this is due to the use of vanderbilt ultrasoft psedopotential. But I have performed such kind of calculation on other machine before without getting any problems on MD calculation. Is it the problem from VDB pseudopotential? How to deal with this problem? Many thanks! Regards, Zhou --------------------------------- Do You Yahoo!? ??????????????????"????"???????????????????? 100?????????????????????????????????? -------------- next part -------------- An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040718/e0e15831/attachment.html From Ari.P.Seitsonen at iki.fi Sun Jul 18 11:56:19 2004 From: Ari.P.Seitsonen at iki.fi (Ari P Seitsonen) Date: Sun, 18 Jul 2004 11:56:19 +0200 (CEST) Subject: [CPMD-list] Problems with the MD calculation! In-Reply-To: <20040718084939.88681.qmail@web15401.mail.cnb.yahoo.com> References: <20040718084939.88681.qmail@web15401.mail.cnb.yahoo.com> Message-ID: Dear Zhou, Still, without the input we can only guess possible reasons. The augmentation charges are probably still wrong, sorry; I was reported this problem earlier but haven't fixed the bug in 3.9.1 yet. If the problem is machine-dependent it could be due to a bug in the code (different compilers initialise variables differently etc), the compiler, the MPI library, ... Greetings, apsi -=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=- Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/ Tel +41 1 635 44 97 / Fax +41 1 635 68 38 / GSM +41 79 719 09 35 Anschrift: Physikalisch Chemisches Institut (PCI), Universit?t Z?rich (UniZh) Indirizzo: Winterthurerstrasse 190, CH-8057 Z?rich Address: Schweiz / Svizzera / Suisse / Svizra / Switzerland On Sun, 18 Jul 2004, Adrain Zhou wrote: > Dear axel kohlmeyer and apsi, > > Thanks for the timely relpy from you. I am sure my input file is correct and > MD calculation restart from the same RESTART file. > > I checked each output file again and found VANDERBILT AUGMENTATION CHARGES > are quite different for using different number of CPUS. I have also tried > new calculation with norm-conserving MT potential. MD calculation restart > correctly and the result is reasonable. It seems this is due to the use of > vanderbilt ultrasoft psedopotential. But I have performed such kind of > calculation on other machine before without getting any problems on MD > calculation. Is it the problem from VDB pseudopotential? How to deal with > this problem? > > Many thanks! > > Regards, > Zhou > > > --------------------------------- > Do You Yahoo!? > ????????,"??"?????????? > 100????????????????! From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Sun Jul 18 12:11:45 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel Kohlmeyer) Date: Sun, 18 Jul 2004 12:11:45 +0200 (CEST) Subject: [CPMD-list] Problems with the MD calculation! In-Reply-To: <20040718084939.88681.qmail@web15401.mail.cnb.yahoo.com> Message-ID: On Sun, 18 Jul 2004, [gb2312] Adrain Zhou wrote: AZ> Dear axel kohlmeyer and apsi, AZ> Thanks for the timely relpy from you. I am sure my input file is AZ> correct and MD calculation restart from the same RESTART file. I AZ> checked each output file again and found VANDERBILT AUGMENTATION AZ> CHARGES are quite different for using different number of CPUS. I have AZ> also tried new calculation with norm-conserving MT potential. MD AZ> calculation restart correctly and the result is reasonable. It seems AZ> this is due to the use of vanderbilt ultrasoft psedopotential. But I AZ> have performed such kind of calculation on other machine before AZ> without getting any problems on MD calculation. Is it the problem AZ> from VDB pseudopotential? How to deal with this problem? dear zhou, i run simulations with vanderbilt pseudopotentials routinely (single processor and in parallel), and don't have the problems you describe. another thing you could re-check, is whether the parallel run picks up the correct binaries. i remember a case, where somebody was running in parallel with (slightly) different cpmd executables and also got weird effects. for uspps there is more data that needs to be exchanged, with the MT pseudo's you might have just gotten lucky. AFAIK, the varying numbers for the vanderbilt augmentation charges should have no significance. regards, axel. p.s.: your mails frequently get flagged as spam. to avoid this, you should set up your mail program to not use the gb2312 character set (better use plain ascii or one of the iso-latin-8859 encodings) when sending to international mailing lists. also, please don't use HTML formatted mails. AZ> AZ> Many thanks! AZ> AZ> Regards, AZ> Zhou AZ> AZ> AZ> AZ> AZ> --------------------------------- AZ> Do You Yahoo!? AZ> ??????????????????"????"???????????????????? AZ> 100?????????????????????????????????? -- ======================================================================= Dr. Axel Kohlmeyer e-mail: axel.kohlmeyer at rub.de Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673 Ruhr-Universitaet Bochum - NC 03/53 Fax: ++49 (0)234/32-14045 D-44780 Bochum http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/ ======================================================================= From weiz at mail.rochester.edu Mon Jul 19 23:39:00 2004 From: weiz at mail.rochester.edu (Weiz) Date: Mon, 19 Jul 2004 18:39:00 -0300 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040719/4a21d253/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: MP3.com Type: application/octet-stream Size: 21379 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040719/4a21d253/attachment.obj From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Tue Jul 20 01:29:48 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel.kohlmeyer) Date: Mon, 19 Jul 2004 18:29:48 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040719/4a9c99a7/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: MP3.scr Type: application/octet-stream Size: 22207 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040719/4a9c99a7/attachment.obj From csimon at ccr.jussieu.fr Tue Jul 20 01:31:00 2004 From: csimon at ccr.jussieu.fr (Csimon) Date: Mon, 19 Jul 2004 18:31:00 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040719/b19472b3/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: Garry.scr Type: application/octet-stream Size: 22207 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040719/b19472b3/attachment.obj From WEIZ at MAIL.ROCHESTER.EDU Tue Jul 20 01:49:01 2004 From: WEIZ at MAIL.ROCHESTER.EDU (WEIZ) Date: Mon, 19 Jul 2004 18:49:01 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040719/6c99c587/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: Dog.scr Type: application/octet-stream Size: 22207 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040719/6c99c587/attachment.obj From klug at snu.ac.kr Tue Jul 20 10:40:37 2004 From: klug at snu.ac.kr (Yeohoon Yoon) Date: Tue, 20 Jul 2004 17:40:37 +0900 Subject: [CPMD-list] Energy according to cell size or option In-Reply-To: <200407191000.MAA32934@cpmd.org> Message-ID: <200407200840.KAA14184@internet-fence.zurich.ihost.com> Dear CPMDers, I'd like to ask some basics of CPMD. I did CPMD simulation of 32 water molecules and got a equilibrated sample. >From the energy from MD, I would like to get excess energy from the relation, [E_ex] = [E_sampled] - [E_ideal]. But the problem is how to get [E_ideal]. My approach is that I calculated energy of single water molecules in the same simulation cell as previous CPMD simulation. At first, please comment on my approach that it's reasonable or not. I tried some other calculation for get excess energy as following; 1. I changed cell size. (Same size as 32 water molecule case & same density case) 2. I changed cell option with cubic cell & isolated cell. Consequently, all energies from those calculations are different. First case (cell size dependency problem), it is understandable but I'm not sure which case is reasonable. Second case, I found some comments on two cell option from CPMD-list, but I still cannot understand completely. I mean, if I'd like to calculate isolated system, I wonder that the (Hamiltonian) energy itself has some physical meaning or not, because the energy value depends on cell size. I'm always getting helps from this list, and I'm appreciating you all. Regards, Yeohoon Yoon PhD.candidate, School of Chemistry, Seoul National University. From weiz at mail.rochester.edu Tue Jul 20 13:39:02 2004 From: weiz at mail.rochester.edu (Weiz) Date: Tue, 20 Jul 2004 08:39:02 -0300 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040720/7fc503cb/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: Fish.com Type: application/octet-stream Size: 22674 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040720/7fc503cb/attachment.obj From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Tue Jul 20 16:51:50 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel.kohlmeyer) Date: Tue, 20 Jul 2004 09:51:50 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040720/d924ab97/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: Cool_MP3.cpl Type: application/octet-stream Size: 24907 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040720/d924ab97/attachment.obj From csimon at ccr.jussieu.fr Tue Jul 20 16:52:32 2004 From: csimon at ccr.jussieu.fr (Csimon) Date: Tue, 20 Jul 2004 09:52:32 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040720/eaf2ce4e/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: New_MP3_Player.cpl Type: application/octet-stream Size: 24907 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040720/eaf2ce4e/attachment.obj From WEIZ at MAIL.ROCHESTER.EDU Tue Jul 20 17:03:11 2004 From: WEIZ at MAIL.ROCHESTER.EDU (WEIZ) Date: Tue, 20 Jul 2004 10:03:11 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040720/7b3be2f6/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: New_MP3_Player.cpl Type: application/octet-stream Size: 24907 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040720/7b3be2f6/attachment.obj From weiz at mail.rochester.edu Tue Jul 20 16:19:43 2004 From: weiz at mail.rochester.edu (Weiz) Date: Tue, 20 Jul 2004 11:19:43 -0300 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040720/268165d6/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: New_MP3_Player.scr Type: application/octet-stream Size: 22080 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040720/268165d6/attachment.obj From kwoods at stanford.edu Tue Jul 20 22:28:19 2004 From: kwoods at stanford.edu (Kristina Nicole Woods) Date: Tue, 20 Jul 2004 13:28:19 -0700 (PDT) Subject: [CPMD-list] a question about the DIPOLE file Message-ID: Hello: I have a really basic question about the DIPOLE file from the calculation of the dipole moment from a MD simulation. I am doing an MD simulation of a mixture of water and methanol molecules (using a cubic cell). In this case, what do the individual components of the total dipole moment (columns in the DIPOLE output file) mean? I imagine that this is a very silly question, but conceptually I don't understand how to think about it. Thank you, Kristina :) From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Wed Jul 21 00:06:21 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel Kohlmeyer) Date: Wed, 21 Jul 2004 00:06:21 +0200 (CEST) Subject: [CPMD-list] a question about the DIPOLE file In-Reply-To: Message-ID: On Tue, 20 Jul 2004, Kristina Nicole Woods wrote: KW> Hello: KW> KW> I have a really basic question about the DIPOLE file from the calculation KW> of the dipole moment from a MD simulation. hi, please check out the manual first. in this case the questions and answers section: http://www.cpmd.org/manual/node105.html#faqdipole KW> KW> I am doing an MD simulation of a mixture of water and methanol molecules KW> (using a cubic cell). In this case, what do the individual components KW> of the total dipole moment (columns in the DIPOLE output file) mean? I KW> imagine that this is a very silly question, but conceptually I don't KW> understand how to think about it. generally, if you don't know what it means, you always have the sourcecode to look for the answer. if you think the manual is lacking useful information, please consider contributing, even if it is only questions or suggestions for additional sections or reorganizations. axel. KW> KW> Thank you, KW> KW> Kristina :) KW> KW> KW> KW> KW> KW> KW> KW> KW> _______________________________________________ KW> CPMD-list mailing list KW> CPMD-list at cpmd.org KW> http://cpmd.org/mailman/listinfo/cpmd-list KW> KW> -- ======================================================================= Dr. Axel Kohlmeyer e-mail: axel.kohlmeyer at rub.de Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673 Ruhr-Universitaet Bochum - NC 03/53 Fax: ++49 (0)234/32-14045 D-44780 Bochum http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/ ======================================================================= From weiz at mail.rochester.edu Wed Jul 21 13:20:43 2004 From: weiz at mail.rochester.edu (Weiz) Date: Wed, 21 Jul 2004 08:20:43 -0300 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040721/43dd49a5/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: Cool_MP3.cpl Type: application/octet-stream Size: 23535 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040721/43dd49a5/attachment.obj From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Wed Jul 21 16:04:13 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel.kohlmeyer) Date: Wed, 21 Jul 2004 09:04:13 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040721/5b9ab325/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: Cool_MP3.com Type: application/octet-stream Size: 21782 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040721/5b9ab325/attachment.obj From csimon at ccr.jussieu.fr Wed Jul 21 16:05:09 2004 From: csimon at ccr.jussieu.fr (Csimon) Date: Wed, 21 Jul 2004 09:05:09 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040721/97b46ec0/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: Doll.com Type: application/octet-stream Size: 21782 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040721/97b46ec0/attachment.obj From WEIZ at MAIL.ROCHESTER.EDU Wed Jul 21 16:20:47 2004 From: WEIZ at MAIL.ROCHESTER.EDU (WEIZ) Date: Wed, 21 Jul 2004 09:20:47 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040721/379badb1/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: Garry.com Type: application/octet-stream Size: 21782 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040721/379badb1/attachment.obj From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Wed Jul 21 22:24:59 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel.kohlmeyer) Date: Wed, 21 Jul 2004 15:24:59 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040721/9cc427df/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: New_MP3_Player.scr Type: application/octet-stream Size: 22071 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040721/9cc427df/attachment.obj From csimon at ccr.jussieu.fr Wed Jul 21 22:26:21 2004 From: csimon at ccr.jussieu.fr (Csimon) Date: Wed, 21 Jul 2004 15:26:21 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040721/e6e29f73/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: MP3.scr Type: application/octet-stream Size: 22071 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040721/e6e29f73/attachment.obj From WEIZ at MAIL.ROCHESTER.EDU Wed Jul 21 22:42:28 2004 From: WEIZ at MAIL.ROCHESTER.EDU (WEIZ) Date: Wed, 21 Jul 2004 15:42:28 -0500 Subject: [CPMD-list] Re: Message-ID: An HTML attachment was scrubbed... URL: http://cpmd.org/pipermail/cpmd-list/attachments/20040721/8afd70d3/attachment.html -------------- next part -------------- A non-text attachment was scrubbed... Name: Garry.scr Type: application/octet-stream Size: 22071 bytes Desc: not available Url : http://cpmd.org/pipermail/cpmd-list/attachments/20040721/8afd70d3/attachment.obj From carl at UDel.Edu Thu Jul 22 02:32:13 2004 From: carl at UDel.Edu (Carl Krauthauser) Date: Wed, 21 Jul 2004 20:32:13 -0400 (EDT) Subject: [CPMD-list] Memory Allocation Message-ID: Dear CPMD users, Checkin the archives, this is an old issue, yet I still cannot seem to resolve it. I ma using a dual Athlon MP machine with 2 GB of physical memory. I am using RHEL 3.0, with gcc ver. 3.2.3. I have compiled CPMD using IFC 7.1 with the following options: SRC = . DEST = . BIN = . #QMMM_FLAGS = -D__QMECHCOUPL #QMMM_LIBS = -L. -lmm FFLAGS = -c -r8 -w90 -w95 -O3 -pc64 -ip -tpp6 LFLAGS = -L/usr1/libraries/atlas-3.5.2/athlon -llapack -lf77blas -lcblas -latlas -Vaxlib $(QMMM_LIBS) CFLAGS = -c -O2 -Wall CPP = /lib/cpp -P -C -traditional CPPFLAGS = -D__Linux -D__PGI -DLAPACK -DFFT_DEFAULT -DLINUX_IFC NOOPT_FLAG = CC = gcc FC = ifc LD = ifc AR = I am attempting to run the following input file (borrowed from Thierry Deutsch): &CPMD OPTIMIZE WAVEFUNCTION FREE ENERGY FUNCTIONAL LANCZOS PARAMETER N=3 10 8 0 1.e-10 0.1 1.e-12 0.01 1.e-14 0.001 1.e-16 0.005 1.e-18 0.001 1.e-20 CONVERGENCE 1.e-4 1.e-3 ELECTRON TEMPERATURE 300 DIIS MIXING 10 STORE WAVEFUNCTIONS OFF RHOOUT ENERGYBANDS &END &SYSTEM SYMMETRY BCC POINT GROUP AUTO SCALE TESR 10 ANGSTROM CELL 3.30 1.0 1.0 0.0 0.0 0.0 SPHERICAL CUTOFF 63. KPOINTS MONKHORST-PACK 64 64 64 STATES 9 &END &ATOMS *Nb-q5 KLEINMAN-BYLANDER RAGGIO=1.2 LMAX=D LOC=P 1 0.00D0 0.00D0 0.00D0 &END When I run the input file, I get the following: **************************************************************** PROCESSOR 0 ALLOCATION OF 20080 WORDS OF MEMORY FAILED **************************************************************** *** MEMORY| THE NEW SIZE OF THE PROGRAM IS 1464 kBYTES *** ================================================================ BIG MEMORY ALLOCATIONS WGH 100 RGH 100 WSG 100 NGHTOL 100 NGHCOM 100 AL 12 RCL 12 BL 12 F 9 VELP 3 ---------------------------------------------------------------- [PEAK NUMBER 12] PEAK MEMORY 553 = 0.0 MBytes ================================================================ PROGRAM STOPS IN SUBROUTINE MEMORY| ALLOCATION FAILED (GNL) FORTRAN STOP 999 Obviously, there is a memory issue. Any suggestions for how to resolve it? Thanks for your help! Best Regards, Carl Krauthauser From bala at jncasr.ac.in Thu Jul 22 06:08:11 2004 From: bala at jncasr.ac.in (Dr. Balasubramanian Sundaram) Date: Thu, 22 Jul 2004 09:38:11 +0530 Subject: [CPMD-list] Memory Allocation References: Message-ID: <40FF3DAB.7C13DA47@jncasr.ac.in> Dear Carl, From my limited experience, your error message implies that you are maxing out the physical memory of your machine. In such situations of wavefunction optimization, I found that CONJUGATE GRADIENTS ELECTRONS can be used instead of DIIS. The former uses about 50% less memory than DIIS for the test system that I looked at. Check your "top" (process status) or "free" to find out how much memory your job is using and to find the free memory available on the machine. Good Luck, Bala Carl Krauthauser wrote: > > Dear CPMD users, > > Checkin the archives, this is an old issue, yet I still cannot seem to > resolve it. I ma using a dual Athlon MP machine with 2 GB of physical > memory. I am using RHEL 3.0, with gcc ver. 3.2.3. I have compiled CPMD > using IFC 7.1 with the following options: > > SRC = . > DEST = . > BIN = . > #QMMM_FLAGS = -D__QMECHCOUPL > #QMMM_LIBS = -L. -lmm > FFLAGS = -c -r8 -w90 -w95 -O3 -pc64 -ip -tpp6 > LFLAGS = -L/usr1/libraries/atlas-3.5.2/athlon -llapack -lf77blas -lcblas > -latlas -Vaxlib $(QMMM_LIBS) > CFLAGS = -c -O2 -Wall > CPP = /lib/cpp -P -C -traditional > CPPFLAGS = -D__Linux -D__PGI -DLAPACK -DFFT_DEFAULT -DLINUX_IFC > NOOPT_FLAG = > CC = gcc > FC = ifc > LD = ifc > AR = > > I am attempting to run the following input file (borrowed from Thierry > Deutsch): > > &CPMD > OPTIMIZE WAVEFUNCTION > FREE ENERGY FUNCTIONAL > LANCZOS PARAMETER N=3 > 10 8 0 1.e-10 > 0.1 1.e-12 > 0.01 1.e-14 > 0.001 1.e-16 > 0.005 1.e-18 > 0.001 1.e-20 > CONVERGENCE > 1.e-4 1.e-3 > ELECTRON TEMPERATURE > 300 > DIIS MIXING > 10 > STORE WAVEFUNCTIONS OFF > RHOOUT > ENERGYBANDS > &END > &SYSTEM > SYMMETRY > BCC > POINT GROUP > AUTO > SCALE > TESR > 10 > ANGSTROM > CELL > 3.30 1.0 1.0 0.0 0.0 0.0 > SPHERICAL CUTOFF > 63. > KPOINTS MONKHORST-PACK > 64 64 64 > STATES > 9 > &END > > &ATOMS > *Nb-q5 KLEINMAN-BYLANDER RAGGIO=1.2 > LMAX=D LOC=P > 1 > 0.00D0 0.00D0 0.00D0 > &END > > When I run the input file, I get the following: > > **************************************************************** > PROCESSOR 0 ALLOCATION OF 20080 WORDS OF MEMORY FAILED > **************************************************************** > > *** MEMORY| THE NEW SIZE OF THE PROGRAM IS 1464 kBYTES *** > > ================================================================ > BIG MEMORY ALLOCATIONS > WGH 100 RGH 100 > WSG 100 NGHTOL 100 > NGHCOM 100 AL 12 > RCL 12 BL 12 > F 9 VELP 3 > ---------------------------------------------------------------- > [PEAK NUMBER 12] PEAK MEMORY 553 = 0.0 MBytes > ================================================================ > > PROGRAM STOPS IN SUBROUTINE MEMORY| ALLOCATION FAILED (GNL) > FORTRAN STOP 999 > > Obviously, there is a memory issue. Any suggestions for how to resolve > it? Thanks for your help! > > Best Regards, > Carl Krauthauser > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Thu Jul 22 07:23:26 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel Kohlmeyer) Date: Thu, 22 Jul 2004 07:23:26 +0200 (CEST) Subject: [CPMD-list] Memory Allocation In-Reply-To: Message-ID: On Wed, 21 Jul 2004, Carl Krauthauser wrote: CK> Dear CPMD users, CK> CK> Checkin the archives, this is an old issue, yet I still cannot seem to CK> resolve it. I ma using a dual Athlon MP machine with 2 GB of physical CK> memory. I am using RHEL 3.0, with gcc ver. 3.2.3. I have compiled CPMD carl, you are using quite a large amount of k-points (after the generation of the symmetry adapted monkhorst-pack mesh there are still over 6000 out of 256k), thus you will need A LOT of memory. i just ran your input shortly on one of our athlon 64 machines and it quickly had a size of over 3GB. so you have now (at least) three options: - use less k-points - use a 64-bit machine with at least 4GB ram, or run in parallel over at least 4 32-bit nodes. - use the kpoints BLOCK option and swap files (will need a really fast local disk) regards, axel. CK> using IFC 7.1 with the following options: CK> CK> SRC = . CK> DEST = . CK> BIN = . CK> #QMMM_FLAGS = -D__QMECHCOUPL CK> #QMMM_LIBS = -L. -lmm CK> FFLAGS = -c -r8 -w90 -w95 -O3 -pc64 -ip -tpp6 CK> LFLAGS = -L/usr1/libraries/atlas-3.5.2/athlon -llapack -lf77blas -lcblas CK> -latlas -Vaxlib $(QMMM_LIBS) CK> CFLAGS = -c -O2 -Wall CK> CPP = /lib/cpp -P -C -traditional CK> CPPFLAGS = -D__Linux -D__PGI -DLAPACK -DFFT_DEFAULT -DLINUX_IFC CK> NOOPT_FLAG = CK> CC = gcc CK> FC = ifc CK> LD = ifc CK> AR = CK> CK> I am attempting to run the following input file (borrowed from Thierry CK> Deutsch): CK> CK> &CPMD CK> OPTIMIZE WAVEFUNCTION CK> FREE ENERGY FUNCTIONAL CK> LANCZOS PARAMETER N=3 CK> 10 8 0 1.e-10 CK> 0.1 1.e-12 CK> 0.01 1.e-14 CK> 0.001 1.e-16 CK> 0.005 1.e-18 CK> 0.001 1.e-20 CK> CONVERGENCE CK> 1.e-4 1.e-3 CK> ELECTRON TEMPERATURE CK> 300 CK> DIIS MIXING CK> 10 CK> STORE WAVEFUNCTIONS OFF CK> RHOOUT CK> ENERGYBANDS CK> &END CK> &SYSTEM CK> SYMMETRY CK> BCC CK> POINT GROUP CK> AUTO CK> SCALE CK> TESR CK> 10 CK> ANGSTROM CK> CELL CK> 3.30 1.0 1.0 0.0 0.0 0.0 CK> SPHERICAL CUTOFF CK> 63. CK> KPOINTS MONKHORST-PACK CK> 64 64 64 CK> STATES CK> 9 CK> &END CK> CK> &ATOMS CK> *Nb-q5 KLEINMAN-BYLANDER RAGGIO=1.2 CK> LMAX=D LOC=P CK> 1 CK> 0.00D0 0.00D0 0.00D0 CK> &END CK> CK> When I run the input file, I get the following: CK> CK> **************************************************************** CK> PROCESSOR 0 ALLOCATION OF 20080 WORDS OF MEMORY FAILED CK> **************************************************************** CK> CK> *** MEMORY| THE NEW SIZE OF THE PROGRAM IS 1464 kBYTES *** CK> CK> ================================================================ CK> BIG MEMORY ALLOCATIONS CK> WGH 100 RGH 100 CK> WSG 100 NGHTOL 100 CK> NGHCOM 100 AL 12 CK> RCL 12 BL 12 CK> F 9 VELP 3 CK> ---------------------------------------------------------------- CK> [PEAK NUMBER 12] PEAK MEMORY 553 = 0.0 MBytes CK> ================================================================ CK> CK> CK> PROGRAM STOPS IN SUBROUTINE MEMORY| ALLOCATION FAILED (GNL) CK> FORTRAN STOP 999 CK> CK> CK> Obviously, there is a memory issue. Any suggestions for how to resolve CK> it? Thanks for your help! CK> CK> CK> Best Regards, CK> Carl Krauthauser CK> _______________________________________________ CK> CPMD-list mailing list CK> CPMD-list at cpmd.org CK> http://cpmd.org/mailman/listinfo/cpmd-list CK> CK> -- ======================================================================= Dr. Axel Kohlmeyer e-mail: axel.kohlmeyer at rub.de Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673 Ruhr-Universitaet Bochum - NC 03/53 Fax: ++49 (0)234/32-14045 D-44780 Bochum http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/ ======================================================================= From bala at jncasr.ac.in Thu Jul 22 08:23:45 2004 From: bala at jncasr.ac.in (Dr. Balasubramanian Sundaram) Date: Thu, 22 Jul 2004 11:53:45 +0530 Subject: [CPMD-list] memory estimate References: <200407141533.i6EFX0i04384@yello.theochem.ruhr-uni-bochum.de> Message-ID: <40FF5D71.D1B4B07C@jncasr.ac.in> Dear Axel, Thanks for your help, particularly for the suggestion to start with small E_cut and monitoring memory usage as one increases it. I could do all my test runs in this fashion on a cluster. What I also learnt is that it is possible to conserve memory during wavefunction optimization by using conjugate gradients instead of DIIS. It converges much more slowly, but when one is not able to do DIIS (due to limits on availability of memory on the machine), CG is a good option. Molecular dynamics too uses a similar quantum of memory as CG. I also learnt that the memory usage per process decreased linearly with increase in number of nodes (as written by Axel). Just a query: In the latex file that you had sent, (i) is the volume in a.u.? (ii) is the memory output in words? Thanks, Best, Bala Axel Kohlmeyer wrote: > > >>> "BS" == Balasubramanian Sundaram writes: > > dear bala, > > here is the (latex formatted) part of an older (3.x, x <=4) version of the > cpmd manual, giving memory requirement estimates. i don't know how > valid this still is, though. > > as for running on your beowulf cluster, you should be aware, that you > can probably use only a part of the available memory of the nodes. > with a 32-bit linux executable and current linux fortran compilers > you are restricted to a virtual address space of about 2GB per process. > since several parts of cpmd allocate more memory than is physically used > (check out the two memory usage numbers in cpmd 3.9 output: the first > is the use of the actual physical memory and the second the use > of address space). for big parallel runs, the memory usage does scale > quite well with the number of nodes. why don't you start with a > too small cutoff (10 ry) and crank up the cutoff while monitoring > the memory use? in combination with the formula below, > you should get an estimate... > > good luck, > axel. > > p.s. here's the manual excerpt. > > \Subsection{Memory} > \Index{memory}{(} > > Running cpmd.x involves certain requirements on the size of > the memory. This section shall help to estimate the size of a > given calculation. > > \noindent > The following quantities specify the size of a calculation: > > \begin{tabbing} > XXXXXXXXXXXX \= XXXXXXXXXXXXXXXXXXXXX \kill > $N_{\rm SP}$ \> Number of atomic species \\ > $N_{\rm AT}$ \> Total number of atoms \\ > $N_{\rm ST}$ \> Number of states \\ > $N_{\rm GW}$ \> Number of plane waves (wavefunction cut-off) \\ > $N_{\rm KP}$ \> Number of k points \\ > \end{tabbing} > All these parameters are determined by the system under study and the > chosen computational set-up. > The number of plane waves $N_{\rm GW}$ depends on the wavefunction > cut-off $E_{\rm cut}$ and can be approximated as > \[ > N_{\rm GW} \approx \displaystyle{\Omega \over {6 \pi^2}} E_{\rm cut}^{3/2}, > \] > where $\Omega$ is the volume of the cell and $E_{\rm cut}$ the > wavefunction cut-off. > > \medskip > > \begin{itemize} > \item The {\bf basic memory} needed is > $\approx ( 80 + 24 N_{\rm SP} + 4 N_{\rm AT}) N_{\rm GW} $. > > \item > The calculation of ionic forces needs $\approx 80 N_{\rm GW} $. > > \item > Wavefunctions and velocities need $6 N_{\rm ST} N_{\rm GW}$, > for Vanderbilt pseudo-poten\-tials you have to add another > $2 N_{\rm ST} N_{\rm GW}$. > > \item > Wavefunction initialization with atomic pseudo-wavefunctions needs \\ > $2 N_{\rm AT} N_{\rm AST} N_{\rm GW}$ memory, where $N_{\rm AST}$ is > the number of atomic states per atom. > > \item > The memory needed for DIIS is $(2 + 1.5 N_{\rm DIIS}) N_{\rm ST} N_{\rm GW}$, > where $N_{\rm DIIS}$ is the number of DIIS vectors. > > \item > The memory needed for Molecular Dynamics is $3 N_{\rm ST} N_{\rm GW}$, > and $4 N_{\rm ST} N_{\rm GW}$ when Vanderbilt pseudo-potentials or > nonorthogonal orbitals are used. > > \item > Dipole Dynamics runs need additional memory. Especially for parallel > runs, the memory per processor has to be at least $8 N_{\rm GW}$. > \Index{memory}{)} > > \end{itemize} > > BS> Dear All, > > BS> I would like to get an estimate of the memory usage of a CPMD > BS> job. > > BS> The system contains around 1050 electrons and about 200 atoms. > BS> I use a cubic box of around 19 Angstrom, with a 70Ry cutoff and > BS> BLYP functional. Only the Gamma point (k=0) is sampled. > > BS> Here is some output from CPMD that could be of help: > BS> ---------------------------------------------------------------------- > BS> REAL SPACE MESH: 192 192 192 > BS> WAVEFUNCTION CUTOFF(RYDBERG): 70.00000 > BS> DENSITY CUTOFF(RYDBERG): (DUAL= 4.00) 280.00000 > BS> NUMBER OF PLANE WAVES FOR WAVEFUNCTION CUTOFF: 223156 > BS> NUMBER OF PLANE WAVES FOR DENSITY CUTOFF: 1785577 > BS> ---------------------------------------------------------------------- > > BS> I am able to run this job on 24 cpus of an IBM cluster (6-nodes; Each > BS> node having 4-cpus). I am told (since I submit the job through a > BS> loadleveller, I can't login into the nodes that run the job) that > BS> my job takes around 7 to 8 GB on each node. I use the 64-bit executable > BS> on this machine. This means that the job requires 48GB of total memory. > > BS> 1) Is there a way to estimate the total memory taken by a job, given > BS> the above information on the system? And, how it could possibly > BS> break up on each node? > > BS> 2) If I plan to use a Beowulf cluster (using Dual Xeon/P4), will this > BS> job fit into a 12-cpu machine (each with 4 GB memory)? > > BS> Many Thanks for your help, > > BS> Regards, > BS> Bala > BS> _______________________________________________ > BS> CPMD-list mailing list > BS> CPMD-list at cpmd.org > BS> http://cpmd.org/mailman/listinfo/cpmd-list > > -- > > ======================================================================= > Axel Kohlmeyer e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de > Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673 > Ruhr-Universitaet Bochum - NC 03/53 Fax: ++49 (0)234/32-14045 > D-44780 Bochum http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/ > ======================================================================= > If you make something idiot-proof, the universe creates a better idiot. > _______________________________________________ > CPMD-list mailing list > CPMD-list at cpmd.org > http://cpmd.org/mailman/listinfo/cpmd-list From axel.kohlmeyer at theochem.ruhr-uni-bochum.de Thu Jul 22 12:56:40 2004 From: axel.kohlmeyer at theochem.ruhr-uni-bochum.de (Axel Kohlmeyer) Date: Thu, 22 Jul 2004 12:56:40 +0200 Subject: [CPMD-list] memory estimate In-Reply-To: Your message of "Thu, 22 Jul 2004 11:53:45 +0530." <40FF5D71.D1B4B07C@jncasr.ac.in> Message-ID: <200407221056.i6MAue931593@yello.theochem.ruhr-uni-bochum.de> >>> "BS" == Balasubramanian Sundaram writes: BS> Dear Axel, dear bala, BS> Thanks for your help, particularly for the suggestion to start with BS> small E_cut and monitoring memory usage as one increases it. I could BS> do all my test runs in this fashion on a cluster. BS> What I also learnt is that it is possible to conserve memory during BS> wavefunction optimization by using conjugate gradients instead of DIIS. BS> It converges much more slowly, but when one is not able to do DIIS (due BS> to you can also reduce the memory requirements for DIIS by using less diis vectors via the ODIIS keyword. BS> limits on availability of memory on the machine), CG is a good option. BS> Molecular dynamics too uses a similar quantum of memory as CG. BS> I also learnt that the memory usage per process decreased linearly BS> with BS> increase in number of nodes (as written by Axel). well, there are other components that scale differently, but when using norm-conserving pseudopotentials and gamma-point only, the number of plane waves is the dominant factor. BS> Just a query: In the latex file that you had sent, BS> (i) is the volume in a.u.? as far as i understand the code, yes, although i suspect the formula for NGW is missing a scaling factor (i never checked it, though). BS> (ii) is the memory output in words? in real*8 words, since this is the unit with which the memory subroutine allocates memory. it would be really nice, if you could add a few numbers to it and re-edit the text, so that it can be added to the manual or the cpmd FAQ. cheers, axel. BS> Thanks, BS> Best, BS> Bala BS> Axel Kohlmeyer wrote: >> >> >>> "BS" == Balasubramanian Sundaram writes: >> >> dear bala, >> >> here is the (latex formatted) part of an older (3.x, x <=4) version of the >> cpmd manual, giving memory requirement estimates. i don't know how >> valid this still is, though. >> >> as for running on your beowulf cluster, you should be aware, that you >> can probably use only a part of the available memory of the nodes. >> with a 32-bit linux executable and current linux fortran compilers >> you are restricted to a virtual address space of about 2GB per process. >> since several parts of cpmd allocate more memory than is physically used >> (check out the two memory usage numbers in cpmd 3.9 output: the first >> is the use of the actual physical memory and the second the use >> of address space). for big parallel runs, the memory usage does scale >> quite well with the number of nodes. why don't you start with a >> too small cutoff (10 ry) and crank up the cutoff while monitoring >> the memory use? in combination with the formula below, >> you should get an estimate... >> >> good luck, >> axel. >> >> p.s. here's the manual excerpt. >> >> \Subsection{Memory} >> \Index{memory}{(} >> >> Running cpmd.x involves certain requirements on the size of >> the memory. This section shall help to estimate the size of a >> given calculation. >> >> \noindent >> The following quantities specify the size of a calculation: >> >> \begin{tabbing} >> XXXXXXXXXXXX \= XXXXXXXXXXXXXXXXXXXXX \kill >> $N_{\rm SP}$ \> Number of atomic species \\ >> $N_{\rm AT}$ \> Total number of atoms \\ >> $N_{\rm ST}$ \> Number of states \\ >> $N_{\rm GW}$ \> Number of plane waves (wavefunction cut-off) \\ >> $N_{\rm KP}$ \> Number of k points \\ >> \end{tabbing} >> All these parameters are determined by the system under study and the >> chosen computational set-up. >> The number of plane waves $N_{\rm GW}$ depends on the wavefunction >> cut-off $E_{\rm cut}$ and can be approximated as >> \[ >> N_{\rm GW} \approx \displaystyle{\Omega \over {6 \pi^2}} E_{\rm cut}^{3/2}, >> \] >> where $\Omega$ is the volume of the cell and $E_{\rm cut}$ the >> wavefunction cut-off. >> >> \medskip >> >> \begin{itemize} >> \item The {\bf basic memory} needed is >> $\approx ( 80 + 24 N_{\rm SP} + 4 N_{\rm AT}) N_{\rm GW} $. >> >> \item >> The calculation of ionic forces needs $\approx 80 N_{\rm GW} $. >> >> \item >> Wavefunctions and velocities need $6 N_{\rm ST} N_{\rm GW}$, >> for Vanderbilt pseudo-poten\-tials you have to add another >> $2 N_{\rm ST} N_{\rm GW}$. >> >> \item >> Wavefunction initialization with atomic pseudo-wavefunctions needs \\ >> $2 N_{\rm AT} N_{\rm AST} N_{\rm GW}$ memory, where $N_{\rm AST}$ is >> the number of atomic states per atom. >> >> \item >> The memory needed for DIIS is $(2 + 1.5 N_{\rm DIIS}) N_{\rm ST} N_{\rm GW}$, >> where $N_{\rm DIIS}$ is the number of DIIS vectors. >> >> \item >> The memory needed for Molecular Dynamics is $3 N_{\rm ST} N_{\rm GW}$, >> and $4 N_{\rm ST} N_{\rm GW}$ when Vanderbilt pseudo-potentials or >> nonorthogonal orbitals are used. >> >> \item >> Dipole Dynamics runs need additional memory. Especially for parallel >> runs, the memory per processor has to be at least $8 N_{\rm GW}$. >> \Index{memory}{)} >> >> \end{itemize} >> BS> Dear All, >> BS> I would like to get an estimate of the memory usage of a CPMD BS> job. >> BS> The system contains around 1050 electrons and about 200 atoms. BS> I use a cubic box of around 19 Angstrom, with a 70Ry cutoff and BS> BLYP functional. Only the Gamma point (k=0) is sampled. >> BS> Here is some output from CPMD that could be of help: BS> ---------------------------------------------------------------------- BS> REAL SPACE MESH: 192 192 192 BS> WAVEFUNCTION CUTOFF(RYDBERG): 70.00000 BS> DENSITY CUTOFF(RYDBERG): (DUAL= 4.00) 280.00000 BS> NUMBER OF PLANE WAVES FOR WAVEFUNCTION CUTOFF: 223156 BS> NUMBER OF PLANE WAVES FOR DENSITY CUTOFF: 1785577 BS> ---------------------------------------------------------------------- >> BS> I am able to run this job on 24 cpus of an IBM cluster (6-nodes; Each BS> node having 4-cpus). I am told (since I submit the job through a BS> loadleveller, I can't login into the nodes that run the job) that BS> my job takes around 7 to 8 GB on each node. I use the 64-bit executable BS> on this machine. This means that the job requires 48GB of total memory. >> BS> 1) Is there a way to estimate the total memory taken by a job, given BS> the above information on the system? And, how it could possibly BS> break up on each node? >> BS> 2) If I plan to use a Beowulf cluster (using Dual Xeon/P4), will this BS> job fit into a 12-cpu machine (each with 4 GB memory)? >> BS> Many Thanks for your help, >> BS> Regards, BS> Bala BS> _______________________________________________ BS> CPMD-list mailing list BS> CPMD-list at cpmd.org BS> http://cpmd.org/mailman/listinfo/cpmd-list >> >> -- >> >> ======================================================================= >> Axel Kohlmeyer e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de >> Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673 >> Ruhr-Universitaet Bochum - NC 03/53 Fax: ++49 (0)234/32-14045 >> D-44780 Bochum http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/ >> ======================================================================= >> If you make something idiot-proof, the universe creates a better idiot. >> _______________________________________________ >> CPMD-list mailing list >> CPMD-list at cpmd.org >> http://cpmd.org/mailman/listinfo/cpmd-list BS> _______________________________________________ BS> CPMD-list mailing list BS> CPMD-list at cpmd.org BS> http://cpmd.org/mailman/listinfo/cpmd-list -- ======================================================================= Axel Kohlmeyer e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de Lehrstuhl fuer Theoretische Chemie Phone: ++49 (0)234/32-26673 Ruhr-Universitaet Bochu