Re: [SIESTA-L] Question about LDA and GGA pseudopotential
On Fri, 2 Feb 2007, Dr. Y. Lin wrote: | BTW: What is the MeshCutoff? I read the manual, and it says: | | MeshCutoff | (real energy): Defines the equivalent plane wave cutoff for the grid. | I thought siesta is not using plane wave basis. Why does the code needs this? | Could somebody point out where I shall get more info? Dear Jin Zhang: chack the basic Siesta paper JPCM 14, 2745, Sec.6 (The paper is algo good for other questions) Concerning the previous discussion, LDA vs. GGA bond lenth, the effect of grid etc: the fine real-space grid is needed in order to stabilize, i.e., to converge (in terms of grid cutoff) your results like total energy, forces and everything that is related to them, including bond lengths. However, a good grid offers no fix for everything else and would not repair what was due to bad pseudopotential, or basis, or both. For system like CH4 one can easily make a very accurate test for LDA truth or GGA truth with a good all-electron code (or, such benchmarks are certainly available), and then you'll see whether your deviation from experiment is a known LDA fault, or a fault in your SIESTA setup which fails to reproduce the LDA truth. Best regards, Andrei Postnikov +-- Dr. Andrei Postnikov Tel. +33-387315873 - mobile +33-666784053 ---+ | Paul Verlaine University - Institute de Physique Electronique et Chimie, | | Laboratoire de Physique des Milieux Denses, 1 Bd Arago, F-57078 Metz, France | +-- [EMAIL PROTECTED] http://www.home.uni-osnabrueck.de/apostnik/ --+ | | Thanks. | | | Jin Zhang | | On 2/2/07, Dr. Y. Lin [EMAIL PROTECTED] wrote: | | On Thu, 1 Feb 2007, Oleksandr Voznyy wrote: | |1) Why is GGA pseudo-potential gives worse result than LDA? |What do you mean worse? |A well converged calculation (i.e. converged in E cutoff, k-grid, and |converged vs pseudo cutoff radii as well) |LDA SHOULD give bonds (or lattice constants) 1% smaller than expt (while |binding energies are strongly overestimated) |while GGA usually overestimates bonds by about 2% (while binding | energies |(for molecules) are much closer to expt). | | First, thanks for the info. Let me explain why I think GGA gives worse | result. | | Let's take CH4 as an example. Here's the fdf input file I have | - Start of the file --- | SystemName Methane Broyden Optimization | SystemLabel general | NumberOfAtoms 5 | NumberOfSpecies 2 | | MeshCutoff 50 Ry | | %block ChemicalSpeciesLabel | 1 6 C # Species index, atomic number, species label | 2 1 H | %endblock ChemicalSpeciesLabel | | #PAO.BasisSize SZ | %blockPAO.BasisSizes |C DZP |H DZP | %endblock PAO.BasisSizes | | AtomicCoordinatesFormat Ang | %block AtomicCoordinatesAndAtomicSpecies |1.0582 0.9353 0.8103 1 |1.4145 1.5662 0. 2 |1.2065 1.4452 1.7588 2 |0. 0.7294 0.6710 2 |1.6121 0. 0.8114 2 | %endblock AtomicCoordinatesAndAtomicSpecies | | XC.functional GGA | XC.authors PBE | PAO.EnergyShift 0.02 Ry | | DM.Number.Pulay 3 | | WriteForces T | | MD.TypeOfRun Broyden | MD.NumCGSteps 400 | MD.Broyden.History.Steps 6 | MD.Broyden.Initial.Inverse.Jacobian 1.0 | --- End of the file --- | | This input file is used along with GGA pseudo potential of Carbon. | For LDA pseudo potential case, I just change the line with Carbon | at block PAO.BasisSizes to | | CDZ | | Basically, this setting reduces the total basis of Carbon atom from 13 to | 8. | | Then I ran siesta to relax the CH4 and measure C-H bond length after | relaxation: | | Exp. LDACarbon GGACarbonCCCBDB calculated |PBEPEB/6-331G** | 1.094 1.097 1.10 1.0977 | | It seems LDA pseudo potential did give a shorter bond-length while still | longer than experimental value. GGA pseudo potential gives an even longer | bond-length, which is obviously, a worse result. | | |2) Is it safe to use LDA pseudo-potential while setting GGA-PBE as |x-correlation in input.fdf? |No, it is not recommended. |However, you can use the cutoff radii from LDA pseudo to generate the | new GGA |pseudo. Have you tried to compare 2 input files that you have? | | | I guess it is not very good to use LDA pseudo-potential while setting | GGA-PBE as x-correlation. Thanks for confirming this. However, I'm not | quite sure I understand your suggestion. | | Let me clarify this: I'm not trying to generate pseudo-potential myself. | So, all the input I said are for Molecular Dynamics or Relaxation. Under | this situation, I'd say, basically, my input files are the same except for | the basis sets. For GGApseudo, I set
Re: [SIESTA-L] Question about LDA and GGA pseudopotential
On Fri, 2 Feb 2007, Jin Zhang wrote: Your meshcutoff is too low (especially for GGA). Try to increase it to 200 or more until convergence reached. Also, a 0.01 deviation is not that bad. Thanks for the suggestion. I'll increase this in my next simulation. However, I've tested the result vs MeshCutoff and posted it previously. Both the energy and bond length changes very little: -- Quote -- Thanks for the suggestion. I just tried to change Meshcutoff from 50 Ry to 200 Ry using GGA pseudo potential. The FreeEnergy at first ionic step is -210.338880 and -210.339573 respectively. Difference is about 1.e-3 eV. However, the optimized bond length between C-H are 1.108 and 1.107 respectively. Doesn't seem to be a good improvement. -- End of Quote -- I understand that the small change of optimized bond length is not a big issue. However, what I don't understand is why GGA gives worse result as counter-intuitive. This makes me hard on choosing the right one for my on-going simulation. BTW: What is the MeshCutoff? I read the manual, and it says: MeshCutoff (real energy): Defines the equivalent plane wave cutoff for the grid. I thought siesta is not using plane wave basis. Why does the code needs this? Could somebody point out where I shall get more info? Thanks. Jin Zhang On 2/2/07, Dr. Y. Lin [EMAIL PROTECTED] wrote: On Thu, 1 Feb 2007, Oleksandr Voznyy wrote: 1) Why is GGA pseudo-potential gives worse result than LDA? What do you mean worse? A well converged calculation (i.e. converged in E cutoff, k-grid, and converged vs pseudo cutoff radii as well) LDA SHOULD give bonds (or lattice constants) 1% smaller than expt (while binding energies are strongly overestimated) while GGA usually overestimates bonds by about 2% (while binding energies (for molecules) are much closer to expt). First, thanks for the info. Let me explain why I think GGA gives worse result. Let's take CH4 as an example. Here's the fdf input file I have - Start of the file --- SystemName Methane Broyden Optimization SystemLabel general NumberOfAtoms 5 NumberOfSpecies 2 MeshCutoff 50 Ry %block ChemicalSpeciesLabel 1 6 C # Species index, atomic number, species label 2 1 H %endblock ChemicalSpeciesLabel #PAO.BasisSize SZ %blockPAO.BasisSizes C DZP H DZP %endblock PAO.BasisSizes AtomicCoordinatesFormat Ang %block AtomicCoordinatesAndAtomicSpecies 1.0582 0.9353 0.8103 1 1.4145 1.5662 0. 2 1.2065 1.4452 1.7588 2 0. 0.7294 0.6710 2 1.6121 0. 0.8114 2 %endblock AtomicCoordinatesAndAtomicSpecies XC.functional GGA XC.authors PBE PAO.EnergyShift 0.02 Ry DM.Number.Pulay 3 WriteForces T MD.TypeOfRun Broyden MD.NumCGSteps 400 MD.Broyden.History.Steps 6 MD.Broyden.Initial.Inverse.Jacobian 1.0 --- End of the file --- This input file is used along with GGA pseudo potential of Carbon. For LDA pseudo potential case, I just change the line with Carbon at block PAO.BasisSizes to CDZ Basically, this setting reduces the total basis of Carbon atom from 13 to 8. Then I ran siesta to relax the CH4 and measure C-H bond length after relaxation: Exp. LDACarbon GGACarbonCCCBDB calculated PBEPEB/6-331G** 1.094 1.097 1.10 1.0977 It seems LDA pseudo potential did give a shorter bond-length while still longer than experimental value. GGA pseudo potential gives an even longer bond-length, which is obviously, a worse result. 2) Is it safe to use LDA pseudo-potential while setting GGA-PBE as x-correlation in input.fdf? No, it is not recommended. However, you can use the cutoff radii from LDA pseudo to generate the new GGA pseudo. Have you tried to compare 2 input files that you have? I guess it is not very good to use LDA pseudo-potential while setting GGA-PBE as x-correlation. Thanks for confirming this. However, I'm not quite sure I understand your suggestion. Let me clarify this: I'm not trying to generate pseudo-potential myself. So, all the input I said are for Molecular Dynamics or Relaxation. Under this situation, I'd say, basically, my input files are the same except for the basis sets. For GGApseudo, I set DZP for Carbon; and for LDApseudo, I set DZ. Thanks for more info. -- You Lin Department of Physics University of South Florida 4202 East Fowler Avenue Tampa, FL 33620 Tel: (813)396-9220 [Office] Homepage: http://shell.cas.usf.edu/~ylin -- You Lin Department of Physics University of South Florida 4202 East Fowler Avenue Tampa, FL 33620 Tel: (813)396-9220 [Office] Homepage: http://shell.cas.usf.edu/~ylin
Re: [SIESTA-L] Question about LDA and GGA pseudopotential
Your meshcutoff is too low (especially for GGA). Try to increase it to 200 or more until convergence reached. Also, a 0.01 deviation is not that bad. Jin Zhang On 2/2/07, Dr. Y. Lin [EMAIL PROTECTED] wrote: On Thu, 1 Feb 2007, Oleksandr Voznyy wrote: 1) Why is GGA pseudo-potential gives worse result than LDA? What do you mean worse? A well converged calculation (i.e. converged in E cutoff, k-grid, and converged vs pseudo cutoff radii as well) LDA SHOULD give bonds (or lattice constants) 1% smaller than expt (while binding energies are strongly overestimated) while GGA usually overestimates bonds by about 2% (while binding energies (for molecules) are much closer to expt). First, thanks for the info. Let me explain why I think GGA gives worse result. Let's take CH4 as an example. Here's the fdf input file I have - Start of the file --- SystemName Methane Broyden Optimization SystemLabel general NumberOfAtoms 5 NumberOfSpecies 2 MeshCutoff 50 Ry %block ChemicalSpeciesLabel 1 6 C # Species index, atomic number, species label 2 1 H %endblock ChemicalSpeciesLabel #PAO.BasisSize SZ %blockPAO.BasisSizes C DZP H DZP %endblock PAO.BasisSizes AtomicCoordinatesFormat Ang %block AtomicCoordinatesAndAtomicSpecies 1.0582 0.9353 0.8103 1 1.4145 1.5662 0. 2 1.2065 1.4452 1.7588 2 0. 0.7294 0.6710 2 1.6121 0. 0.8114 2 %endblock AtomicCoordinatesAndAtomicSpecies XC.functional GGA XC.authors PBE PAO.EnergyShift 0.02 Ry DM.Number.Pulay 3 WriteForces T MD.TypeOfRun Broyden MD.NumCGSteps 400 MD.Broyden.History.Steps 6 MD.Broyden.Initial.Inverse.Jacobian 1.0 --- End of the file --- This input file is used along with GGA pseudo potential of Carbon. For LDA pseudo potential case, I just change the line with Carbon at block PAO.BasisSizes to CDZ Basically, this setting reduces the total basis of Carbon atom from 13 to 8. Then I ran siesta to relax the CH4 and measure C-H bond length after relaxation: Exp. LDACarbon GGACarbonCCCBDB calculated PBEPEB/6-331G** 1.094 1.097 1.10 1.0977 It seems LDA pseudo potential did give a shorter bond-length while still longer than experimental value. GGA pseudo potential gives an even longer bond-length, which is obviously, a worse result. 2) Is it safe to use LDA pseudo-potential while setting GGA-PBE as x-correlation in input.fdf? No, it is not recommended. However, you can use the cutoff radii from LDA pseudo to generate the new GGA pseudo. Have you tried to compare 2 input files that you have? I guess it is not very good to use LDA pseudo-potential while setting GGA-PBE as x-correlation. Thanks for confirming this. However, I'm not quite sure I understand your suggestion. Let me clarify this: I'm not trying to generate pseudo-potential myself. So, all the input I said are for Molecular Dynamics or Relaxation. Under this situation, I'd say, basically, my input files are the same except for the basis sets. For GGApseudo, I set DZP for Carbon; and for LDApseudo, I set DZ. Thanks for more info. -- You Lin Department of Physics University of South Florida 4202 East Fowler Avenue Tampa, FL 33620 Tel: (813)396-9220 [Office] Homepage: http://shell.cas.usf.edu/~ylin -- Dept. of Physics, Peking University Beijing, P.R.China Phone: 86-10-62753190
Re: [SIESTA-L] Question about LDA and GGA pseudopotential
On Thu, 1 Feb 2007, Oleksandr Voznyy wrote: 1) Why is GGA pseudo-potential gives worse result than LDA? What do you mean worse? A well converged calculation (i.e. converged in E cutoff, k-grid, and converged vs pseudo cutoff radii as well) LDA SHOULD give bonds (or lattice constants) 1% smaller than expt (while binding energies are strongly overestimated) while GGA usually overestimates bonds by about 2% (while binding energies (for molecules) are much closer to expt). First, thanks for the info. Let me explain why I think GGA gives worse result. Let's take CH4 as an example. Here's the fdf input file I have - Start of the file --- SystemName Methane Broyden Optimization SystemLabel general NumberOfAtoms 5 NumberOfSpecies 2 MeshCutoff 50 Ry %block ChemicalSpeciesLabel 1 6 C # Species index, atomic number, species label 2 1 H %endblock ChemicalSpeciesLabel #PAO.BasisSize SZ %blockPAO.BasisSizes C DZP H DZP %endblock PAO.BasisSizes AtomicCoordinatesFormat Ang %block AtomicCoordinatesAndAtomicSpecies 1.0582 0.9353 0.8103 1 1.4145 1.5662 0. 2 1.2065 1.4452 1.7588 2 0. 0.7294 0.6710 2 1.6121 0. 0.8114 2 %endblock AtomicCoordinatesAndAtomicSpecies XC.functional GGA XC.authors PBE PAO.EnergyShift 0.02 Ry DM.Number.Pulay 3 WriteForces T MD.TypeOfRun Broyden MD.NumCGSteps 400 MD.Broyden.History.Steps 6 MD.Broyden.Initial.Inverse.Jacobian 1.0 --- End of the file --- This input file is used along with GGA pseudo potential of Carbon. For LDA pseudo potential case, I just change the line with Carbon at block PAO.BasisSizes to CDZ Basically, this setting reduces the total basis of Carbon atom from 13 to 8. Then I ran siesta to relax the CH4 and measure C-H bond length after relaxation: Exp. LDACarbon GGACarbonCCCBDB calculated PBEPEB/6-331G** 1.094 1.097 1.10 1.0977 It seems LDA pseudo potential did give a shorter bond-length while still longer than experimental value. GGA pseudo potential gives an even longer bond-length, which is obviously, a worse result. 2) Is it safe to use LDA pseudo-potential while setting GGA-PBE as x-correlation in input.fdf? No, it is not recommended. However, you can use the cutoff radii from LDA pseudo to generate the new GGA pseudo. Have you tried to compare 2 input files that you have? I guess it is not very good to use LDA pseudo-potential while setting GGA-PBE as x-correlation. Thanks for confirming this. However, I'm not quite sure I understand your suggestion. Let me clarify this: I'm not trying to generate pseudo-potential myself. So, all the input I said are for Molecular Dynamics or Relaxation. Under this situation, I'd say, basically, my input files are the same except for the basis sets. For GGApseudo, I set DZP for Carbon; and for LDApseudo, I set DZ. Thanks for more info. -- You Lin Department of Physics University of South Florida 4202 East Fowler Avenue Tampa, FL 33620 Tel: (813)396-9220 [Office] Homepage: http://shell.cas.usf.edu/~ylin
Re: [SIESTA-L] Question about LDA and GGA pseudopotential
1) Why is GGA pseudo-potential gives worse result than LDA? What do you mean worse? A well converged calculation (i.e. converged in E cutoff, k-grid, and converged vs pseudo cutoff radii as well) LDA SHOULD give bonds (or lattice constants) 1% smaller than expt (while binding energies are strongly overestimated) while GGA usually overestimates bonds by about 2% (while binding energies (for molecules) are much closer to expt). 2) Is it safe to use LDA pseudo-potential while setting GGA-PBE as x-correlation in input.fdf? No, it is not recommended. However, you can use the cutoff radii from LDA pseudo to generate the new GGA pseudo. Have you tried to compare 2 input files that you have?