Top Page > Browsing

Question about Geometry Optimization Date: 2015/08/22 18:31 Name: Eike F. Schwier   <schwier@hiroshima-u.ac.jp> Dear "MD.type RF" experts, I have come across a rather strange behaviour during one of my bulk unit-cell optimisations and would like to ask, if anybody can explain to my why this is occurring and if there are any parameters I may change / set / optimize to prevent this from happening. In my relaxation of the FRAC coordinates (RF) of TlBiSe2, I found that there was a lateral force on the Se atoms (not on Tl or Bi). If Se positions are allowed to relax laterally (MD.Fixed.XYZ 0 0 0) all Se move ~0.1 A away from their original positions until the convergence criterion is reached. This is due to a force Fx,Fy~0.008 Ha/bohr. If I the restart the optimisation from the "relaxed" coordinates, the Se atoms move back to (almost) their original positions before the first lateral relaxation. This seemed puzzling as I do not expect any more relaxation, if I restart the RF from its already relaxed coordinates. Btw: If I choose EF instead of RF I get the same behaviour. Se positions in Ang (original, first relax, second relax) 1,x 1,y 2,x 2,y 3,x 3,y 0 0 0 0 2.12 1.22398 0.10537 0.04579 -0.10537 -0.04578 2.22532 1.26981 0.00031 0.00096 -0.00031 -0.00096 2.12024 1.22495 4,x 4,y 5,x 5,y 6,x 6,y 2.12 1.22398 0 2.44797 0 2.44797 2.01463 1.17812 0.10537 2.49384 -0.10533 2.40213 2.11978 1.22314 0.00022 2.4488 -0.00024 2.447 I tried to change the basisset, the cutoff and the k-mesh, but the initial force on the Se atoms remains nonzero regardless of my choice of parameters. However, if I displace the first Se atom (dX >= 0.+0.00001001 FRAC) the lateral forces Fx and Fy become zero (Fx,Fy~0.00005 Ha/bohr) If I choose a dX <= 0.00001 forces are non-zero again. My suspicion is, that I break some internal symmetry in the calculation with the Se displacement larger than 0.00001 FRAC. But I do not understand why a broken symmetry should change the forces to such an extend. Furthermore, if the groundstate is indeed one with the Se atoms displaced and the symmetry broken, then why do the Se atoms move back to the symmetry positions if I restart the calculation? Is the initial symmetry conserved in the RF (EF) run, even if the fractional coordinates are changed to break it? best regards, Eike # # Definition of Atomic Species # Species.Number 3 <Definition.of.Atomic.Species Se Se7.0-s3p2d1 Se_PBE13 Bi Bi8.0-s3p2d2f1 Bi_PBE13 Tl Tl8.0-s3p3d2f1 Tl_PBE13 Definition.of.Atomic.Species> Atoms.Number 12 Atoms.SpeciesAndCoordinates.Unit FRAC # Ang|AU <Atoms.SpeciesAndCoordinates 1 Se +0.00001500 -0.00000000 +0.24202732 3 3 2 Se -0.00000000 -0.00000000 +0.75797044 3 3 3 Se +0.66666666 +0.33333333 +0.57535558 3 3 4 Se +0.66666666 +0.33333333 +0.09131124 3 3 5 Se +0.33333333 +0.66666666 +0.90868607 3 3 6 Se +0.33333333 +0.66666666 +0.42464308 3 3 7 Bi -0.00000000 -0.00000000 +0.50000000 7.5 7.5 8 Bi +0.66666666 +0.33333333 +0.83333333 7.5 7.5 9 Bi +0.33333333 +0.66666666 +0.16666666 7.5 7.5 10 Tl -0.00000000 -0.00000000 -0.00000000 9.5 9.5 11 Tl +0.66666666 +0.33333333 +0.33333333 9.5 9.5 12 Tl +0.33333333 +0.66666666 +0.66666666 9.5 9.5 Atoms.SpeciesAndCoordinates> Atoms.UnitVectors.Unit Ang # Ang|AU <Atoms.UnitVectors 4.24 0 0 -2.12 3.67194771204602 0 0 0 22.33000 Atoms.UnitVectors> <MD.Fixed.XYZ 1 1 1 0 2 1 1 0 3 1 1 0 4 1 1 0 5 1 1 0 6 1 1 0 7 1 1 1 8 1 1 1 9 1 1 1 10 1 1 1 11 1 1 1 12 1 1 1 MD.Fixed.XYZ> # # SCF or Electronic System # scf.XcType GGA-PBE # LDA|LSDA-CA|LSDA-PW|GGA-PBE scf.ElectronicTemperature 300.0 # default=300 (K) scf.energycutoff 300.0 # default=150 (Ry) scf.maxIter 160 # default=40 scf.EigenvalueSolver band # Recursion|Cluster|Band scf.Kgrid 5 5 3 # means n1 x n2 x n3 scf.Generation.Kpoint MP # regular|MP scf.Mixing.Type rmm-diisk # Simple|Rmm-Diis|Gr-Pulay scf.Init.Mixing.Weight 0.3 # default=0.30 scf.Min.Mixing.Weight 0.001 # default=0.001 scf.Max.Mixing.Weight 0.600 # default=0.40 scf.Mixing.History 20 # default=5 scf.Mixing.StartPulay 10 # default=6 scf.criterion 1e-9 # default=1.0e-6 (Hartree) scf.restart off # # MD or Geometry Optimization # MD.Type RF # MD.maxIter 1 # default=1 MD.Opt.criterion 1.0e-5 scf.fixed.grid 0.00000000000000 0.00000000000000 0.00000000000000

Page: [1]

Re: Question about Geometry Optimization ( No.1 ) Date: 2016/04/11 12:08 Name: T. Ozaki Hi, Sorry for very late response. I did the calculation with Ver. 3.8 using the input file provided by you, and got the following output. <Steepest_Descent> SD_scaling= 1.259801151395 <Steepest_Descent> |Maximum force| (Hartree/Bohr) = 0.000007788889 <Steepest_Descent> Criterion (Hartree/Bohr) = 0.000010000000 And I did the second calculation using a dat# file, and got the same result. Could you provide your input files for the first and second calculations, if this is still an important problem for you? Regards, TO Re: Question about Geometry Optimization ( No.2 ) Date: 2016/04/14 14:15 Name: Eike F. Schwier  <schwier@hiroshima-u.ac.jp> Dear Ozaki-sensei, thank you very much for the reply. I will test if I can reproduce the problem with openMX-3.8 and post the input files if the lateral forces remain present. best regards, Eike Re: Question about Geometry Optimization ( No.3 ) Date: 2016/05/12 11:38 Name: Eike F. Schwier  <schwier@hiroshima-u.ac.jp> Dear Ozaki-sensei, I checked the RF relaxation with openmx-3.8 and I can reproduce my initial problem. Below I have attached my input file for the relaxation in the first RF run (starting from the experimental fractional coordinates). I also attached the resulting coordinates from the first and second (restarted) RF run. As you can see the Se positions deviate by up to 0.1 A in x,y from the experimental / high symetry coordination at the end of the first RF run. After restarting with the newly found positions the Se deviation is reduced to 0.0001 A after reaching the convergence criterium. The deviation of the Se, I suppose would come closer and closer to zero, if I were to increase the calculation precision. Surprisingly, the total energy of the first run is lower by 7mHa then the total energy of the second run and each SCF-loop took longer in the second run, compared to the initial RF run. Which makes my think, the symmetry of the system may be different between the first run and the second run (change of the k sampling points weighting in the Monckhorst- Pack Mesh should lead to change in total energy). Does the system remember the symmetry of the first RF loop and keep using it for the full RF run even if symetries are broken? In order increase the speed of both RF runs I started with a "nomd" SCF with scf.Mixing.History 35 scf.Mixing.StartPulay 25 and then restarted with the reduced parameters 20/10 and RF. best regards, Eike PS: I also noticed that the Bi and Tl coordinated even though they were fixed changed slightly in the last two to three digits. # # Input file number 1 # Species.Number 3 <Definition.of.Atomic.Species Se Se7.0-s3p2d1 Se_PBE13 Bi Bi8.0-s3p2d2f1 Bi_PBE13 Tl Tl8.0-s3p3d2f1 Tl_PBE13 Definition.of.Atomic.Species> Atoms.Number 12 Atoms.SpeciesAndCoordinates.Unit FRAC # Ang|AU <Atoms.SpeciesAndCoordinates 1 Se +0.00000000 -0.00000000 +0.24202732 3 3 2 Se -0.00000000 -0.00000000 +0.75797044 3 3 3 Se +0.66666666 +0.33333333 +0.57535558 3 3 4 Se +0.66666666 +0.33333333 +0.09131124 3 3 5 Se +0.33333333 +0.66666666 +0.90868607 3 3 6 Se +0.33333333 +0.66666666 +0.42464308 3 3 7 Bi -0.00000000 -0.00000000 +0.50000000 7.5 7.5 8 Bi +0.66666666 +0.33333333 +0.83333333 7.5 7.5 9 Bi +0.33333333 +0.66666666 +0.16666666 7.5 7.5 10 Tl -0.00000000 -0.00000000 -0.00000000 9.5 9.5 11 Tl +0.66666666 +0.33333333 +0.33333333 9.5 9.5 12 Tl +0.33333333 +0.66666666 +0.66666666 9.5 9.5 Atoms.SpeciesAndCoordinates> Atoms.UnitVectors.Unit Ang <Atoms.UnitVectors 4.24 0 0 -2.12 3.67194771204602 0 0 0 22.33000 Atoms.UnitVectors> <MD.Fixed.XYZ 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 7 1 1 1 8 1 1 1 9 1 1 1 10 1 1 1 11 1 1 1 12 1 1 1 MD.Fixed.XYZ> scf.XcType GGA-PBE scf.ElectronicTemperature 300 scf.energycutoff 300 scf.maxIter 160 scf.EigenvalueSolver band scf.Kgrid 5 5 3 scf.Generation.Kpoint MP scf.Mixing.Type rmm-diisk scf.Init.Mixing.Weight 0.3 scf.Min.Mixing.Weight 0.001 scf.Max.Mixing.Weight 0.600 scf.Mixing.History 20 # 35 scf.Mixing.StartPulay 10 # 25 scf.criterion 1e-9 scf.restart on MD.Type RF MD.maxIter 200 MD.Opt.criterion 1.0e-5 scf.fixed.grid 0.00000000000000 0.00000000000000 0.00000000000000 # # coordinates after initial RF run # time= 5.000 (fs) Energy= -699.36784 (Hartree) 1 Se 0.1048307 0.0453472 5.3913548 3.00000 3.00000 2 Se -0.1048357 -0.0453437 16.9386738 3.00000 3.00000 3 Se 2.2248200 1.2694267 12.8346324 3.00000 3.00000 4 Se 2.0150856 1.1784504 2.0519801 3.00000 3.00000 5 Se 0.1049096 2.4935010 20.2780510 3.00000 3.00000 6 Se -0.1048250 2.4025237 9.4953872 3.00000 3.00000 7 Bi -0.0000006 0.0000011 11.1650152 7.50000 7.50000 8 Bi 2.1200240 1.2240375 18.6083878 7.50000 7.50000 9 Bi -0.0000247 2.4479132 3.7216545 7.50000 7.50000 10 Tl -0.0000021 0.0000004 0.0000220 9.50000 9.50000 11 Tl 2.1200000 1.2239610 7.4433323 9.50000 9.50000 12 Tl -0.0000037 2.4479875 14.8867019 9.50000 9.50000 # # coordinates after RF run with relaxed coordinates from first run # time= 4.500 (fs) Energy= -699.36153 (Hartree) 1 Se -0.0001864 0.0000985 5.4046056 3.00000 3.00000 2 Se 0.0001834 -0.0000984 16.9254391 3.00000 3.00000 3 Se 2.1197940 1.2240674 12.8476880 3.00000 3.00000 4 Se 2.1202008 1.2238682 2.0390057 3.00000 3.00000 5 Se -0.0002032 2.4480808 20.2910410 3.00000 3.00000 6 Se 0.0002036 2.4478818 9.4823477 3.00000 3.00000 7 Bi -0.0000002 0.0000005 11.1650194 7.50000 7.50000 8 Bi 2.1200029 1.2239923 18.6083366 7.50000 7.50000 9 Bi -0.0000030 2.4479564 3.7217140 7.50000 7.50000 10 Tl -0.0000013 0.0000000 0.0000260 9.50000 9.50000 11 Tl 2.1200073 1.2239873 7.4433848 9.50000 9.50000 12 Tl -0.0000094 2.4479605 14.8866577 9.50000 9.50000 Re: Question about Geometry Optimization ( No.4 ) Date: 2016/05/15 11:38 Name: T. Ozaki Hi, I reproduced your problem, and found that this comes from the use of MP mesh. When I commented out the line as: #scf.Generation.Kpoint MP with the first trial, I obtained the expected result as follows: 1 Se -0.00019 0.00005 5.40454 -0.000001206769 0.000000345288 0.000001203253 2 Se 0.00019 -0.00005 16.92544 0.000001156990 -0.000000363958 0.000000503216 3 Se 2.11982 1.22403 12.84770 0.000000381678 -0.000000087042 0.000000563203 4 Se 2.12018 1.22393 2.03896 -0.000000604929 0.000000175567 0.000001035809 5 Se -0.00018 2.44801 20.29103 0.000000669838 -0.000000118779 -0.000000005786 6 Se 0.00018 2.44792 9.48229 -0.000000371595 0.000000091127 0.000000299348 7 Bi -0.00000 -0.00000 11.16501 -0.000000010938 0.000000000111 -0.000001854282 8 Bi 2.12000 1.22398 18.60834 -0.000001150646 0.000000667656 -0.000019821237 9 Bi -0.00000 2.44797 3.72168 0.000001147790 -0.000000637480 0.000013407077 10 Tl -0.00000 -0.00000 0.00001 0.000000000279 0.000000022954 -0.000003008282 11 Tl 2.12000 1.22398 7.44335 -0.000000589635 0.000000390216 -0.000021168343 12 Tl 0.00000 2.44796 14.88667 0.000000647392 -0.000000366004 0.000017526501 So, could you use the regular mesh, until the bug is fixed? Thank you for your cooperation in advance. Regards, TO Re: Question about Geometry Optimization ( No.5 ) Date: 2016/05/16 19:35 Name: Eike F. Schwier  <schwier@hiroshima-u.ac.jp> Dear Ozaki-sensei, thank you very much for your kind help. Commenting out the MP mesh option indeed reduces the forces (x,y as well as z) to almost zero. In i further increase the k-mesh the lateral forces get towards zero as expected for k-convergence. best regards, Eike

Page: [1]