A fluctuation in max. forces is normal. Attached is a max. force graph of a few CG geometry optimizations I did of isolated molecules.
Further details are in the following PDF: (keyword EggBox Effect) http://www.home.uni-osnabrueck.de/apostnik/Lectures/SIESTA-tuto.pdf So the lower the max. force is, the less CG steps are required in my experience and only then can you achieve low MD.MaxForceTol values. To reduce the max. forces you have to increase the finess of the real space grid, by either increasing the MeshCutOff value or making use of GridCellSampling: http://departments.icmab.es/leem/siesta/Documentation/Manuals/siesta-3.1-manual/node47.html Best wishes, Frank > I'm doing geometry optimisation if I correctly set it (as I understand > it's > by default) > > Best Wishes, Maxim. > > > 2013/11/22 Herbert Fruchtl <[email protected]> > >> Just to clarify: You are doing MD, not a geometry optimisation, and you >> are surprised that the forces don't go to zero? What are you trying to >> find? For a minimum energy structure, you need a geometry optimisation. >> In >> an MD simulation the total (potential+kinetic) energy should remain >> constant (depending on your ensemble...), but the forces will vary. >> >> Herbert >> >> >> On 22/11/2013 08:58, ÐакÑим ÐÑÑенÑÑев wrote: >> >>> Dear Sebastian! >>> >>> My calculations convverge at each MD cycle (about 300-400 SCF cycles), >>> but max force oscillate (about 0.1-0.04 eV/anstrom) and I can not >>> finish >>> MD, and am already at 63 cycle and the calculation cycle is still >>> going, >>> although the system is quite large and contains 47 atoms. My counter >>> question to all: does this oscillation of forces normal? >>> I would like to strict DM.Tolerance to 0.00001 (it is VERY important to >>> accurate calculation of forces) and to limit MD.MaxCGDispl . >>> >>> Best Wishes, >>> Maxim >>> >>> >>> 2013/11/21 Sebastian Caicedo Davila >>> <[email protected] >>> <mailto:[email protected]>> >>> >>> >>> Hello Diana!! >>> >>> Thanks a lot for your suggestion! I'll give it a try and maybe I'll >>> be bothering you again, if it is not an imposition, of course. >>> >>> Gracias de nuevo, y saludos desde tu tierra! >>> >>> Sebastian Caicedo Davila >>> Bionanoelectronics Research Group >>> School of electrical and electronics engineering >>> Universidad del Valle >>> Cali - Colombia. >>> >>> >>> >>> On Nov 21, 2013, at 11:02 AM, <[email protected] >>> <mailto:[email protected]>> <[email protected] >>> <mailto:[email protected]>> wrote: >>> >>> Hello Sebastian, >>>> >>>> I have had similar problems with convergence of the scf cycle. >>>> They were solved when I changed from Pulay mixing to Broyden >>>> mixing for the charge density. >>>> >>>> Usually if something doesn't converge in 250 cycles, it probably >>>> never will. Also you can play with the DM.Tolerance. For very >>>> difficult systems I set it to 0.0008 and then restart from that DM >>>> file with a stringer convergence criterion. >>>> >>>> Here is the settings I am using for my calculations: >>>> >>>> #--------------------Convergence of SCF--------------------------- >>>> ---- >>>> MeshCutoff 200.0 Ry # Default value PW cutoff >>>> for grid >>>> >>>> >>>> MaxSCFIterations 250 # Maximum number of SCF >>>> iter >>>> DM.NumberBroyden 3 >>>> DM.MixingWeight 0.01 # New DM amount for next >>>> SCF cycle >>>> DM.Tolerance 0.0001 # Tolerance in maximum >>>> difference >>>> DM.UseSaveDM .false. # to use continuation >>>> files >>>> >>>> DM.NumberKick 100 >>>> DM.KickMixingWeight 0.01 >>>> #---------------------------------- >>>> >>>> Suerte! >>>> >>>> PS: yo tambien soy caleña! >>>> >>>> Diana M. Otálvaro >>>> PhD Candidate >>>> >>>> Computational Material Science >>>> MESA+ Institute of Nanotechnology >>>> University of Twente. >>>> Enschede, Nederland >>>> ------------------------------------------------------------ >>>> ------------ >>>> *From:*[email protected] <mailto:[email protected]> >>>> [[email protected] <mailto:[email protected]>] on >>>> >>>> behalf of ÐакÑим ÐÑÑенÑÑев [[email protected] >>>> <mailto:[email protected]>] >>>> *Sent:*Wednesday, November 20, 2013 7:29 AM >>>> *To:*[email protected] <mailto:[email protected]> >>>> *Subject:*Re: [SIESTA-L] Convergence problem on heterostructure. >>>> >>>> >>>> I forgot to tell you - besides DM.MixingWeight 0.05 I also set >>>> PAO.EnergyShift to 65 meV as >>>> herehttp://www.nature.com/nnano/journal/v7/n9/extref/ >>>> nnano.2012.127-s2.txt . >>>> >>>> >>>> Best wishes, >>>> Dr. Maxim Arsent'ev >>>> Laboratory of inorganic synthesis >>>> Institute of Silicate Chemistry of RAS >>>> >>>> >>>> 2013/11/18 Mahmoud Hammouri<[email protected] <mailto:[email protected]>> >>>> >>>> >>>> You could also try to increase the non-equilibrium contour >>>> points. >>>> >>>> Mahmoud >>>> >>>> >>>> On Nov 18, 2013, at 11:22, "Sebastián Cauce do Dávila" >>>> <[email protected] >>>> <mailto:[email protected]>> wrote: >>>> >>>> Thank you very much for your suggestion. I will give it a >>>> try! >>>>> >>>>> Sebastián Caicedo Dávila >>>>> Bionanoelectronics research group >>>>> School of electrical and electronics engineering >>>>> Universidad del Valle >>>>> Cali - Colombia >>>>> >>>>> >>>>> On 18/11/2013, at 11:50, ÐакÑим ÐÑÑенÑÑев >>>>> <[email protected] <mailto:[email protected]>> wrote: >>>>> >>>>> I also try some siesta only six months, but for a more >>>>>> heterogeneous structure I noticed that the reduction >>>>>> DM.MixingWeight reduces oscillations in convergence, so try >>>>>> to play with it (decrease DM.MixingWeight). See >>>>>> http://www.nature.com/nnano/journal/v7/n9/extref/nnano. >>>>>> 2012.127-s2.txt >>>>>> Maybe someone will tell something else. >>>>>> >>>>>> >>>>>> 2013/11/18 Sebastian Caicedo >>>>>> D.<[email protected] >>>>>> <mailto:[email protected]>> >>>>>> >>>>>> >>>>>> Dear Siesta/Transiesta users, >>>>>> >>>>>> I've been using the software for a couple months and >>>>>> made some calculations of InAs and InP unit cell quite >>>>>> successfully. Now I intend to work on transiesta >>>>>> calculations of a heterostructure of both these >>>>>> materials (InAs bulk with InP barriers). As a first >>>>>> approach I want to simulate a couple of InAs unit cells >>>>>> and the barriers shall be unit cells of InP, just to get >>>>>> a qualitative idea. The electrodes .TSHS files were >>>>>> obtained from the previous calculations of the InAs unit >>>>>> cell. I wanted first to run a siesta calculation of the >>>>>> whole heterostructure, so I can take a look at the >>>>>> eigenvalues and define a correct value for the parameter >>>>>> TS.ComplexContour.Emin, as suggested in the manual. >>>>>> >>>>>> The structure has 120 atoms and I'm using LDA >>>>>> Pseudopotentials, obtained from the siesta webpage. (In >>>>>> Pseudo includes semicore electrons, so we have enough >>>>>> electrons to build a zinc-blende structure). The issue >>>>>> is that after 1000 SCF steps, the energy doesn't >>>>>> converge at all. I'm doing a Gamma-point calculation, >>>>>> but even when I make the k-grid 1x1x15 it doesn't work >>>>>> either. >>>>>> >>>>>> I used the same MeshCutoff as for the individual >>>>>> materials unit cell an 200Ry and I increased the Energy >>>>>> Shift, so the calculation would be a bit faster though >>>>>> less precise. >>>>>> >>>>>> I attach the Input (INPUT.fdf, STRUCTURE.fdf) and output >>>>>> files. >>>>>> >>>>>> I would really appreciate any suggestions. >>>>>> >>>>>> sincerely >>>>>> >>>>>> Sebastian Caicedo Davila >>>>>> Bionanoelectronics Group >>>>>> School of Electrical and Electronics Engineering >>>>>> Universidad del Valle >>>>>> Cali-Colombia >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> -- >>>>>> Best wishes, >>>>>> Dr. Maxim Arsent'ev >>>>>> Laboratory of inorganic synthesis >>>>>> Institute of Silicate Chemistry of RAS >>>>>> >>>>> >>>> >>>> >>>> >>>> -- >>>> Best wishes, >>>> Dr. Maxim Arsent'ev >>>> Laboratory of inorganic synthesis >>>> Institute of Silicate Chemistry of RAS >>>> >>> >>> >>> >>> >>> -- >>> Best wishes, >>> Dr. Maxim Arsent'ev >>> Laboratory of inorganic synthesis >>> Institute of Silicate Chemistry of RAS >>> >> >> >> -- >> Herbert Fruchtl >> Senior Scientific Computing Officer >> School of Chemistry, School of Mathematics and Statistics >> University of St Andrews >> -- >> The University of St Andrews is a charity registered in Scotland: >> No SC013532 >> > > > > -- > Best wishes, > Dr. Maxim Arsent'ev > Laboratory of inorganic synthesis > Institute of Silicate Chemistry of RAS >
TotalForces.pdf
Description: Adobe PDF document
