[Pw_forum] vc-relax
Stefano de Gironcoli
degironc at sissa.it
Tue Nov 25 18:48:36 CET 2008
there are two main algorithms in variable cell shape relaxation: a
damped dynamics and a bfgs-like relaxation.
in the damped dynamics case there are two flavors, Parrinello-Rahman and
Wentzcovitch, and the value of the cell mass is different in the two
cases because the underlying Lagrangian is different. There is a default
value that should be not too wrong. A good value should be something
that gives similar frequencies for the internal vibrations and for the
cell oscillationos so that they are optimally coupled.
In the (recently introduced) bfgs algorithm there is no need of
specifying the cell mass.
Stefano de Gironcoli - SISSA and DEMOCRITOS
Timothy Mason wrote:
> I have two perhaps fundamental questions related to vc-relax.
>
> 1) Can someone give me an idea of what role wmass plays in VC relax?
> it does seem to affect my convergence but it would be nice to have a
> better understanding so I'm not just shooting in the dark. How does
> the algorithm use this parameter?
>
> 2) this question is perhaps more fundamental. Even though the toal
> forces and stresses eventually oscillate about a point, the energy
> itself ALWAYS decreases although asymptotically. is this an
> indication that I actually am converging but it will take a long time?
>
> Thanks in advance for your help.
>
> Tim Mason (university of Missouri St. Louis)
>
> --- On *Fri, 11/14/08, Eric Abel /<etabel at hotmail.com>/* wrote:
>
> From: Eric Abel <etabel at hotmail.com>
> Subject: RE: vc-relax PWSCF
> To: thmason24 at yahoo.com
> Date: Friday, November 14, 2008, 11:00 PM
>
> Not sure what your end goal is, but in general vc-relax shouldn't
> be necessary. Rather, the lattice parameters, as well as the
> atomic positions can be very accurately determined using x-ray or
> neutron diffraction. These should therefore be the starting point
> for your study. From there, I would then relax the structure
> using the simple 'relax' routine, which keeps the lattice
> parameters fixed, but varies the atomic positions within the unit
> cell. This ensures that all of the atoms are in their potential
> minimum. If the relaxed structure is drastically different from
> the experimental structure, this indicates something wrong with
> the pseudopotentials or other input parameters mentioned below.
> This step isn't so necessary if all you are investigating is
> electronic properties, but it is absolutely essentialy for
> anything to do with lattice dynamics.
>
> Anyway, it sounds like you have some foundation calculation to do
> (it took me a good two weeks non-stop to optimize my input
> parameters). I am happy to continue helping you, but you should
> also utilize the pw_forum. By now there is a huge knowledgebase
> of Q & A on all topics. This is where I learned most of what I
> know. My discussions with Marzari and Co. only fine tuned some of
> the details.
>
> Good Luck, and feel free to ask me for any additional guidance.
>
> Eric
>
>
> ------------------------------------------------------------------------
>
> Date: Fri, 14 Nov 2008 06:38:53 -0800
> From: thmason24 at yahoo.com
> Subject: RE: vc-relax PWSCF
> To: etabel at hotmail.com
>
>
> Dear Eric,
>
> the reason I want to use VC relax is that our first guess also
> includes a first guess of the lattice vectors and I have no reason
> to think that they are optimized. Am I missing the point of
> VC-relax?
>
> 1) Good info on the Kpoint density. Hopefully 8 8 8 will give me
> the convergence I need.
> sounds like these computations willl be more expensive than I
> anticipated.
>
> I'm quite new to this so I'm not sure what's all involved in
> thouroughly research the space. I make sure it converges on ecut
> and Kpoint for an scf run. I believe its not metallic so I leave
> smearing alone.
>
> 2) I thought 'damp' was suppose to do better at the final
> convergence so i chose that over 'sd'. I believe I did try a run
> with 'bfgs' and didn't see a difference.
>
>
> i really appreciate you taking the time to look.
>
>
>
>
> --- On *Thu, 11/13/08, Eric Abel /<etabel at hotmail.com>/* wrote:
>
> From: Eric Abel <etabel at hotmail.com>
> Subject: RE: vc-relax PWSCF
> To: thmason24 at yahoo.com
> Date: Thursday, November 13, 2008, 10:01 PM
>
> Well, at first glance I don't see anything grossly wrong.
> The first question I feel compelled to ask, is why you wish to
> do a vc-relax? As you saw, I asked a lot of questions on the
> issue, and (being a student at MIT) had the opportunity to sit
> down with Dr. Marzari, and some of his grad students who were
> actively developing PWSCF code, and I found that sometimes we
> make life too dificult for ourselves by performing unnecessary
> calculations.
>
>
>
>
>
> Anyway, assuming vc-relax is indeed something you care about,
> I do have a few things stick out however:
>
> 1. Your k-point density seems a little coarse. I wouldn't
> run at anything less than 8 8 8, especially for vc-relax. You
> may even need a grid as tight as 16 16 16, or even 24 24 24.
> Have you thoroughly investigated this space? The relax will
> be more sensitive to this than just the band calculations.
>
> 2. I was running with 'bfgs' ion dynamics. I believe this is
> the default. Do you have a specific reason for running with
> 'damp'?
>
> 3. You should also play with the mixing beta, smearing
> parameters, and ecutwfc/ecutrho. Dr. Marzari showed me nice
> plots of the convergence of the total force as a function of
> all of these parameters, which really illustrated to me the
> importance of finding asymtote, as the results can be
> drastically affected.
>
>
>
> ------------------------------------------------------------------------
>
> Date: Thu, 13 Nov 2008 06:22:53 -0800
> From: thmason24 at yahoo.com
> Subject: RE: vc-relax PWSCF
> To: etabel at hotmail.com
>
>
> Great.
>
> my struggle is that I can't get my structure to converge as
> tightly as I'd like it to. I'd like to get the forces down
> to 0.0004 Ryd/a.u but they generally start oscillating around
> 0.02. Quite confusing to me since the total energy continues
> to decline although asymptotically. I'm including my input
> file as a reference. Did you run into this and remember how
> you solved it?
>
> thanks so much for taking a look.
>
> &CONTROL
> calculation = "vc-relax",
> prefix = 'n14.out',
> tprnfor = .TRUE.
> pseudo_dir = '/home/thmmqc/pwscf/psps/US_GGA',
> outdir='/home/thmmqc/tmp'
> forc_conv_thr = 0.0004,
> nstep = 100
> /
>
> &SYSTEM
> ibrav = 0, nat = 8, ntyp = 3,
> ecutwfc = 40
> ecutrho = 480
> celldm(1) = 1.889725989
> /
>
> &ELECTRONS
> mixing_mode = 'plain',
> mixing_beta = 0.7,
> conv_thr = 1.0D-8,
> /
>
> &IONS
> ion_dynamics = 'damp',
> ion_damping = 0.2,
> ion_positions = 'from_input',
> upscale = 100,
> /
>
> &CELL
> cell_dynamics = "damp-pr",
> /
>
> ATOMIC_SPECIES
> Li 6.941000 Li.pbe-n-van.UPF
> N 14.006740 N.pbe-rrkjus.UPF
> H 1.007940 H.pbe-rrkjus.UPF
>
> CELL_PARAMETERS
> 5.67649703 0.09921379 0.92493840
> 1.07129347 6.96479879 -0.12086294
> -1.34503851 0.22123471 3.75667723
>
> ATOMIC_POSITIONS {crystal}
> Li 0.116794697000 0.651673872000 0.623162333000
> Li 0.175318460000 0.437682929000 0.133625687000
> N 0.091275887000 0.737412816000 0.118429764000
> N 0.198322191000 0.351690647000 0.637630892000
> H -0.076285944000 0.821236835000 0.075907461000
> H 0.202774564000 0.839292956000 0.150103014000
> H 0.081850214000 0.253355665000 0.605178356000
> H 0.363204688000 0.263855155000 0.679010850000
>
> K_POINTS automatic
> 4 4 4 0 0 0
>
>
>
>
>
>
>
>
>
>
> --- On *Thu, 11/13/08, Eric Abel /<etabel at hotmail.com>/* wrote:
>
> From: Eric Abel <etabel at hotmail.com>
> Subject: RE: vc-relax PWSCF
> To: thmason24 at yahoo.com
> Date: Thursday, November 13, 2008, 2:29 AM
>
> "done quite a bit" might be an overstatement. I got it to
> work, but that was over a year ago. You can shoot some
> questions my way, and I'll do my best to try to help.
>
> Eric
>
>
> ------------------------------------------------------------------------
>
> Date: Tue, 11 Nov 2008 12:46:53 -0800
> From: thmason24 at yahoo.com
> Subject: vc-relax PWSCF
> To: etabel at hotmail.com
>
>
> Hi Eric,
>
> My name is Tim Mason. I'm a grad student at University
> of Missouri St. Louis.
>
> I saw your name on the PWSCF forums. You seem to have
> done quite a bit with VC-relax. I'm not sure how
> experienced you ended up getting but would you be willing
> to answer a few basic questions?
>
> thanks,
>
> Tim
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>
More information about the users
mailing list