[QE-users] cell_dofree with BFGS relaxation

Daniel Marchand daniel.marchand at gmail.com
Wed May 1 22:17:37 CEST 2019


Thanks, I saw the notifcations and will be testing out the patch tomorrow.
Thanks!

On Wed, May 1, 2019 at 5:40 PM Lorenzo Paulatto <paulatz at gmail.com> wrote:

> I think I found a possible solution, I put a little patch in the gitlab
> tracker, I'm not sure if you get notifications.
>
> cheers
>
> On 4/30/19 2:49 PM, Daniel Marchand wrote:
> > Thanks again. So for the BFGS output, for whatever reason there is a
> > final SCF step, I"m not sure why. And yes, correct, the
> > damped_dyanmics(n=50) ~= "BFGS"(n=14).
> >
> > However my concern is that for the "BFGS" the z component of all axis
> > changes, which for my use-case is very bad. However, for damp only the z
> > component of axis 3 (v3_z) is allowed to relax. Unfortunately, in my use
> > case (the one in the attached example) the damp method seems to converge
> > much much slower than BFGS. E.g. for the metrics that would affect my
> needs:
> >
> > grep -A3 CELL_PARAMETERS pw.out.bfgs <fast convergence in cell shape,
> > but we see shifts in the z-component of axis 1 & axis 2, which for my
> > needs is very bad>
> > grep -A3 CELL_PARAMETERS pw.out.damp_pr <slow convergence in cell shape,
> > but we see that the z-component of axis 1 & axis 2 are held fixed, which
> > for my needs is very good>
> >
> > Is there any way to make bfgs hold the z-component of axis 1 & 2 fixed,
> > like how we see for the damp_pr example?
> >
> > Best,
> >
> > Daniel
> >
> > On Tue, Apr 30, 2019 at 2:16 PM Lorenzo Paulatto <paulatz at gmail.com
> > <mailto:paulatz at gmail.com>> wrote:
> >
> >     there are several quircks:
> >
> >     I see you did not specify ion_dynamics="damp" when doing damped
> >     dynamics, I'm not sure what happens of the ions in this case.
> >
> >     The "BFGS" output you sent me is contains 14 steps of BFGS
> >     optimization,
> >     followed by one step of (damped?) molecular dynamics
> >
> >     The damped dynamics run stops because it reaches the maximum of 50
> >     iterations (nstep=50) not because it has converged, but the cell it
> >     obtains is pretty much identical to the BFGS one
> >
> >     hth
> >
> >     On 30/04/2019 13:55, Daniel Marchand wrote:
> >      > Thanks, Lorenzo, for the rapid response. I have a bit of a
> >     concern w.r.t
> >      > the z-only relaxation. When I use BFGS the z component of all the
> >     axis
> >      > are allowed to relax. However, I do not want this, I want only
> the z
> >      > component of axis 3 (v3_z) to be moved, as per the description in
> >     the
> >      > documentation. This only seems possible with the damp algorithm.
> >     Here I
> >      > have attached sample input for both the bfgs and damp_pr
> algorithms
> >      >
> >
> https://drive.google.com/file/d/17Bt8HHL55yG75cba6awvl8E_wo20KDYI/view?usp=sharing
> .
> >
> >      > Perhaps I have one of the settings wrong?
> >      >
> >      > Best,
> >      >
> >      > Daniel
> >      >
> >      >
> >      > On Tue, Apr 30, 2019 at 1:07 PM Lorenzo Paulatto
> >     <paulatz at gmail.com <mailto:paulatz at gmail.com>
> >      > <mailto:paulatz at gmail.com <mailto:paulatz at gmail.com>>> wrote:
> >      >
> >      >      > 1) Is this still the case? When I enable this setting in
> >     my own
> >      >      > calculations (QE v6.3) I see the cell does indeed allow
> only
> >      >      > z-relaxation in the cell, but it does so for all the
> lattice
> >      >     directions
> >      >      > rather than just the a_3 lattice direction.
> >      >
> >      >     In general cell_dofree works with BFGS, but it is true that
> >     some times
> >      >     damped dynamics is more reliable.
> >      >
> >      >     I.e. I recently did some relaxations with
> >     cell_dofree="shape", as BFGS
> >      >     was not findign the minimum for some volumes, I just put
> >      >         cell_dynamics="damp-pr"
> >      >         ion_dynamics = 'damp'
> >      >     with everything default.
> >      >
> >      >     cheers
> >      >
> >      >     --
> >      >     Lorenzo Paulatto - Paris
> >      >     _______________________________________________
> >      >     Quantum Espresso is supported by MaX
> >      >     (www.max-centre.eu/quantum-espresso
> >     <http://www.max-centre.eu/quantum-espresso>
> >      >     <http://www.max-centre.eu/quantum-espresso>)
> >      >     users mailing list users at lists.quantum-espresso.org
> >     <mailto:users at lists.quantum-espresso.org>
> >      >     <mailto:users at lists.quantum-espresso.org
> >     <mailto:users at lists.quantum-espresso.org>>
> >      > https://lists.quantum-espresso.org/mailman/listinfo/users
> >      >
> >      >
> >      > _______________________________________________
> >      > Quantum Espresso is supported by MaX
> >     (www.max-centre.eu/quantum-espresso
> >     <http://www.max-centre.eu/quantum-espresso>)
> >      > users mailing list users at lists.quantum-espresso.org
> >     <mailto:users at lists.quantum-espresso.org>
> >      > https://lists.quantum-espresso.org/mailman/listinfo/users
> >      >
> >
> >     --
> >     Lorenzo Paulatto - Paris
> >     _______________________________________________
> >     Quantum Espresso is supported by MaX
> >     (www.max-centre.eu/quantum-espresso
> >     <http://www.max-centre.eu/quantum-espresso>)
> >     users mailing list users at lists.quantum-espresso.org
> >     <mailto:users at lists.quantum-espresso.org>
> >     https://lists.quantum-espresso.org/mailman/listinfo/users
> >
> >
> > _______________________________________________
> > Quantum Espresso is supported by MaX (www.max-centre.eu/quantum-espresso
> )
> > users mailing list users at lists.quantum-espresso.org
> > https://lists.quantum-espresso.org/mailman/listinfo/users
> >
>
> --
> Lorenzo Paulatto - Paris
> _______________________________________________
> Quantum Espresso is supported by MaX (www.max-centre.eu/quantum-espresso)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20190501/533b5da3/attachment.html>


More information about the users mailing list