error messages in optimisation run.

Total energy, DFT+U, BigDFT,...

Moderator: amadon

Post Reply
umasubbiah
Posts: 4
Joined: Tue Aug 19, 2014 4:17 pm

error messages in optimisation run.

Post by umasubbiah » Sun Aug 24, 2014 11:46 pm

I am trying to optimise the ions for a 108 atom amorphous system. I am using the input file given in test case 81 in the package. For most of the higher pressures, the code runs smoothly.. with energy converged, but with large forces (> tolmxf 5Ha/Bohr). The stress conditions however seem to show isotropic pressure. Are these total energy values and pressure true ground state values..? 1. Can I use the rprim and xred to continue with linear response calculations?

2. When I increase the scaling factor to reduce the pressure, close to ambient conditions, with increased dilatmx and nband and nstep,(with the same ecut as the higher pressures) I get the following error messages.. and the code stopped with 3 out of 10 (ntime=10) iterations for ionic optimisation.

What should I change to do the calculations at the lower pressures?

!WARNING
message: |
Problem in abi_xheev, info= 7
src_file: Subroutine Unknown
src_line: 0
...


--- !WARNING
message: |
abi_xpotrf, info=1
src_file: Subroutine Unknown
src_line: 0

...

Can you kindly guide me on how to proceed?

Sincerely,
Uma

Jordan
Posts: 282
Joined: Tue May 07, 2013 9:47 am

Re: error messages in optimisation run.

Post by Jordan » Tue Aug 26, 2014 6:58 pm

I am not gonna help you much but you should try to restart you relaxation using the last coordinates (xred or xcart) and the last lattice parameters (acell) from the log file. Just replace your values in you input file by those one.
Run the calculation again.
I expect that you should be able to run again a few time iteration and that the code will fail with the same error message.

I have quite the same issue with the error message

Code: Select all

--- !ERROR
message: |
    abi_xpotrf, info=11
src_file: abi_xorthonormalize.f90
src_line: 106
...

and I can restart the calculation every time for a new run of about 5 or 7 time iterations.

What linear algebra do you use ? What compiler ?

Cheers

Jordan

umasubbiah
Posts: 4
Joined: Tue Aug 19, 2014 4:17 pm

Re: error messages in optimisation run.

Post by umasubbiah » Fri Aug 29, 2014 4:52 pm

Thank you Jordan..
I did not compile the current version of ABINIT. I am using a cluster where the administrators had already done it. I shall try a few things and post my observations on this thread.
Uma

amir
Posts: 19
Joined: Fri Oct 24, 2014 8:44 pm

Re: error messages in optimisation run.

Post by amir » Sun Jan 10, 2016 9:28 pm

I have the same problem when I use toldff or tolrff as stopping criterion, but when I use toldfe it works fine. I also realized that if I use toldff in optimizing a unit cell it does not converge, but when I use toldfe it works fine. However, in the description of toldff it says "This is to be used when trying to equilibrate a structure to its lowest energy configuration (ionmov=2), or in case of molecular dynamics (ionmov=1) ". In the tutorial input files, when optimizing the cell, toldfe is used. Can you explain how I should choose my stopping criteria? Thanks.

amir
Posts: 19
Joined: Fri Oct 24, 2014 8:44 pm

Re: error messages in optimisation run.

Post by amir » Wed Jan 13, 2016 11:31 pm

I have the answer to a part of my question, why toldff can't be used in optimizing the unit cell. Under the description optcell it says: if all the reduced coordinates of atoms are fixed by symmetry, one cannot use toldff to stop the SCF cycle. (Suggestion : use toldfe with a small value, like 1.0d-10)

Jordan
Posts: 282
Joined: Tue May 07, 2013 9:47 am

Re: error messages in optimisation run.

Post by Jordan » Tue Jan 19, 2016 2:16 pm

It hard to answer the question but as a starting reflexion, for general purpose converge the density, then if you are interested in the energy only, consider using toldfe.
In the case of structural optimization, you just want the forces and streses to be as minimal as possible so use toldff or tolrdf to converge the forces/stress.

Usually converging the density is stricter than converging on the energy.
If you keep the same value for toldfe and toldff, then in the first case you converge the energy to let say 1e-10Ha. In the later you converge the forces to 1e-10Ha/Bohr which is really really small for the forces. Then the SCF cycles needs many more iterations to converge and lobpcg becomes unstable.

Post Reply