[CP2K-user] Convergence problems with DFT+U in NPT ensemble

[Ivan Scivetti]

Dear CP2K community,

I am computing a transition metal oxide for which I am applying DFT+U 
corrections in a supercell composed of 384 atoms.
To analyse structural properties at room temperature (295K) and standard 
pressure (1 atm) I need to carry out MD simulations. Due to the complexity 
of the system (of hexagonal symmetry), I start with the *NVT* ensemble and 
eventually reach equilibration. Electronic convergence runs smoothly and 
efficiently with the OT scheme. Since I do not know the range of the 
supercell volumes that corresponds to a pressure of 1atm, I started from 
the experimental values but, as expected, I get average pressures of the 
order of 500 to 1000 bars depending on the composition of the system. I 
could keep trying with different volumes and approach to the desire 1 atm 
pressure. This task, however, is very demanding and rather cumbersome.
I then decided to consider the *NPT_F* ensemble. However, the electronic 
DFT+U energy starts increasing significantly during the first steps of the 
simulation and, eventually, convergence problems start to manifest due to 
the Mulliken population analysis, with a negative contribution to the DFT+U 
energy. I could also try the LOWDIN method for charges but forces are not 
available for this scheme. Thus, it seems that my settings for the NPT_F 
ensemble are not appropriate for this DFT+U simulation. I describe the 
relevant MD settings of the input file below:

*&Motion*
*  &MD*
*    Timestep 0.5*
*    Steps 2000*
*    Ensemble NPT_F*
*    Temperature 300.0*
*    &Thermostat*
*       REGION massive*
*       TYPE CSVR*
*       &CSVR*
*         TIMECON [fs] 10.*
*       &END CSVR*
*    &End Thermostat*

*   &Barostat*
*     Pressure 1.0*

*     TIMECON [fs] 10.*

*   &End barostat*

*&End Motion*

 The message I obtain is the following
* *** WARNING in dft_plus_u.F:1457 :: DFT+U energy contibution is negative 
****
* *** possibly due to unphysical Mulliken charges. Check your input, if    
****
* *** this warning persists or try a different method!                    
 ****

Again, this problems does not arise with NVT simlation, only with NPT. I am 
wondering if I need of an extra flag in the motion/thermostat session to 
"tame" the MD run and the electronic convergence. I have carried out an 
extensive research in the web but it seems there is no previous record of 
such a type of simulation with CP2K. 

Any help would be appreciated.
Thank you very much in advance.

Ivan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190218/07cd87ad/attachment.htm>