Re: [QE-users] NVE simulation in cp.x - large fluctuation in temperature

2018-04-14 Thread Jie Peng 
Ari:

Indeed! You are right on spot. The too small size unit cell that leads to
inaccurate structure and energy is responsible for large temperature
fluctuation here.

I am trying to increase the size of the supercell now.

Thanks!

On Sat, Apr 14, 2018 at 6:06 AM, Ari P Seitsonen 
wrote:

>
> Dear Jie,
>
>   I do not know about the other options, but
>
>  1) simulating a periodic system of only three atoms with only the Gamma
> point cannot lead to an accurate description of the electronic structure,
> and thus the forces on the ionic cores,
>
>  2) the standard deviation in the fluctuations is supposed to be
> ~sqrt[2/(3*N_atoms)) * T_average ~= 350 K in your case, and the
> instantaneous fluctuations are of course larger
>
>   Just my quick thoughts...
>
> Greetings from Sunny Zurich,
>
>apsi
>
> -=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=
> -=*=-=*=-=*=-=*=-
>   Ari Paavo Seitsonen / ari.p.seitso...@iki.fi / http://www.iki.fi/~apsi/
> Ecole Normale Supérieure (ENS), Département de Chimie, Paris
> Mobile (F) : +33 789 37 24 25(CH) : +41 79 71 90 935
>
>
> On Fri, 13 Apr 2018, Jie Peng wrote:
>
> Dear all
>> I have been running MD simulations on HfS2 using cp.x code in Quantum
>> espresso. I start from initial
>> configuration obtained from pwscf vc-relax, and relax the system using
>> cp.x by consecutive steps of:
>> electron relaxation->ionic relaxation->cell relaxation. Then, I just
>> directly start a NVE simulation
>> starting from the equilibrium configuration. I expect the system to
>> almost stay stationary or the
>> temperature should be very small since I am allowing dynamics in a system
>> that is already in
>> equilibrium. However, what I see is a huge fluctuation in the tmpp output
>> of cp.x, as I attach a figure
>> showing variation of tmpp (Ionic temperature) with simulation time
>> [IMAGE]
>>
>>
>> I did this because it is suggested in the user guide you should apply an
>> initial displacement to the
>> atoms in your system after the relaxation since otherwise there will not
>> be any dynamics. But what I
>> see here is a large fluctuation of the system temperature.
>>
>> The thinking or questions here are
>>
>> 1.Does the tmpp represents the physical temperature of the system here? I
>> think it should be since it
>> is the temperature corresponding to kinetic energy of the ions.
>>
>> 2.It above point is true, why is the temperature varying so fiercely? Am
>> I setting incorrect
>> parameters, for instance the timestep or the fictitious mass? But I took
>> those from previous simulation
>> steps where I did the relaxation, and they all worked well since they
>> successfully drived my system to
>> equilibrium, satisfying the convergence threshold on total energy, forces
>> acting on atoms, and the
>> fictitious electron kinetic energy. I am confused at this point.
>>
>> The input file for NVE simulation is attached here:
>>
>> 
>> calculation='cp',
>> title='Halfnium disulfide'
>> restart_mode='restart',
>> ndr=53,
>> ndw=54,
>> nstep=5,
>> iprint=10
>> isave=100,
>> tstress = .true.
>> tprnfor = .true.
>> dt=10,
>> wf_collect=.true.
>> etot_conv_thr=1e-6
>> forc_conv_thr=1e-3
>> ekin_conv_thr=1e-5
>> prefix='HfS2',
>> pseudo_dir='/home/jpeng/HfS2/potential'
>> outdir='./tmp/',
>>  /
>>  
>> ibrav= 4,
>> a=3.6529
>>c=5.6544
>> nat=  3, ntyp= 2,
>> ecutwfc =50
>> vdw_corr='DFT-D',
>>  !   lspinorb=.true.
>>  !   noncolin=.true.
>>  !   ecutrho=300
>>  !   nbnd=14
>> !occupations='smearing'
>> !smearing='gaussian'
>> !   degauss=0.01
>>  !  nspin=2
>>  !   starting_magnetization(1)=0.1
>> ! Hf  95.94  Hf.pbe-mt_fhi.UPF
>> ! S  32.065  S.pbe-mt_fhi.UPF
>> /
>>  
>> electron_dynamics='verlet'
>> electron_velocities='zero'
>> emass=400
>> emass_cutoff=1
>> /
>>  
>> ion_dynamics = 'verlet'
>> ion_damping=0.1
>> !ion_nstepe=10
>>  /
>>  
>> cell_dynamics = 'none'
>>
>> /
>> ATOMIC_SPECIES
>>  Hf  95.94  Hf.pbe-mt_fhi.UPF
>>  S  32.065  S.pbe-mt_fhi.UPF
>> ATOMIC_POSITIONS (crystal)
>> Hf  -0.0  -0.0  -0.0
>> S0.7   0.3   0.257234636
>> S0.3   0.7  -0.257234636
>>
>> Anyone could help me on it? Thank you very much.
>>
>> Best
>> Jie
>> --
>> 
>> --
>> --
>> Jie Peng
>> PhD student
>> 2134 Glenn Martin Hall, Mechanical Engineering, University of Maryland
>> College Park, Maryland, USA
>> Phone:(+1) 240-495-9445
>> Email: jiep...@umd.edu
>>
>>
>>
> ___
> users mailing list
> users@lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users
>



--

Re: [QE-users] NVE simulation in cp.x - large fluctuation in temperature

2018-04-14 Thread Stefano de Gironcoli 

Dear Jie Peng,

 suppose you were running a model harmonic system in 1 dimension.

  M a = - K x

 at fixed energy  E.

 The kinetic energy would fluctuate harmonically between 0 (at 
maximum/minimum elongation) and E at the equilibrium distance .


 On average the Kinetic energy would be E/2 and its fluctuation some 
big fraction of E^2


 Something like sigma^2 = 1/T \int_0^T (E cos^2(2pi t/T) -E/2)^2 dt = 
E^2 1/T \int_0^T (cos(4pi t/T)/2)^2 dt = (E/2)^2 1/2pi \int_0^2pi 
cos^2(x) dx = (E/2)^2 / 2


  or

  sigma =  1/sqrt(2) * E/2 = 1/sqrt(2) avg EKin

  with 1 degree of freedom the mean square fluctuation of the kinetic 
energy is 70% of its average !


  you have 3 atoms in your cell hence 9 degrees of freedom. Assuming 
each contributes independently to the average this goes down by a factor 
1/sqrt(9)=1/3


  actually more likely just 1/sqrt(6) as the total momentum is 
conserved so only 6 modes at Gamma are actually excited...


  If you perform your simulation in a bigger supercell with more atoms 
(more degrees of freedom) the average will be more stable ( 
proportionally to  1/sqrt(#deg.of.freedom-3 )  ... moreover the thermal 
excitations of vibrational modes will be sampled more faithfully.


  best

stefano


On 13/04/2018 21:39, Jie Peng wrote:

Dear all

I have been running MD simulations on HfS2 using cp.x code in Quantum 
espresso. I start from initial configuration obtained from pwscf 
vc-relax, and relax the system using cp.x by consecutive steps of: 
electron relaxation->ionic relaxation->cell relaxation. Then, I just 
directly start a NVE simulation starting from the equilibrium 
configuration. I expect the system to almost stay stationary or the 
temperature should be very small since I am allowing dynamics in a 
system that is already in equilibrium. However, what I see is a huge 
fluctuation in the /tmpp/ output of cp.x, as I attach a figure showing 
variation of tmpp (Ionic temperature) with simulation time


I did this because it is suggested in the user guide you should apply 
an initial displacement to the atoms in your system after the 
relaxation since otherwise there will not be any dynamics. But what I 
see here is a large fluctuation of the system temperature.


The thinking or questions here are

1.Does the tmpp represents the physical temperature of the system 
here? I think it should be since it is the temperature corresponding 
to kinetic energy of the ions.


2.It above point is true, why is the temperature varying so fiercely? 
Am I setting incorrect parameters, for instance the timestep or the 
fictitious mass? But I took those from previous simulation steps where 
I did the relaxation, and they all worked well since they successfully 
drived my system to equilibrium, satisfying the convergence threshold 
on total energy, forces acting on atoms, and the fictitious electron 
kinetic energy. I am confused at this point.


The input file for NVE simulation is attached here:

//
/    calculation='cp',/
/    title='Halfnium disulfide'/
/    restart_mode='restart',/
/    ndr=53,/
/    ndw=54,/
/    nstep=5,/
/    iprint=10/
/    isave=100,/
/    tstress = .true./
/    tprnfor = .true./
/    dt=10,/
/    wf_collect=.true./
/    etot_conv_thr=1e-6/
/    forc_conv_thr=1e-3/
/    ekin_conv_thr=1e-5/
/    prefix='HfS2',/
/    pseudo_dir='/home/jpeng/HfS2/potential'/
/    outdir='./tmp/',/
/ //
/ /
/    ibrav= 4,/
/    a=3.6529/
/   c=5.6544/
/    nat=  3, ntyp= 2,/
/    ecutwfc =50/
/    vdw_corr='DFT-D',/
/ !   lspinorb=.true./
/ !   noncolin=.true./
/ !   ecutrho=300/
/ !   nbnd=14/
/!    occupations='smearing'/
/!    smearing='gaussian'/
/!   degauss=0.01/
/ !  nspin=2/
/ !   starting_magnetization(1)=0.1/
/! Hf  95.94  Hf.pbe-mt_fhi.UPF/
/! S  32.065  S.pbe-mt_fhi.UPF/
///
/ /
/    electron_dynamics='verlet'/
/    electron_velocities='zero'/
/    emass=400/
/    emass_cutoff=1/
///
/ /
/    ion_dynamics = 'verlet'/
/    ion_damping=0.1/
/!    ion_nstepe=10/
/ //
/ /
/    cell_dynamics = 'none'/
//
///
/ATOMIC_SPECIES/
/ Hf  95.94  Hf.pbe-mt_fhi.UPF/
/ S  32.065  S.pbe-mt_fhi.UPF/
/ATOMIC_POSITIONS (crystal)/
/Hf      -0.0  -0.0  -0.0/
/S        0.7   0.3   0.257234636/
/S        0.3   0.7  -0.257234636/

Anyone could help me on it? Thank you very much.

Best
Jie
--

Jie Peng
PhD student
2134 Glenn Martin Hall, Mechanical Engineering, University of Maryland
College Park, Maryland, USA
Phone:(+1) 240-495-9445
Email: jiep...@umd.edu 



___
users mailing list
users@lists.quantum-espresso.org
https://lists.quantum-espresso.org/mailman/listinfo/users


___
users mailing list
users@lists.quantum-espresso.org
https://lists.quantum-espresso.org/mailman/listinfo/users

Re: [QE-users] NVE simulation in cp.x - large fluctuation in temperature

2018-04-14 Thread Ari P Seitsonen 


Dear Jie,

  I do not know about the other options, but

 1) simulating a periodic system of only three atoms with only the Gamma 
point cannot lead to an accurate description of the electronic structure, 
and thus the forces on the ionic cores,


 2) the standard deviation in the fluctuations is supposed to be 
~sqrt[2/(3*N_atoms)) * T_average ~= 350 K in your case, and the 
instantaneous fluctuations are of course larger


  Just my quick thoughts...

Greetings from Sunny Zurich,

   apsi

-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-
  Ari Paavo Seitsonen / ari.p.seitso...@iki.fi / http://www.iki.fi/~apsi/
Ecole Normale Supérieure (ENS), Département de Chimie, Paris
Mobile (F) : +33 789 37 24 25(CH) : +41 79 71 90 935


On Fri, 13 Apr 2018, Jie Peng wrote:


Dear all
I have been running MD simulations on HfS2 using cp.x code in Quantum espresso. 
I start from initial
configuration obtained from pwscf vc-relax, and relax the system using cp.x by 
consecutive steps of:
electron relaxation->ionic relaxation->cell relaxation. Then, I just directly 
start a NVE simulation
starting from the equilibrium configuration. I expect the system to almost stay 
stationary or the
temperature should be very small since I am allowing dynamics in a system that 
is already in
equilibrium. However, what I see is a huge fluctuation in the tmpp output of 
cp.x, as I attach a figure
showing variation of tmpp (Ionic temperature) with simulation time
[IMAGE]

I did this because it is suggested in the user guide you should apply an 
initial displacement to the
atoms in your system after the relaxation since otherwise there will not be any 
dynamics. But what I
see here is a large fluctuation of the system temperature.

The thinking or questions here are

1.Does the tmpp represents the physical temperature of the system here? I think 
it should be since it
is the temperature corresponding to kinetic energy of the ions.

2.It above point is true, why is the temperature varying so fiercely? Am I 
setting incorrect
parameters, for instance the timestep or the fictitious mass? But I took those 
from previous simulation
steps where I did the relaxation, and they all worked well since they 
successfully drived my system to
equilibrium, satisfying the convergence threshold on total energy, forces 
acting on atoms, and the
fictitious electron kinetic energy. I am confused at this point.

The input file for NVE simulation is attached here:


    calculation='cp',
    title='Halfnium disulfide'
    restart_mode='restart',
    ndr=53,
    ndw=54,
    nstep=5,
    iprint=10
    isave=100,
    tstress = .true.
    tprnfor = .true.
    dt=10,
    wf_collect=.true.
    etot_conv_thr=1e-6
    forc_conv_thr=1e-3
    ekin_conv_thr=1e-5
    prefix='HfS2',
    pseudo_dir='/home/jpeng/HfS2/potential'
    outdir='./tmp/',
 /
 
    ibrav= 4,
    a=3.6529
   c=5.6544
    nat=  3, ntyp= 2,
    ecutwfc =50
    vdw_corr='DFT-D',
 !   lspinorb=.true.
 !   noncolin=.true.
 !   ecutrho=300
 !   nbnd=14
!    occupations='smearing'
!    smearing='gaussian'
!   degauss=0.01
 !  nspin=2
 !   starting_magnetization(1)=0.1
! Hf  95.94  Hf.pbe-mt_fhi.UPF
! S  32.065  S.pbe-mt_fhi.UPF
/
 
    electron_dynamics='verlet'
    electron_velocities='zero'
    emass=400
    emass_cutoff=1
/
 
    ion_dynamics = 'verlet'
    ion_damping=0.1
!    ion_nstepe=10
 /
 
    cell_dynamics = 'none'
 
/
ATOMIC_SPECIES
 Hf  95.94  Hf.pbe-mt_fhi.UPF
 S  32.065  S.pbe-mt_fhi.UPF
ATOMIC_POSITIONS (crystal)
Hf      -0.0  -0.0  -0.0
S        0.7   0.3   0.257234636
S        0.3   0.7  -0.257234636

Anyone could help me on it? Thank you very much.

Best
Jie
--
--
--
Jie Peng
PhD student
2134 Glenn Martin Hall, Mechanical Engineering, University of Maryland
College Park, Maryland, USA
Phone:(+1) 240-495-9445
Email: jiep...@umd.edu


___
users mailing list
users@lists.quantum-espresso.org
https://lists.quantum-espresso.org/mailman/listinfo/users