The ABINIT Users Mailing List ( CLOSED )

[TORRENT Marc <marc.torrent@cea.fr>]

Hi Vincent,

Apparently, there are two separated problems when using tphysel:

1- Etotal vs EtotalDC:

Now you use a sufficiently small tolerance on density residual.
Are now Etotal and Etotal DC equal when tphysel is used ?
(I cannot read it with your pdf file...)
I mean 10 digits or more... Are they closer than the results obtained 
with toldfe=1.d-9.
If yes, the problem is closed.
If not this is a bug.... and the origin of it is mysterious...


2- Etotal/EtotalDC vs etotal:

When tphysel<>0:
As I wrote yesterday, I can easily explain the difference.
Etotal is e_intern - tphysel*entropy (implemented in prtene.F90)
etotal is e_intern - tsmear*entropy (implemented in etotfor.F90)

The first question is: which is the correct one ?
Looking at your graph, it seems that Etotal is better (it converges wr 
to tsmear).
Let Matthieu answer this question...

The second question is: why is it implemented like this ?
Again, let Matthieu answer this question...
Perhaps it is sufficient to simply modify etotfor.F90 routine and put in 
it the same formula as in prtene.F90.



Marc



Vincent Chevrier a écrit :
> Hi,
>
> I'm really sorry but I swapped the tphysel values in the previous
> postings. It SHOULD have read:
> toldfe=1e-9, tphysel=0
>     >>>>>>>>> Etotal=   -2.24784859172166E+02
>     >>>> Etotal (DC)=   -2.24784864956616E+02
>     etotal              -2.2478486496E+02
>
> toldfe=1e-9, tphysel=0.002131
>     >>>>>>>>> Etotal=   -2.24785888288957E+02
>     >>>> Etotal (DC)=   -2.24785888278458E+02
>     etotal              -2.2478510947E+02
>
> And so Etotal != etotal when tphysel != 0.
>
> As Matthieu suggested I varied tsmear and as Marc suggested used a
> tolvrs=1e-15. See the attached plot. It therefore appears that the value
> of etotal in the .out file is missing some contribution when using
> tphysel. (I'm only plotting tsmear=[0.005:0.015] but the same trend
> continues up to 0.05 and I didn't test beyond that).
>
> One should therefore use Etotal or EtotalDC when using tphysel and NOT
> etotal? Is this a bug?
>
> Thanks
> Vincent
>
>   
> > Guilty, as charged. The concept behind tphysel is in PRB 65 035111:
> >
> > The result of a combination of a physical (Femi Dirac tphysel)
> > temperature and an artificial (tsmear, e.g. Marzari cold-smearing)
> > smearing is an effective smearing scheme, where the temperature for
> > the entropic term is tphysel and the smearing function is a
> > mathematical convolution (determines the occupation numbers, which
> > gives the value of the entropy) which depends both on tsmear and
> > tphysel. This is why in the tphysel case you specify both but it is
> > tphysel which should appear in front of the entropy, not tsmear.
> >
> > Note that for physical temperature calculations you always have the
> > reference of setting tsmear=T and occopt 3, and then converging wrt
> > k-points. This avoids the additional cross-convergence of tsmear and
> > nkpt, but obviously is heavier (that's the whole point of
> > tphysel+tsmear, to improve k-point convergence).
> >
> >     
> > > toldfe=1e-9, tphysel=0
> > >     >>>>>>>>> Etotal= -2.24785888288957E+02
> > >     >>>> Etotal (DC)=   -2.24785888278458E+02
> > >     etotal                      -2.2478510947E+02
> > >       
> > this is wierd: the Etotal-s are already free energies, with the
> > entropic term subtracted. So etotal should be equal to them. Perhaps
> > it is recalculated somewhere else before the final output? And that
> > point would have a small bug. Could you test this difference as a
> > function of tsmear for instance, to see if there is a trend?
> >
> >
> >     
> > > toldfe=1e-9, tphysel=0.002131
> > >     >>>>>>>>> Etotal= -2.24784859172166E+02
> > >     >>>> Etotal (DC)=   -2.24784864956616E+02
> > >     etotal                      -2.2478486496E+02
> > >       
> > Here etotal looks identical to Etotal(DC), as it should be, but the
> > difference with Etotal is about 1.e-6, instead of 1.e-8 as above... I
> > see no particular reason that the DC estimate should be worse in this
> > case wrt the direct calculation of the energy. Has anyone seen this in
> > norm-conserving calculations? Again, here, could you test if there is
> > a trend wrt tphysel? Perhaps with a very reduced cutoff if the runs
> > take time (and if the discrepancy is still there, of course).
> >
> > cheers,
> >
> > Matthieu
> >
> > On Thu, Sep 3, 2009 at 2:12 PM, TORRENT Marc<marc.torrent@cea.fr> wrote:
> >     
> > > Dear Vincent,
> > >
> > > Some explanations (but not all).
> > >
> > > etotal is used to converge the scf cycle and contains an entropy energy
> > > computed with tsmear.
> > > It is obtained by direct scheme if iscf<10, by double-counting scheme if
> > > iscf>=10:
> > >
> > > iscf=7 :  etotal=e_intern      - tsmear*entropy
> > > iscf=17:  etotal=eDC_intern - tsmear*entropy
> > >
> > > Etotal (EtotalDC) is just printed at the end of the SCF cycle and
> > > recomputed
> > > at that level.
> > > It uses tphysel if not zero, else tsmear:
> > >
> > > If tphysel<>0:
> > > Etotal=e_intern      - tphysel*entropy
> > > EtotalDC=eDC_intern - tphysel*entropy
> > >
> > > If tphysel=0:
> > > Etotal=e_intern      - tsmear*entropy
> > > EtotalDC=eDC_intern - tsmear*entropy
> > >
> > > So, it explains the differences between etotal and Etotal.
> > > Why is it implemented like this: I don't know.
> > > The developer who introduced tphysel should be able to explain that.
> > >
> > >
> > > Concerning the difference between Etotal and EtotalDC when tphysel<>0...
> > > it
> > > sounds strange.
> > > But I cannot explore further if I'm not sure of the convergence of your
> > > SCF
> > > cycle.
> > > toldfe is not the right criterium.
> > > you should use tolvrs and put it to a very low value (i.e 1.d-15 or
> > > lower).
> > >
> > > Regards,
> > >
> > > Marc
> > >
> > >
> > >
> > > Etotal and Etotal
> > >
> > > Vincent Chevrier a écrit :
> > >
> > > Dear Bernard,
> > > thank you for your reply. I've rerun with toldfe=1e-9. But etotal does
> > > not
> > > always agree with Etotal(DC). See below, in all cases tsmear=0.01 and
> > > the
> > > runs only differ in the parameters stated.
> > >
> > > toldfe=1e-5, tphysel=0
> > >     >>>>>>>>> Etotal= -2.24785447837035E+02
> > >     >>>> Etotal (DC)=   -2.24785888234037E+02
> > >     etotal                      -2.2478510911E+02
> > >
> > > toldfe=1e-5, tphysel=0.002131
> > >     >>>>>>>>> Etotal= -2.24785745210972E+02
> > >     >>>> Etotal (DC)= -2.24784864246079E+02
> > >     etotal            -2.2478486425E+02
> > >
> > >       
> > > when tphysel=0.002131 then etotal = Etotal(DC), but when tphysel=0 then
> > > etotal != Etotal(DC). Why is this, or more specifically what is:
> > >
> > > 1 - the cause of the difference between Etotal(DC) and etotal
> > > 2 - the effect of tphysel on the difference between Etotal(DC) and
> > > etotal
> > >
> > > Thanks
> > > Vincent
> > >
> > >
> > >
> > > Dear Vincent,
> > >
> > > With iscf=17 (the default value of iscf),  ETOT
> > > is equal to the so-called "double counting" expression of the energy
> > >
> > >
> > > written "Etotal (DC)" in the .out file (you could check it in your .out
> > > file).
> > >
> > >
> > > Moreover, Etotal and Etotal (DC) (both written in the .out file) are
> > >
> > >
> > > expected to agree if
> > >
> > >
> > > your calculation is well converged.
> > > In your case, your convergence criterion
> > > is weak (toldfe 1.d-05) so Etotal and Etotal (DC)
> > > are not equal. You could
> > > increase you convergence criterion to check this.
> > >
> > > Best regards
> > > Bernard
> > >
> > >
> > > Vincent Chevrier a écrit :
> > >
> > >
> > > Hi,
> > > I am wondering what the difference is between Etotal and etotal for PAW
> > >
> > >
> > > calcs. I can't remember ever seeing a difference stated on the ABINIT
> > > website. I'm especially interested in knowing how tphysel affects this
> > > difference.
> > >
> > >
> > > I completed two runs with the same structure the only difference being
> > >
> > >
> > > tphysel=0 and tphysel=0.00213.
> > >
> > >
> > > >From the .out file:
> > > For tphysel=0:
> > >     Kinetic energy  =  1.72081044902126E+01
> > >     Hartree energy  =  1.17823500013938E+01
> > >     XC energy       = -3.59810665254476E+01
> > >     Ewald energy    = -1.07436906857906E+02
> > >     PspCore energy  =  8.93535605840166E+00
> > >     Loc. psp. energy= -4.47984809673791E+01
> > >     Spherical terms = -7.44950150296981E+01
> > >     >>>>> Internal E= -2.24785658830422E+02
> > >     -kT*entropy     =  2.10993387856660E-04
> > >     >>>>>>>>> Etotal= -2.24785447837035E+02
> > > and
> > >     etotal  -2.2478510911E+02
> > > For tphysel=0.002131
> > >     Kinetic energy  =  1.72075689848333E+01
> > >     Hartree energy  =  1.17845516746048E+01
> > >     XC energy       = -3.59818457081452E+01
> > >     Ewald energy    = -1.07436906857906E+02
> > >     PspCore energy  =  8.93535605840166E+00
> > >     Loc. psp. energy= -4.48011025224259E+01
> > >     Spherical terms = -7.44958705044103E+01
> > >     >>>>> Internal E= -2.24788248875047E+02
> > >     -kT*entropy     =  2.50366407550064E-03
> > >     >>>>>>>>> Etotal= -2.24785745210972E+02
> > > and
> > >     etotal  -2.2478486425E+02
> > > below is the input file. Can anybody help me understand:
> > > 1 - the cause of the difference between Etotal and etotal
> > > 2 - the effect of tphysel on the difference between Etotal and etotal
> > >
> > >
> > > Thanks
> > >
> > >
> > > Vincent
> > > Input file:
> > >     ndtset   2
> > >    tphysel1  0.002131
> > >    tphysel2  0.0
> > >     occopt   4
> > >     tsmear   0.01
> > >  pawecutdg   15
> > >       ecut   15
> > >     kptopt   1
> > >    nshiftk   1
> > >     shiftk   0.5 0.5 0.5
> > >   kptrlatt   6  0  0
> > >              0  3  0
> > >              0  0 10
> > >      nstep   50
> > >     toldfe   1.0d-5
> > >     ntypat   2
> > >      znucl   3 14
> > >      natom   34
> > >      typat   1  1  1  1  2  1  1  1
> > >              2  1  1  1  2  1  1  1
> > >              2  1  1  1  1  2  1  1
> > >              1  2  1  1  1  2  1  1
> > >              1  2
> > >      acell   1.5065427965E+01  2.8615352526E+01  8.4083802112E+00 Bohr
> > >
> > >
> > > rprim   1.0 0.0 0.0
> > >
> > >
> > >              0.0 1.0 0.0
> > >              0.0 0.0 1.0
> > >       xred   7.4996953092E-13  5.0000000000E-01  0.0000000000E+00
> > >              9.6480532055E-02  1.9420479791E-01  0.0000000000E+00
> > >
> > >
> > > 2.7080350417E-01  3.4610078252E-01  0.0000000000E+00
> > > 1.5292044305E-01  2.5787557735E-02  0.0000000000E+00
> > > 4.1748605305E-01  1.6031572532E-01  0.0000000000E+00
> > > 2.5955975937E-01  9.5181187368E-02  5.0000000000E-01
> > > 4.0978302750E-01  2.5635768786E-01  5.0000000000E-01
> > > 9.6925526499E-02  3.9393640703E-01  5.0000000000E-01
> > > 4.2837868374E-01  4.3177429281E-01  5.0000000000E-01
> > > 9.0351946795E-01  8.0579520209E-01  0.0000000000E+00
> > > 7.2919649583E-01  6.5389921748E-01  0.0000000000E+00
> > > 8.4707955695E-01  9.7421244227E-01  0.0000000000E+00
> > > 5.8251394695E-01  8.3968427468E-01  0.0000000000E+00
> > > 7.4044024063E-01  9.0481881263E-01  5.0000000000E-01
> > > 5.9021697250E-01  7.4364231214E-01  5.0000000000E-01
> > > 9.0307447350E-01  6.0606359297E-01  5.0000000000E-01
> > > 5.7162131627E-01  5.6822570719E-01  5.0000000000E-01
> > > 5.0000000000E-01  0.0000000000E+00  0.0000000000E+00
> > > 4.0351946795E-01  6.9420479791E-01  0.0000000000E+00
> > > 2.2919649583E-01  8.4610078252E-01  0.0000000000E+00
> > > 3.4707955695E-01  5.2578755773E-01  0.0000000000E+00
> > > 8.2513946947E-02  6.6031572532E-01  0.0000000000E+00
> > > 2.4044024063E-01  5.9518118737E-01  5.0000000000E-01
> > > 9.0216972503E-02  7.5635768786E-01  5.0000000000E-01
> > > 4.0307447350E-01  8.9393640703E-01  5.0000000000E-01
> > > 7.1621316266E-02  9.3177429281E-01  5.0000000000E-01
> > > 5.9648053206E-01  3.0579520209E-01  0.0000000000E+00
> > > 7.7080350417E-01  1.5389921748E-01  0.0000000000E+00
> > > 6.5292044305E-01  4.7421244227E-01  0.0000000000E+00
> > > 9.1748605305E-01  3.3968427468E-01  0.0000000000E+00
> > > 7.5955975937E-01  4.0481881263E-01  5.0000000000E-01
> > > 9.0978302750E-01  2.4364231214E-01  5.0000000000E-01
> > > 5.9692552650E-01  1.0606359297E-01  5.0000000000E-01
> > > 9.2837868374E-01  6.8225707188E-02  5.0000000000E-01
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >       
> >
> > --
> > ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > Dr. Matthieu Verstraete
> >
> > Universite de Liège
> > Institut de Physique, Bat. B5
> > Allée du 6 aout, 17
> > B- 4000 Sart Tilman, Liège
> > Belgium
> >
> > Phone : +32 4 366 37 50
> > Fax   : +32 4 366 36 29
> >
> > Mail : matthieu.jean.verstraete@gmail.com
> >
> >