Page 1 of 1

Responce calculations and PAW datasets

Posted: Wed Sep 15, 2010 7:19 pm
by ruslan
Hello, abinit users!

I am trying to do a responce calculation with PAW potentials. Converging density is quickly done but the problem occurs when i try to converge wavefunctions using iscf -3

Code: Select all

# 2. converge responce to d/dk
iscf1 -3
kptopt1 2
rfphon1 0
rfelfd1 2
tolwfr1 1.0d-22

# 3. responce to Q=0
getddk2 2
kptopt2 2
rfelfd2 3

It seems, that the system does not converge at all, as can be seen in output

Code: Select all

     iter   2DEtotal(Ha)        deltaE(Ha) residm    vres2
-ETOT  1  -45.736192492181     -4.574E+01 7.977E-01 0.000E+00
 ETOT  2  -46.354527419797     -6.183E-01 5.299E-01 0.000E+00
 ETOT  3  -46.401130927362     -4.660E-02 1.272E+00 0.000E+00
 ETOT  4  -46.397162406766      3.969E-03 1.191E+00 0.000E+00
 ETOT  5  -46.398055674238     -8.933E-04 1.497E+00 0.000E+00
 ETOT  6  -46.347604885697      5.045E-02 2.263E+00 0.000E+00
 ETOT  7  -46.358587559944     -1.098E-02 2.309E+00 0.000E+00
 ETOT  8  -46.318810908718      3.978E-02 3.369E+00 0.000E+00
 ETOT  9  -46.298235890568      2.058E-02 2.828E+00 0.000E+00
 ETOT 10  -46.277683366679      2.055E-02 4.298E+00 0.000E+00
 ETOT 11  -46.275402375590      2.281E-03 3.143E+00 0.000E+00
 ETOT 12  -46.220355643065      5.505E-02 5.058E+00 0.000E+00
 ETOT 13  -46.238456591518     -1.810E-02 3.807E+00 0.000E+00
 ETOT 14  -46.206598389753      3.186E-02 5.678E+00 0.000E+00
 ETOT 15  -46.186604453668      1.999E-02 4.359E+00 0.000E+00
 ETOT 16  -46.172633619957      1.397E-02 6.182E+00 0.000E+00
 ETOT 17  -46.177863714078     -5.230E-03 4.804E+00 0.000E+00
 ETOT 18  -46.132110261736      4.575E-02 6.591E+00 0.000E+00
 ETOT 19  -46.153101989510     -2.099E-02 5.201E+00 0.000E+00
 ETOT 20  -46.121826117739      3.128E-02 6.923E+00 0.000E+00
 ETOT 21  -46.114789289932      7.037E-03 5.609E+00 0.000E+00
 ETOT 22  -46.107450928316      7.338E-03 7.191E+00 0.000E+00
 ETOT 23  -46.114623621245     -7.173E-03 5.963E+00 0.000E+00
 ETOT 24  -46.070461895234      4.416E-02 7.408E+00 0.000E+00
 ETOT 25  -46.097296090667     -2.683E-02 6.266E+00 0.000E+00
 ETOT 26  -46.077519928066      1.978E-02 7.584E+00 0.000E+00
 ETOT 27  -46.069253403734      8.267E-03 6.523E+00 0.000E+00
 ETOT 28  -46.062380503081      6.873E-03 7.727E+00 0.000E+00
 ETOT 29  -46.076245680092     -1.387E-02 6.738E+00 0.000E+00
....

Is it a general limitation of current abinit implementation or am i missing some option?

Second question is more general and regards doing responce calculations with wavefunctions loaded from _WFK file (.fhi pseudopotentials). Restarting produces get following warning:
-P-0000 hdr_check: WARNING -
-P-0000 Restart of self-consistent calculation need translated wavefunctions.
-P-0000 Indeed, critical differences between current calculation and
-P-0000 restart file have been detected in:
-P-0000 * the number, position, or weight of k-points
-P-0000 * the format of wavefunctions (istwfk)

allthough i set istwfk *1 during initial run. Is this warning rather harmless or does it have negative consequences regarding calculations? Is there a way to get rid of this warning? Till now i did RF calculations this way:
- convergence of density with kptopt 3 and shiftk 0 0 0 to produce _WFK input file (can i take advantage of symmetry in this run? should i converge wavefunctions with iscf -2 afterwards?)
- individual runs for each q-point with irdwfk 1 and kptopt 3

Third question is about number of grid points in responce calculation. I found that responce calculations for my system with ngkpt>12 cannot be done due to Out-of-Memory errors (a job with 96 processors on cray XT6m for q=0.5,0.5,0 of A15 structure):

Code: Select all

 newkpt: loop on k-points done in parallel
aborting job:
Fatal error in MPIDI_CRAY_SMPClus_Allreduce: Other MPI error, error stack:
MPIDI_CRAY_SMPClus_Allreduce(209): MPIDI_CRAY_SMPClus_Allreduce(sendbuf=0x2aaaf52db010, recvbuf=0x2aaafef1c010, count=20480000, type=MPI_DOUBLE_PRECISION, op=MPI_SUM, comm=MPI_COMM_WORLD) failed
MPIR_Allreduce(193)..............: Unable to allocate 163840000 bytes of memory for temporary buffer (probably out of memory)
[NID 00171] 2010-08-19 14:46:56 Apid 19696: initiated application termination
Application 19696 exit codes: 1
Application 19696 exit signals: Killed
Application 19696 resources: utime 0, stime 0

Increasing number of processors did not really help. What can be done to reduce memory needs of this calculation?

Any help is appreciated,
Ruslan Zinetullin

Re: Responce calculations and PAW datasets

Posted: Fri Sep 17, 2010 10:39 am
by ruslan
Additionally, during a RF calculation for Gamma point i get strange error from rhofermi3:
scfcv3, nstep= 200
scfcv3, nstep= 200
scfcv3, nstep= 200
scfcv3, nstep= 200

rhofermi3 : ERROR -
Not yet allowed within PAW !
-P-0000
-P-0000 leave_new : decision taken to exit ...

Calculations with q/=0 are currently running without problems. What should i do to get q=0,0,0 working?

I also wanted to mention, that a parameter pawxcdev had to be set to zero to get PAW RF running. I think it should be documented somewhere, at least in the description of it.

Greetings,
Ruslan Zinetullin

Re: Responce calculations and PAW datasets

Posted: Fri Sep 17, 2010 11:15 am
by ruslan
Responce calculations for q/=Gamma stopped with following error:
-P-0000 - newkpt: read input wf with ikpt,npw= 76 7556, make ikpt,npw= 10 7556
-P-0000 - newkpt: read input wf with ikpt,npw= 77 7522, make ikpt,npw= 11 7522
-P-0000 leave_test : synchronization done...
newkpt: loop on k-points done in parallel
pareigocc : MPI_ALLREDUCE


Subroutine Unknown:0:BUG
Does not support idir>0 for that choice.
Action : contact ABINIT group.

.Delivered 1 WARNINGs and 0 COMMENTs to log file.


any suggestions?

Ruslan Zinetullin

Re: Responce calculations and PAW datasets

Posted: Mon Sep 20, 2010 10:51 am
by torrent
Dear Ruslan,

Response calculations are still not open within PAW in the official version and GIVE WRONG RESULTS !

The developper team is currently testing and correcting this implementation.

See releases notes :
D.4 The use of Density Functional Perturbation Theory with PAW
(phonons and electric field) is in progress"


Keep in touch...
and have a look in releases notes in next versions.