[Pw_forum] (Slightly) different energies based on ordering of atomic positions

[Mike Mehl]

I've been doing frozen phonon calculations with pwscf, and found a 
difference in the output of the code based on how I ordered the list of 
atomic positions.

Specifically:

Both of the attached runs compute the energy of an 8 atom supercell of 
IrV in the L1_0 structure with no displacements away from the 
equilibrium positions, so they should give the same energy.  They don't.

The difference between the runs is that the atomic positions were 
generated from two different irreducible representations from Harold 
Stokes's FINDSYM program, and the input differs slightly.  That is, the 
atomic positions for the run "m000.U1.in" are

Ir   0.00000000000000     0.00000000000000     0.00000000000000
Ir   0.00000000000000     5.21947856004048     0.00000000000000
Ir   0.00000000000000    15.65843568012144     0.00000000000000
Ir   0.00000000000000    10.43895712008096     0.00000000000000
V    2.60973928002024    13.04869640010120    -3.46622752200000
V    2.60973928002024    13.04869640010120     3.46622752200000
V    2.60973928002024     7.82921784006072    -3.46622752200000
V    2.60973928002024     7.82921784006072     3.46622752200000

and for "m000.U2.in" we have

Ir   0.00000000000000    10.43895712008096     0.00000000000000
Ir   0.00000000000000     0.00000000000000     0.00000000000000
Ir   0.00000000000000     5.21947856004048     0.00000000000000
Ir   0.00000000000000    15.65843568012144     0.00000000000000
V    2.60973928002024    13.04869640010120    -3.46622752200000
V    2.60973928002024    13.04869640010120     3.46622752200000
V    2.60973928002024     7.82921784006072    -3.46622752200000
V    2.60973928002024     7.82921784006072     3.46622752200000

e.g., the Ir atoms are arranged in a different order.

As I said, I get different results.  The relevant differences in the 
outputs seem to be:

$ diff m000.U1.out m000.U2.out

<      Found symmetry operation: I + (  0.0000  0.2500  0.2500) 
This is a supercell, fractional translation are disabled
---
 >      Found symmetry operation: I + (  0.0000  0.5000  0.5000)
This is a supercell, fractional translation are disabled

<
<      the Fermi energy is    14.9323 ev
<
< !    total energy              =    -346.21707104 Ry
<      Harris-Foulkes estimate   =    -346.21707104 Ry
<      estimated scf accuracy    <          9.2E-11 Ry
---
 >
 >      the Fermi energy is    14.9317 ev
 >
 > !    total energy              =    -346.21709373 Ry
 >      Harris-Foulkes estimate   =    -346.21709373 Ry
 >      estimated scf accuracy    <          6.2E-11 Ry

Of course, the eigenvalues are different as well.

I suspect that the problem is that the first non-symmorphic symmetry 
operation found is different in the two runs, so different space group 
operations are thrown out of the two calculations.

Admittedly, this is a contrived situation (displacing the atoms away 
from equilibrium consistent with the symmetry of the phonon mode removes 
the problem), and the energy difference is inconsequential, so I'm 
hesitant to label it as a bug. Nevertheless I find it a bit 
disconcerting to get two different answers from what should be identical 
input.

-- 
Michael J. Mehl
Head, Center for Computational Materials Science
Naval Research Laboratory Code 6390
Washington DC
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: m000.U1.in
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20100415/550c3323/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: m000.U2.in
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20100415/550c3323/attachment-0001.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: m000.U1.out
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20100415/550c3323/attachment-0002.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: m000.U2.out
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20100415/550c3323/attachment-0003.ksh>