[QE-users] PHonon regression

[Scott Brozell]

Hi,

I am reporting a regression in QE 7.3 for PHonon/examples/example04
which calculates the normal modes of the molecule CH4.
We have typical hardware running various versions of RHELS, eg:
Model name:            Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz
Red Hat Enterprise Linux Server release 7.9 (Maipo)

There is no difference in the total energies from ch4.scf.out.
For a build with intel 19.0.5 (including mkl) and mvapich2 2.3.3
the frequencies in ch4.nm.out are initially all asterisks, but after
a minor change:
(
diff q-e-qe-7.3/PHonon/PH/dyndia.f90 q-e-qe-7.3formatchange/PHonon/PH/dyndia.f90
96c96
< 9010 format   (5x,'freq (',i5,') =',f15.6,' [THz] =',f15.6,' [cm-1]')
---
> 9010 format   (5x,'freq (',i5,') =',g15.6,' [THz] =',g15.6,' [cm-1]')
)
the frequencies are seen to be astronomical:
 **************************************************************************
         freq (    1) =  -0.479981+102 [THz] =  -0.160105+104 [cm-1]
         freq (    2) =  -0.206365+102 [THz] =  -0.688359+103 [cm-1]
         freq (    3) =  -0.192288+102 [THz] =  -0.641402+103 [cm-1]
         freq (    4) =  -0.174235+102 [THz] =  -0.581187+103 [cm-1]
         freq (    5) =  -0.124589+102 [THz] =  -0.415585+103 [cm-1]
         freq (    6) =  -0.621996+101 [THz] =  -0.207476+103 [cm-1]
         freq (    7) =  -0.599140+101 [THz] =  -0.199851+103 [cm-1]
         freq (    8) =  -0.137753E+94 [THz] =  -0.459496E+95 [cm-1]
         freq (    9) =   0.294112E+94 [THz] =   0.981053E+95 [cm-1]
         freq (   10) =   0.195386+101 [THz] =   0.651739+102 [cm-1]
         freq (   11) =   0.455999+101 [THz] =   0.152105+103 [cm-1]
         freq (   12) =   0.644921+101 [THz] =   0.215123+103 [cm-1]
         freq (   13) =   0.343296+102 [THz] =   0.114511+104 [cm-1]
         freq (   14) =   0.346276+102 [THz] =   0.115505+104 [cm-1]
         freq (   15) =   0.421103+102 [THz] =   0.140465+104 [cm-1]
 **************************************************************************

For a qe7.3 build with gnu 8.4.0, mvapich2 2.3.3, and mkl 2019.0.5
the frequencies in ch4.nm.out are merely bad:
 **************************************************************************
     freq (    1) =      -9.773705 [THz] =    -326.015720 [cm-1]
     freq (    2) =      -8.889882 [THz] =    -296.534532 [cm-1]
     freq (    3) =      -5.549399 [THz] =    -185.108021 [cm-1]
     freq (    4) =       1.585071 [THz] =      52.872283 [cm-1]
     freq (    5) =       5.255137 [THz] =     175.292486 [cm-1]
     freq (    6) =       6.649912 [THz] =     221.817198 [cm-1]
     freq (    7) =      34.550754 [THz] =    1152.489094 [cm-1]
     freq (    8) =      37.202830 [THz] =    1240.952824 [cm-1]
     freq (    9) =      37.503239 [THz] =    1250.973413 [cm-1]
     freq (   10) =      43.470790 [THz] =    1450.029471 [cm-1]
     freq (   11) =      43.472451 [THz] =    1450.084874 [cm-1]
     freq (   12) =      89.328051 [THz] =    2979.663039 [cm-1]
     freq (   13) =      90.261402 [THz] =    3010.796276 [cm-1]
     freq (   14) =      92.248874 [THz] =    3077.091230 [cm-1]
     freq (   15) =      94.624278 [THz] =    3156.326165 [cm-1]
 **************************************************************************

Note that the reference outputs are from v.6.0, but we have had
consistent agreement through 6.7.  7.3 is the first v7 we've installed.
This smells fishy even though it is only a toy calculation,
but i wanted to notify the community and get feedback before
submitting a gitlab issue.

Best regards,

Scott

-- 
Scott Brozell, Ph.D.
Scientific Applications Senior Engineer
Scientific Applications Group
Ohio Supercomputer Center
1224 Kinnear Rd.
Columbus, OH 43212