[Pw_forum] structural relaxation with espresso 2.1.2 in parallel

Matteo Cococcioni matteoc at MIT.EDU
Fri Feb 18 03:21:31 CET 2005

Dear pwscf users

in our lab we have found some problems in doing structural relaxation with the
latest version of the espresso package (2.1.2). We find that, using bfgs (or
old-bfgs) algorithm, the structural optimization sistematically stops after the
second ionic step. The process doesn't crash but freezes without any error
message, after writing:

     NEW-OLD atomic charge density approx. for the potential
We are compiling the code with the latest intel compiler (ifc_8.1.023) and using
the lam mpi parallel environment (lam 7.0.6 compiled with icc_8.1.026 besides
the above fortran compiler).

The problem only appears in parallel jobs and seems to be somehow related to the
architecture of the machine. In fact it only shows up if we run the job on more
than one node. On the contrary, if we run a parallel job on the two proc of a
dual xeon for instance, the problem doesn't appear.

We have tried to increase the parameter mp_msgsiz_max in the Modules/mp.f90
routine as suggested few days ago by Paul Tangney and Paolo Giannozzi, but it
didn't solve the problem. Actually the problem seems to be unrelated to the
size of the system: we tried with very small ones (as the one whose input is
attached below) and we had the same situation.

We thought it could be a problem related to the use of several local disks on
different machines, but using a shared directory as outdir didn't help. 

Does anyone have any idea to solve this problem?

Many thanks in advance for any help.

Best regards,



%%%%%%%%%%%%%%%%%%%%%% Input file %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

    calculation = 'relax'
    pseudo_dir = './'
    ibrav=  8
    celldm(1) = 4.0d0
    celldm(2) = 1.0d0
    celldm(3) = 2.0d0
    nat=  2
    ntyp= 2
    nspin = 2
    ecutwfc =25.0
    ecutrho =200.0
    occupations = 'smearing'
    xc_type = 'PBE'
    electron_maxstep = 200
    mixing_beta = 0.3
    conv_thr =  1.0d-8
 C 12.0107d0 CUSPBE.RRKJ3
 O 15.9994d0 O.PBE.US.RRKJ3
  C  2.0 2.0 2.0  1 1 1
  O  2.0 2.0 5.0  1 1 1
K_POINTS {automatic}
1 1 1 0 0 0

More information about the users mailing list