The ABINIT Users Mailing List ( CLOSED )

[aastaneh@rc.usf.edu]

Hi-

I built abinis/abinip from scratch so that I am able to link against OpenMPI.
Everything compiled fine, and the internal tests passed.
I did a `make tests_acc` in the tests directory, and found these failed 
tests: (I embedded the nature of the test below each one):


Test Suite v2:

Case_21 within 1.5 of tolerance (relative error : 5.84012795725582e-08 , 
accepted 5.0e-8 ) 
Tests 18 to 21 treat a BaTiO3 DDB, and reproduce the convergence
   study with sampling of the Brillouin zone for q wavevectors,
   as done in Ghosez et al, Ferroelectrics 206-207, 205 (1998),
   as well as the analysis of IFCs in term of dipole-dipole 
   contributions and short-ranged contributions. Eigenvectors are also
   analyzed.


Case_99 within 1.5 of tolerance (relative error : 0.000244558571777914 , 
accepted 2.0e-4 ) 
Test 99 RF of Niobium, at a rather large wavelength


Test Suite v3:

Case_83            failed (too large absolute error : 1.00000000000057e-07 , 
accepted 6.0e-8 ) 
Test 83 : test of non-linear features


Case_97            failed (too large absolute error : 0.0014246 , accepted 
0.0 ) 
Tests 96 and 97 : Lattice Wannier Functions


Test Suite v4:

Case_71            failed (too large absolute error : 3.29999999982428e-10 , 
accepted 0.0 ) 
qTests 50 to 83 concern linear and non-linear response computations
 as well as finite-electric field computations, and their analysis


These are all non-parallel tests above, since the developers do not currently 
support parallel tests in this fashion.
How should I respond to these test failures? Is there another library that I 
need to try linking to?

I ran configure with these options against a SMP Dual-Core Opteron system:

./configure \
--prefix=/opt/apps/abinit-5.4.4 \
--enable-64bit-flags \
--enable-mpi \
--enable-linalg \
--with-linalg-ldflags="-L/opt/lib/acml/pgi -lacml" \
--with-mpi-cc="mpicc" \
--with-mpi-fc="mpif90" \
--with-mpi-cxx="mpicxx" \
--with-mpi-run="mpirun"  \
CC=mpicc FC=mpif90

Thank you for your time.
Amin Astaneh
USF Research Computing
http://rc.usf.edu