On Tue, 2012-03-27 at 10:41 +0200, Carlo Nervi wrote: > By using "mpirun -n 2" more discrepancies appears (sometimes very > different number of iterations, i.e. 86 instead of 57 in lda+U_gamma test).
the lda+U test is especially sensitive to small numerical differences, and the Gamma-only one seems to be the most sensitive of all. I suspect that the origin of such numerical differences between is mainly in mathematical libraries, blas and lapack. If so, there isn't much that can be done P. -- Paolo Giannozzi, IOM-Democritos and University of Udine, Italy