Dear everyone, I've installed the parallel version of SIESTA. Strange enough, although all tests are successful, the program drops out when trying to calculate the BaTiO3.fdf test:
*...* ** *InitMesh: MESH = 36 x 36 x 36 = 46656 InitMesh: Mesh cutoff (required, used) = 200.000 223.210 Ry * Maximum dynamic memory allocated = 29 MB Error in Cholesky factorisation in cdiag Error in Cholesky factorisation in cdiag Stopping Program from Node: 0 Error in Cholesky factorisation in cdiag Error in Cholesky factorisation in cdiag Error in Cholesky factorisation in cdiag Error in Cholesky factorisation in cdiag Error in Cholesky factorisation in cdiag Error in Cholesky factorisation in cdiag Stopping Program from Node: 1 Stopping Program from Node: 7 Stopping Program from Node: 5 Stopping Program from Node: 6 Stopping Program from Node: 2 Stopping Program from Node: 4 Stopping Program from Node: 3* A similar error occurs in a graphite calculation. Both input files are OK, since they work with serial SIESTA. The program was compiled using MPIF90 and BLAS/LAPACK that I had built myself (no ESSL or MKL or whatever, since Sébastien Le Roux said in his guide that he couldn't compile SIESTA with MKL on his environment, which is quite similar to mine). I used the following compiler flags: -w -O3. Following the example of OpenMX, which works incorrectly when compiled with MPICC -O3 (version 9), I've recompiled SIESTA with -O2, and it still doesn't work for the BaTiO3 and graphite cases. Does anyone have a clue about this problem? What compile options should I use with the 9.0 Intel compiler? Is this really a compilation problem at all? Thanks, Vasilii