fspiga / hpcs-dirac-exomol Goto Github PK

CONFIGURE
---------

ScaLAPACK is required. ELPA is optional (it has to be compiled manually, 
the flag -D__ELPA has to be specified).

Under "makes/" there are several Makefiles, all related to Darwin (HPCS)


INPUT DEFINITION
----------------

Input parameters are passed as options to the command line, e.g.

./diag_INTEL.x <GEN-MAT> <NROOTS> <DIAGONALIZER> <GENERATOR>

- "GEN-MAT" is the size of the matrix 
- "NROOTS" is the number of eigen-state to compute, starting from index 1.
- "DIAGONALIZER" is the type of solver to use
	PDSYEVD (default), PDSYEVX, ELPA-1STAGE, ELPA-2STAGE
- "GENERATOR" is strategy used to generate the random input matrix 
	SYM-POSITIVE-O3, SYM-POSITIVE-O2, RANDOM-LOCAL (default)
	
NOTE: if PDSYEVD, NROOTS is ignored (by default all the eigenvalues and 
      eigenvectors are computed)
      
      
KNOWN PROBLEMS
--------------

- PDSYEVX sometimes fails ( INFO .ne. ZERO) due to not enough workspace
  allocated to compute the eigen-vectors. Reasons still unknown, it is 
  related to the number of NROOTS
- ELPA does not report explicitly any error code (e.g. INFO) back to the 
  main application
- RANDOM-LOCAL is really faster but it is not clear if the resultin matrix 
  is definite positive or semi-positive. SYM-POSITIVE-O3 is the slowest but
  should produce an input with N eigen-state