[SIESTA-L] MD: failing diagonalization

[Anders Odell]

Dear Siesta users,

I'm doing a molecular dynamics run on a H2 molecule, microcanonical 
ensemble. For some reason the calculation stops after some timesteps 
(usually a few hundred á 0.5 fs) due to failed diagonalization:

siesta:                 ==============================
                           Begin MD step =    341
                       ==============================

siesta: iscf   Eharris(eV)      E_KS(eV)   FreeEng(eV)   dDmax  Ef(eV)
siesta:    1      -31.4677      -31.4727      -31.4727  0.0274  0.1838
DSYGV failed to converge.            1
 off-diagonal elements of an intermediate
tridiagonal form did not converge to zero
Terminating due to failed diagonalisation
Stopping Program
***** MPI-error in rank 0 Routine MPI_Abort : Terminating after call to 
MPI_Abort *****
...

The energy and dynamics looks good up to this point so there is nothing 
indicating that something is wrong. I use a constrained number of SCF 
iterations in each timestep. In this example 3 SCF iter/timestep. Could 
anyone help me with an explanation for this and a possible remedy?

Thanks!
/Anders