UT theoretical chemistry code forum

The ground state energy of a neutral oxygen vacancy in an oxide is closed shell singlet and hence the total magnetic moment is zero. I ran CI-NEB with ISPIN=1 to compute the migration barrier of this vacancy and obtained a value of 1.81 eV. Forces were converged to a value below 0.01 eV/A.
Out of curiosity I ran the same CL-NEB with ISPIN=2 and obtained a saddle point configuration that is triplet and hence the magnetic moment is 2. The surprising thing is that the migration barrier in that case is 2 eV! Notice that I'm using 3 images in both cases and only image 2 (The saddle point) is triplet. Furthermore, both saddle points (singlet and triplet) have almost the same atomic configuration.
My question is: Is not it supposed that CL-NEB combined with ISPIN=2 search for the lowest possible migration barrier? Why did not my second calculation converge to a singlet with a migration barrier 1.81 eV?
Any hint will be appreciated.

Perhaps the energy of the endpoint(s) were lowered by more than the saddle when you considered spin-polarization?

In both NEB calculations, I used the same end points (Same POSCAR and OUTCAR). These end points correspond to the ground state which is in this case singlet (closed shell). While relaxing these two end points, both ISPIN=1 and ISPIN=2 converge to the singlet solution.
So, in fact what happened is that NEB with ISPIN=2 gave me a saddle point configuration that is 0.2 eV higher than the one obtained with ISPIN=1.

It is certainly possible to have excited electronic states, and your ISPIN=2 calculation could converge to such a higher-energy state. This can be due to the initial spin, which is set very high by default in an ISPIN=2 calculation. Try starting your converged ISPIN=1 NEB calculation with an ISPIN=2 calculation using MAGMOM=0 for all ions. This should relax to either the same or to a lower saddle energy (as you are expecting).

Thank you for your suggestions. I tend also to believe also that it is sort of trapping in an excited state.