relaxation+spin

[Cyrille]

Dear all

I have a funny problem when trying to relax a molecule (or more precisely a 2D network of molecules).
I have fist performed a spin polarized calculation LSDA and everything goes well. The relaxation is smooth and II get a magnetic solution. Then I have tried to do a non magnetic calculation. But instead of using LDA I have used LSDA and starting from zero magnetization. The first scf cycle does not converge (after 100 iterations) but the calculation continues anyway (after forces have been evaluated and system relaxed). However in this new scf cycle the system has acquired a magnetization as if a "random magnetic kick" had been given to the system... Finally the relaxation continues and at the end the system converges towards the magnetic solution..

Why is it so?

Thanks in advance

Cyrille

[Daniele Stradi]

Dear Cyrille,

the behavior you observe is due to the fact that a new density matrix has been set up in the new relaxation step.

In the OptimizeGeometry function,

http://www.quantumwise.com/documents/manuals/latest/ReferenceManual/index.html/ref.optimizegeometry.html

there is a parameter called "scf_restart_step_length". The default is set to 0.1 Angstrom. If any atomic displacement between two consecutive optimization steps is larger than this value, a new density matrix will be set up, and the one calculated from the previous geometry will not be used.

 

[Cyrille]

OK thanks Daniele that's clear.
I can increase the number of electronic scf step to be (almost) sure that I do not encounter this situation again.

Cyrille