continuing run for hessian matrix calculation

Message

tlchan · #1 Post by **tlchan** » Thu Jan 26, 2006 4:23 pm

I'm running a hessian matrix calculation(ibrion=5) for a big system. The calculation takes a long time, and I wonder how I can continue the calculation in case of a machine crash. Does vasp save the intermediate calculations of the hessian matrix? Thank you very much.

#2 Post by **admin** » Mon Jan 30, 2006 2:29 pm

no, VASP does not save the intermediate steps. in case of a crash you have to start all over again. Please note that
--) use IBRION = 5 only for post-processing a converged run
--) by choosing seective dynamics, you can pick out the modes which are reasonable (decreasing the number of vibrational degrees of freedom will speed up the run significantly)

tlchan · #3 Post by **tlchan** » Fri Feb 10, 2006 11:22 pm

After the calculation, VASP outputs the eigenvectors and eigenvalues of the dynamical matrix(at gamma point, I think) in OUTCAR. This output is divided into 6 columns:x, y, z, dx, dy and dz. I believe the x, y, z are the components of the eigenvectors. But I have no idea what are the dx, dy and dz. What are these 3 columns correspond to? Thank you.

#4 Post by **admin** » Mon Feb 13, 2006 1:06 pm

no, x,y and z are the positions of the ions (as read from POSCAR), dx, dy, and dz are the eigenvectors .

alex · #5 Post by **alex** » Mon Feb 27, 2006 12:33 pm

If you are interested, I wrote myself an extension to finite_diff.F and main.F
It's somehow alpha-stage, it does not work always, but most of the time. And it's tricky, if you have more than one restart to do.

jkawasaki · #6 Post by **jkawasaki** » Thu Jun 14, 2007 3:00 pm

VASP shows both mass weighted and non-mass weighted values for the eigenvectors (dx,dy,dz). what are the units for the non-mass weighted eigenvectors? are they in angstrom? i did frequency calculations for water, and VASP calculated non-mass weighted eigenvectors of magnitude 0.7. this seems very high compared to the 0.96 angstrom bond length of h-o in water.

jkawasaki · #7 Post by **jkawasaki** » Thu Jun 14, 2007 3:28 pm

nvm, just realized that eigenvectors should be dimensionless