some question about calculating Hessian matrix

Hi All,

Recently I use lammps_local_package(https://cmrl.lab.uic.edu/lammps_hessian.html) to compute the Hessian matrix, howerver i can’t get a convergence value when i change the epsilon value(displacement value for the centeral finite difference method). For the command module: hessian group_ID epsilon etol ftol nsteps maxiter.

I have used the following two command:

hessian SIA 5.0e-6 1.0e-12 1.0e-12 10000 10000

hessian SIA 5.0e-8

Then i diagonalized the hessian.out file to get the eigenvalues and i found that the eigenvalues of these two matrics are very different. Then i have tried a smaller epsilon value but there is no convergence.

Besides my lammps version is lammps-12Dec18.

Thank you so much for your time and consideration.

Sincerely,

Yingzhao He

Hi,

What are the units of the finite displacement? If the units are A, then the numbers 5.0e-6 and 5.0e-8 are bad choices.
Also, what are the units of the convergence parameters (1.0e-12 1.0e-12) for the energy minimization process here?

Bruce

Thank you for your reply. My units style is metal so the distance units is A. The reason why i choose these epsilon value is that in the Journal of Nuclear Materials437(1-3), 438-444 ) they set the Δ to 1.0e-7nm to calculate the elements of Hessian matrix. And the etol is the stopping tolerance for energy (energy units) for minimizer, ftol is the stopping tolerance for force (force units) for minimizer.

I think that is too small. Usually one uses 0.005-0.015 A.

Bruce

何莹钊 <m201771134@…7448…5…> 于 2019年5月27日周一 04:21写道: