ABINIT Discussion Forums

Dear All. I'm trying to study phonon spectrum of matrix isolated NaNO2 samples. I've calculated fully relaxed single crystal orthorombic body-centered structure, and phonons are close to experimental data. Now I need to calculate the phonon spectrum using the same functional and norm-conserving PPs for isolated singcle molecule. I started to increase lattice parameters by 20% at each step and try to relax the system with acell fixed. Geometry optimization is OK, but when I starting to calculate phonons like it is shown in RF1 tutorial I'm fail with calculating ground state, since the variable for stop criteria in RF1 for ground state is tolvrs1 which is in order of 1.0d-20 the last is never reached (the best result is 1d-5 or worse). Is it because system is not fully relaxed? May I use the other variable for stop criteria, like toldfe? Is there more general way to calculate free molecule in vacuum? Any help is approtiated a much!

roginovicci wrote:Is it because system is not fully relaxed?

Yes, most probably. The relaxed cell may be between your 20% increment. Why don't you use bfgs algorithm and let abinit relax the cell and positions by itself?

roginovicci wrote:May I use the other variable for stop criteria, like toldfe?

Yes, but don't. Tolvrs is tested to be the best.

roginovicci wrote:Is there more general way to calculate free molecule in vacuum?

Well, I'm sure there are posts on this forum about simulating molecules (you just have to find them). In short, you have to simulate the molecule and a vacuum around it as a "periodical cell". So, you have first to converge to thickness of the vacuum layer to ensure that there is no interaction between neighboring "cells". And then relax the molecule.

Here's something for you to read:

1] M. C. Payne, M. P. Teter, D. C. Allan, T. A. Arias and J. D. Joannopoulos.
Iterative minimization techniques for ab initio total-energy calculations: molecular dynamics and conjugate gradients.
Rev. Mod. Phys. 64 (4), 1045 (October 1992).

2] Electronic Structure: Basic Theory and Practical Methods, Richard M. Martin

I'm speculating a bit here but it seems to me that using DFPT as formulated in abinit to compute the vibrational frequencies of an isolated molecule is a bad idea, at the very least, horribly inefficient. You will spend a lot of time computing phonons associated with cell/cell vibrations and not enough time computing localized vibrations. It seems much better to use a molecular code like GAMESS, which can relax the molecule and compute the vibrational frequencies, all using DFT, for this part of the problem. I understand the desire to use one code for everything, so that you have a better handle on the approximations, but sometimes it's necessary to use more than one.

Many thanx dear all for answers. Let me make it clear. Actually I wonder how system behave becoming from 3d to 2d and single dimentional. I made some work arround in 2d development i.e. layers with some vacuum. And now I wonder what will happen when increase lattice paramiters. There is some intresting behaviour in geometry of molecule and I was wondering how does phonos spectrum acts like. Actually, I found that SCF do converges but extremly slow it needs more than 400 cycles to converge well.
I think increasing acell does really adds vacuum, but I'll try release whole cell with initial lattice parameters in 25% higher then 50% etc to see if there fully relaxed state with lots of vacuum exist.

And I know the GAMESS code as GAUSSIAN and CRYSTAL and others able to calculate single molecule, but I wondering the dinamic process in transition 3d-1d state. Probably CRYSTAL09 would be preferable tool to research the problem, but I do really like ABINIT in many ways, so I started here.

And once again thank you jzwanzig and ilukacevic very much for the help and literature adviced!