Optimize geometry

[ramkrishna]

Dear Sir,
        In optimize geometry window in 12.2 beta version, I have seen two types of optimizer method are there (quasinewton, fire). Can you please discuss a little bit about them? Which one will be better option?

Regards
Ramkrishna

[zh]

please refer to:
1. SIAM Rev. 52, pp. 3-54 (52 pages)
Numerical Methods for Electronic Structure Calculations of Materials
Yousef Saad, James R. Chelikowsky, and Suzanne M. Shontz
http://epubs.siam.org/sirev/resource/1/siread/v52/i1/p3_s1

2.  http://users.jyu.fi/~pekkosk/resources/FIRE.pdf
3. Daniel Sheppard, Rye Terrell, and Graeme Henkelman, Optimization methods for finding minimum energy paths, J. Chem. Phys. 128, 134106 (2008)
http://jcp.aip.org/resource/1/jcpsa6/v128/i13/p134106_s1

[Nordland]

Speaking only from my own empirical experience with them, I find the QuasiNewton works the best in the most cases, but there is systems where the FIRE method is better.

The quasi-newton has a major draw-back, that the time it takes to perform the next steps is order N^3 in terms of the number of atoms. So if you are doing a big system with a fast method to calculate the energy and forces, the time spent in the optimizer is the dominant part, so I am doing a classical calculation or doing a neb optimization, I will from time to time use the FIRE.

[Anders Blom]

Another time where FIRE can be better is if you want to take incremental steps, like 10 at the time, to see how things move along. QuasiNewton needs to build up a Hessian over a sequence of iterations, and each restart resets the Hessian, which makes QN perform badly or not even converge properly. This is why we use the FIRE method in the QuickOptimizer in the Builder in ATK 12.2.

[ramkrishna]

Thank you for these informations.

[Gundu]

Dear sir,

           In geometry optimization of BNNR(boron nitride nanoribben) with ATK-SE(11.8.2), i use Huckul's method and slater-koster methods seperataly for functionalizing the edges of ribbens with different chemical groups such as F2(flourine molecule ), select the h-boron and h-nitrogen in Huckul basis set with weitghing scheme of wolfberge and without choosing No SCF iteration and the run program, after some time i found that program terminated itself with completion  please me give detail regarding this problem, if any other method to approch for fuctionlizing edges of nano-ribben please suggest me

regards
khatri

[Anders Blom]

You have not provided enough information to allow a proper answer. Perhaps you ran out of memory, perhaps the computer overheated. Without some indication of error message or similar it's not possible to even guess.