Geometry Optimization in Computational Chemistry

[hiltac]

I've read an article about computational chemistry in which the authors were performing a geometry optimization. For this purpose, they firstly optimized the geometry at HF/6-311++G(d,p) level with full relaxation on the potentiel energy surface (what does that mean exactly ? Can we extract this information from the basis set or from the HF method in general?).
And after that, they re-optimized at B3LYP level (with the same basis set). Why perform HF and then B3LYP ?
Thank you for your help !

 

[e.bar.goum]

It's my understanding that you need HF before you can do B3LYP, since B3LYP takes a linear combination of Hartree-Fock exchange functional. You'd do this, because you can optimize better with B3LYP than with ab-initio models like HF.

My understanding of this is pretty limited though, hopefully someone else can give a deeper answer.

See here: http://en.wikipedia.org/wiki/Hybrid_functional

 

[cgk]

@e.bar.guom: No, it does not work that way. HF and DFT are usually done with the same programs, and hybrid functionals like B3LYP will typically use the HF code paths for their exact exchange contributions. However, there is no reason to perform a HF calculation before a hybrid DFT one. The self-consistent HF solution never enters anywhere in the Hybrid DFT calculation.

@OP: I would consider "full relaxation on the potentiel energy surface" as a sciency term for "without any artificial constraints on the geometry" (e.g., symmetry constraints, fixed bond lengths, etc)---i.e., to mean exactly what one would understand under "we performed a geometry optimization" without qualifying this any further. But context might be important, maybe they mean some more involved global optimization of the structure (but if they do, they would make this clear in their paper).

Regarding your main question: No, optimizing geometries first at HF level and then at DFT level normally does not make much sense. What can make sense it to first pre-optimize a geometry on DFT level using pure functionals (i.e., functionals without exact exchange) and then do a further optimization using hybrid functionals or HF. The reason being that calculations with pure functionals are much faster hybrid DFT/HF calculations and still provide good geometries in most cases, so a pre-optimization with pure functionals can result in a much better starting geometry for the more expensive HF/HybridKS calculations, and thus less total computation time. However, the other way around this does not work, and for mixing HF and B3LYP this would not work in any case.

The only reasons I can imagine for first doing a HF optimization would be to work around implementation problems in the quantum chemistry package they are using (e.g., numerical noise in the DFT integration could interfere with the geometry optimizer, so the optimizer might work more reliably with HF, which has no numerical integrations. Or the package they are using has 2nd derivatives for HF, but not for DFT, or similar things).

 

[hiltac]

Thank you for your answer.

And what about the electron correlation? If we use DFT after HF method we can introduce the electron correlation whereas if we begin with DFT we can lose it by using HF, am I wrong?

 

[DrDu]

Why don't you give the reference to the article you were reading? Maybe it contains some clue as to why the did what they did.​