Agreed with Oleg. Basis set Limit is "lower" within DFT framework than with post-HF methods.
I would suggest Dunning basis set for post-HF methods instead of Pople. These (aug-)cc-pVxZ was designed for post-HF methods. I would otherwise also suggest Def2-xZVP(P) (x = T,Q,..).
Then, TZ or QZ with post-HF method depends on many other parameters (calculated property, system size, auxiliary basis sets for RI, etc...).
The commonly accepted "gold standard" method is CCSD(T)/CBS, where CBS is a methodology to do a number of lower level HF and MP2 calculations and extrapolate them to a complete basis set (CBS) limit. This does require several calculations and may not be necessary for many chemical structures and properties. For single calculations, CCSD(T) with the aug-cc-pVTZ basis set is probably the best you can normally do. Note, however, that the "best" method is costly (as are many other "best" things in life!) - the CCSD(T) method scales as the 6th power of the number of basis functions (N^6), so the calculations can get very long very easily!
Hi, for the CC and or MPx I would suggest to use the extrapolation to the CBS limit. It is not that difficult to use as it might seem and you will get much improved results for virtually the same cost as for one bigger cc-pVxZ calculation. Also even these basis sets look "big" the BSSE cannot be ignored and "one calculation" changes to five anyway.
For the DFT I would say a good/new functional treating dispersion combined with reasonably big Pople or cc basis set would be the choice. So b3lyp alone is not a good choice in general but it was parameterized on a "small" Pople basis set so with DFT (imho) there is no simple general rule but better look at the papers of the functional authors and use what they were using.
Also from your question it is not clear if you are talking about calculating single point energies or optimization of geometry. But the CCSD(T)/CBS and scaled MP3 MP2.X/CBS (if the system is too big) are working for both these purposes - and you omit inconsistent description of geometry and calculating "very accurate energy" on a "wrong geometry". (you can look at our papers for a more detailed idea)
Mr. Wijesingha, it would be more useful discussion if you first determine which type of sytem you have and which structural and/or physical properties you would like to predict/optimize. because of both ab initio and DFT methods provide variety of theoretical levels where are balanced the accuracy and computational costs. In this respect even in your question it is not shown the Truhlar's M05, M06 and related functionals, which are currently accepred as among the most accurate in prediction for example of optical properties. For inorganic compounds for example good accuracy of the physical properties could obtain with SDD or LANL2DZ basis set. Therefore it is really of importance to clarify which system, which phase, which properties and more, you exactly would like to compute by ab initio or DFT methods.
Despite of this, please find the following links and references, which are very useful for the choice of the proper theoretical level depending of the variety of molecular systems, phases and properties of interest:
2. The M06 suite of density functionals for main group thermochemistry, thermochemical kinetics, noncovalent interactions, excited states, and transition elements: two new functionals and systematic testing of four M06 functionals and 12 other functionals, Yan Zhao, Donald G. Truhlar, Theoretical Chemistry Accounts
April 2008, Volume 119, Issue 5-6, p 525,
3. Coupled cluster, B2PLYP and M06-2X investigation of the thermochemistry of five-membered nitrogen containing heterocycles, furan, and thiophene, Pablo A. Denis, Theoretical Chemistry Accounts, May 2011, Volume 129, Issue 2, pp 219-227
Especially you can payed attantion on:
4. Encyclopedia of Computational Chemistry, 5 volumes, ISBN: 0-471-96588-X (http://www.wiley.com/legacy/wileychi/ecc/opening.html)
where very comprehensively are described the variety of DFT and ab initio methods, the variety of basis sets and diverce number of molecular systems, where the accuracy on the theoretical methods have been tested.
There is not really a 'correct' answer to this. The choice of basis set depends on system, methodology, and properties of interest. For example, polarised triple-zeta basis functions are nowadays standard, and should be good enough for *most* things. However, if you are interested in excited states or anionic systems, where the electron density tends to be more diffuse, you may need to augment your basis set with diffuse functions. If you are working on very small systems, then you may be looking for extremely high accuracy and it's not uncommon to find quintuple-zeta or even larger basis sets used, usually in conjunction with some CC based method.
Also, bear in mind that MP2 and CI use virtual orbitals to include electron correlation. Therefore your virtual space has to be large enough to be able to describe this correlation well.
An example where a small basis set (such as split-valence) would be appropriate would be in the early stages of a geometry optimisation, for a structure well away from it's energetic minimum. Here, a large basis set would be overkill in terms of computational expense. You can use the smaller basis to get a reasonable geometry, then switch to the larger basis set to get an accurate final structure.
I think, we are discussing about two subjects. Although, they are connected to each other but we should note to their priority. First you should specify which types of ab-initio, semi-classical or semi-empirical calculation method is suitable for studying the system. Of course you can refer to published articles which are similar to yours and choose the method. After choosing the method, you should choose the basis functions. The choice of basis functions depends on the system. For example 3-21G is suitable basis function for carbon based clusters. LANL2DZ is suitable basis function for elements which are placed after third row of periodic table. Of course for using the large basis function such as LANL2DZ for big clusters you should use suitable computer. Therefore not only you should note to he system but also you should no e to your hardware. Some times the elements can guide you for choosing the best basis function. For example if you deal with Mn-element since the element includes d-orbitals you should use a suitable basis function for adding the d-orbitals in calculations. Finally, after calculation comparison between your results and other published similar theoretical and experimental results can clarify that whether you have chose the correct calculation method and basis function or not?
There is no correct answer , it depends in the molecule ex does it have a metal , is it an organic molecule , how big it is , has many rings , does it have a halogen , you calculate and if you are interested in the energy or geometry optimization with respect to computation time . The only way is to test but mostly b3lyp with an augmented basis set would give good results
I do not agree with Zoi. I explained a general rule. At first, we should choose the ab-initio method. Then we should choose the basis function. Since, some scientists have used some methods and basis functions and have shown that they can be suitable for some kinds of molecules before, now we are aware about them and use them automatically. But for new idea, we should choose and then proof our choices are suitable. After us, others may use our choice and follow us.
It depends on the system you are simulating in addition to the objective of the calculation. You need to look for previous publications that tackled cases comparable to your particular system, method, and objective. Another factor to consider is the available computational resources, like available memory or the time consumed. You may find that for a particular computation a more complete basis set is too demanding while a lesser complete set could give acceptable results with less computational resources. It is really a case-dependent answer.