4
$\begingroup$

I'm just getting started with quantum chemistry using some simple test problems. I have a few of the software packages installed (ASE, Psi4, GPAW etc) and everything is working - no issues with the software itself - but there is a bewildering variety of methods, from semi-empirical (PM7) to DFT (B3LYP) to coupled cluster (CCSD(T)); plus options for each such as basis functions, and nothing in the documentation gives any specific advice on what to use when.

Is there concrete guidance on which method is appropriate for which type of problems?

What I'm mostly interested is biological systems: molecules in water at room temperature, and both covalent and non-covalent interactions; most of the atoms are CHNOSP. An example would be taking a system of a few hundred atoms and optimizing the coordinates (either locally or trying to find the global minimum). A slightly more ambitious project would be to optimize the coordinates a few hundred atoms within a much larger system of thousands of atoms (where the other atoms have partial charges and produce electrostatic forces, but are otherwise fixed).

$\endgroup$
4
  • 3
    $\begingroup$ The simulation of chemical systems is literally the subject of a whole field inside chemistry, so taking an introductory course would be much more helpful (and you could start asking actual questions). Also the size of your system and resources automatically exclude most if not all quantum chemistry methods. $\endgroup$ – Greg Dec 20 '20 at 4:52
  • 1
    $\begingroup$ This is more computational than quantum. $\endgroup$ – Alchimista Dec 20 '20 at 9:35
  • 1
    $\begingroup$ A large part of computational chemistry is about benchmarking methods. I.e. testing how good different methods/basis sets are for a problem. If you have a specific problem in mind try searching the literature for it and check what methods were already tested/used. The best practical method for a problem is hard to predict and most of the time determined empirically. People simply test a bunch of methods and compare it to known results. Another "filter" is also what you can actually do with your hardware. The theoretically best method is often simply not feasible especially for large systems. $\endgroup$ – Hans Wurst Dec 21 '20 at 17:15
  • $\begingroup$ Related: chemistry.stackexchange.com/q/27302/4945 The question is probably better suited on Matter Modeling. $\endgroup$ – Martin - マーチン Dec 27 '20 at 16:26
4
$\begingroup$

A good guide comes from this series of slides by Jan Řezáč:

enter image description here

It seems that semiempirical methods (of which PM6-D3H4 in MOPAC is probably most accurate for the types of systems encountered in biology) are the only methods able to handle structure optimization of a whole protein (or whole domain of a larger protein). DFT can handle an order of magnitude fewer atoms, and MP2 or CCSD(T) are mainly used for creating benchmark data sets to test the accuracy of methods, such as S66.

$\endgroup$
2
  • $\begingroup$ I get a virus warning for the pdf file. From when are these slides? There are multiple forests of methods missing from that scheme... $\endgroup$ – Martin - マーチン Dec 27 '20 at 16:29
  • $\begingroup$ @Martin-マーチン What software gives the warning? Virustotal thinks it's okay, virustotal.com/gui/url/… Slides are from 2014. Could you fill in the methods that are missing? $\endgroup$ – Alex I Dec 28 '20 at 2:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.