7
$\begingroup$

I had some experience with the ACFDT-RPA, but despite its promising features it's not practical most of the time, and for that reason, I'm looking for a new method for reference calculations.

QMC is the first alternative for me. Since I'm not familiar with QMC, what are its advantages/disadvantages over ACFDT-RPA? For example, is it computationally more expensive? Is restarting the calculation possible? (it's not possible with the ACFDT-RPA since it's implemented as a post-processing method due to its cost).

Principally, both methods give the exact ground state energy.

$\endgroup$
7
  • 1
    $\begingroup$ +1. But there's 15 different types of QMC listed here and they have varying computational costs. MP2 costs O(N^5) for N orbitals, and I don't know about ACFDT-RPA, but if you want to know the cost of ACFDT-RPA you might have to ask a separate question: mattermodeling.meta.stackexchange.com/q/193/5. You are asking whether QMC is more expensive than "all other correlated methods" which is too open-ended in my opinion. A better question might be to choose a particular QMC and a particular correlated method and ask which costs more. $\endgroup$ Sep 13, 2020 at 20:24
  • $\begingroup$ I got your point, but I have no experience with the QMC calculations, that's why I asked. If it's not something expensive than MP2 or ACFDT-RPA in general, I will probably be interested in it. Your suggestion also sounds good. I'll ask a question like that then by editing this question. $\endgroup$
    – Alfred
    Sep 13, 2020 at 20:36
  • $\begingroup$ QMC methods tend to be "heuristics", so the run time varies. It also depends on how small you want your error bar (QMC methods tend to be stochastic so they don't give just one number, but instead a number with an error bar). DMC can in principle be as expensive as FCI, if you want to use FCI to fix your wavefunction nodes. Maybe "what are the advantages/disadvantages of QMC over ACFDT-RPA" is a better question. Computational cost is not easily compared because in DMC the cost can be low or high depending on the accuracy you want, and depending on the nature of the wavefunction and the system. $\endgroup$ Sep 13, 2020 at 21:17
  • $\begingroup$ The question has now improved. I'd say that the advantages are that you can restart the calculation, parallelize very well, and if you can wait long enough you might be able to converge to arbitrary accuracy, while the disadvantages will be the requirement of a lot of resources (many CPUs and a lot of time) and stochasticity (you don't get a number, but a number with some error-bar which will be quite big if you don't run the calculation for long enough). $\endgroup$ Sep 13, 2020 at 22:36
  • 1
    $\begingroup$ It depends on so many things: how many electrons, how many orbitals, how strongly correlated are the electrons, etc. $\endgroup$ Sep 14, 2020 at 2:28

1 Answer 1

1
$\begingroup$

Advantage for QMC:

While this question is quite open ended due to there being at least 15 different types of QMC, one thing that all 15 of those methods have in common is that as far as I know they can all be restarted when the calculation has to stop for whatever reason. Since you pointed out that ACFDT-RPA calculations cannot be restarted, the ability to restart a calculation provides us with at least one advantage that QMC has over ACFDT-RPA.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .