7
$\begingroup$

While I am aware that VASP and QE are widely used for solid-state materials and PySCF/OpenMolcas are primarily designed for molecular systems.

I want to know why should I use PySCF/OpenMolcas for molecular calculations over VASP/QE and whether those packages could replace VASP/QE for periodic systems?

$\endgroup$
4
  • 2
    $\begingroup$ +1. The next four questions in this post (now removed) made the question a bit too broad when the main difference is the type of basis set used (there's no point in comparing "community support" for the Windows operating system vs for the Google Chrome browser, because their purposes are completely different, but if you do want to compare the community support availability between PySCF and OpenMOLCAS or between VASP and QuantumESPRESSO, it would be a perfectly appropriate new question; and if you want to compare it between PySCF and VASP, we can do that too but it's less interesting). $\endgroup$ Commented Feb 25 at 14:06
  • $\begingroup$ @NikeDattani, I think they are part of the question labeled as comparison points. Not to focus on 'community support' but as you mentioned that, I agree it is not an interesting point that's why I put it at last but I believe it makes sense to have such comparison among tools that has potential similarities. Besides what you mentioned in the answer which is pretty much informative, I was interested in the scalability and paralleziation part. $\endgroup$ Commented Feb 25 at 14:39
  • 1
    $\begingroup$ I think these are all suitable for different questions, otherwise the question becomes two broad for this platform. If you want to know about the scalability/paralellization of OpenMolcas, that can be asked in a post. If you want to know about the comparison between PySCF and OpenMolcas in terms of scalability/paralellization, that can also be asked, but it might still be considered very broad. Are you wondering about the parallelizability of the coupled cluster routine, the integrals routine, the Hartree-Fock routine, the FCI routine, or what? These are packages with 100s of programs. $\endgroup$ Commented Feb 25 at 15:34
  • $\begingroup$ @NikeDattani, I see your point and thanks for the answer. $\endgroup$ Commented Feb 26 at 0:47

1 Answer 1

6
$\begingroup$

The main difference is that PySCF, OpenMolcas, and other popular software such as Gaussian, ORCA, Molpro, Psi4, MRCC, CFOUR, Dalton, GAMESS, COLUMBUS, ACES, etc., are very well-developed for using Gaussian basis sets. In a typical molecule, there will be nuclei at certain irregular places, and the electronic wavefunction will look very different near those nuclei compared to away from those nuclei, and Gaussian basis sets that are centered at each nucleus, offer an excellent way to help model the shape of the wavefunction when we have such irregularities at very specific places.

In PySCF and OpenMOLCAS, we also have access to well-developed and extremely high-accuracy (compared to VASP and QuantumESPRESSO) wavefunction-based methods for solving the Schroedinger equation: (FCI), , and in OpenMolcas we even have methods built on top of CASSCF such as CASPT2, RASPT2, and GASPT2. FCI offers an exact solution to the Schroedinger equation for a given basis set, so not only is the Gaussian basis set capturing the fine details of the shape of the equation more accurately than a plane-wave basis set typically would, the actual solution to the Schroedinger equation based on that basis set is also capable to be more accurate in the Gaussian-based software.

VASP and Quantum ESPRESSO use plane-wave basis sets which don't capture the local irregularities of the wavefunction near the nuclei as well as Gaussian basis sets do. Furthermore, the capabilities to do calculations such as FCI, coupled cluster, etc. are not nearly as well-developed in VASP and Quantum ESPRESSO. VASP and Quantum ESPRESSO are designed to model comparatively huge materials with far more nuclei and electrons, so the methods available are mainly low-cost methods such as DFT with plane-wave basis sets: these will not be able to compete in accuracy with the above-mentioned wavefunction-based methods with Gaussian basis sets, and when modeling smaller molecules, the standards of accuracy tend to be a lot higher (it's not okay to just get a band gap accurate to within ±0.1 eV of the exact solution to the Schroedinger equation, the goal will be much closer to ±1 cm-1 accuracy or at least ±1 kcal/mol accuracy). A plane-wave basis set can certainly compete with a Guassian basis sets for accuracy if the former contains far more functions than the latter, but you would need orders of magnitude more functions, so the vast majority of the people that model molecules do not use plane-wave basis sets.

PySCF has some functionality to do calculations with periodic boundary conditions, but it is not nearly as developed as the functionality for that in VASP and QuantumESPRESSO. Contrarily, I do not yet recommend using VASP or Quantum ESPRESSO for calculations with Gaussian basis sets.

$\endgroup$
7
  • 1
    $\begingroup$ Are there actually studies on the precision obtainable with Gaussian basis sets? I'm not familiar with these types of basis sets, but at first they sound not systematically extendable to me. There have been precision studies with many DFT codes for crystals, but to the best of my knowledge, not a single code with a Gaussian basis set took part in these. $\endgroup$ Commented Feb 25 at 23:32
  • 1
    $\begingroup$ @GregorMichalicek please ask that in a new post. $\endgroup$ Commented Feb 25 at 23:33
  • 1
    $\begingroup$ @GregorMichalicek CP2K/Quickstep was evaluated in one such study acwf-verification.materialscloud.org $\endgroup$ Commented Feb 27 at 8:52
  • $\begingroup$ @Kristof Bal : Thank you for pointing that out. I should have been aware of this, as I also participated in that study. I somehow didn't associate that code with a pure Gaussian basis set, but you are right. Of course, there are two causes for possible imprecision in that study: The basis set and the chosen pseudopotential. Unfortunately, in that study CP2K/quickstep has unique characteristics for both of them. So, one can not clearly decide on the cause for disagreement to the reference data set. But at least we get an upper boundary. $\endgroup$ Commented Feb 27 at 20:22
  • 1
    $\begingroup$ @GregorMichalicek they are superior to planewaves for calculations on irregular small molecules as opposed to regular periodic materials. 100% (rounded) of the calculations of the former that have been done worldwide over the last several decades have been done with Gaussian basis sets and 0% with planewaves, for good reasons. No one does FCI or CASPT2 with planewaves. If you want to ask for references to studies that compare the two, please ask for that in the appropriate place. This question was about why one would use OpenMOLCAS vs QE and the answer is because it uses Gaussians. $\endgroup$ Commented Feb 27 at 20:48

You must log in to answer this question.

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