using Numba

[Konjkov]

Hello Rick

I am writing a package for quantum chemical calculations using the Monte Carlo method with python3 and numba

While I work in the local repository, I noticed that using the numba allows you to write easy-to-read and at the same time fast code. Unfortunately, Monte Carlo calculations are not self-sufficient and require input data on orbitals calculated in the classical quantum-mechanical program. I wrote a converter for this MOLDEN2QMC, however, it is very inconvenient that there is no single standard for storing molecular orbitals, especially in the case of MCSCF methods and orbital-optimised one.

I wanted to ask if there are any plans for the further development of the pyquante package and wat kind of performance tools you plan to use (I saw an example code on JULIA), but this does not make it possible to reuse all available python code.

Is the numba your choice and how long does it take to rewrite the RHF/UHF pyquante code with numba to check that they are as fast as the C++/FORTRAN code?

PS: It would also be interesting to study the systems expanded in the basis of the Slater orbitals not only in Gaussian with CASSCF method, I can say that there is no program that does such calculations including ADF

PPS: Despite the fact that I write code in my free time, the Python3 allows me to write fast enough to be productive.

Best, Vladimir

[rpmuller]

Vladimir,

Unfortunately, I moved into management a few years back, and have very little time for active development of pyquante. Numba is indeed very interesting, and I don't think it would take much work, but I think those are directions that should be explored by others.

Rick

On Tue, Jul 14, 2020 at 1:51 AM Vladimir notifications@github.com wrote:

Hello Rick

I am writing a package for quantum chemical calculations using the Monte Carlo method with python3 and numba https://numba.pydata.org/numba-doc/latest/user/5minguide.html

While I work in the local repository, I noticed that using the numba allows you to write easy-to-read and at the same time fast code. Unfortunately, Monte Carlo calculations are not self-sufficient and require input data on orbitals calculated in the classical quantum-mechanical program. I wrote a converter for this MOLDEN2QMC https://github.com/Konjkov/molden2qmc, however, it is very inconvenient that there is no single standard for storing molecular orbitals, especially in the case of MCSCF methods and orbital-optimised one.

I wanted to ask if there are any plans for the further development of the pyquante package and wat kind of performance tools you plan to use (I saw an example code on JULIA), but this does not make it possible to reuse all available python code.

Is the numba your choice and how long does it take to rewrite the RHF/UHF pyquante code with numba to check that they are as fast as the C++/FORTRAN code?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/rpmuller/pyquante2/issues/9, or unsubscribe https://github.com/notifications/unsubscribe-auth/AADSDMTAM5OUPK7X4GFDUUTR3QE7NANCNFSM4OZH3UMA .

-- Rick Muller rpmuller@gmail.com 505-750-7557

[Konjkov]

Hello Rick

I have significant progress in developing quantum chemistry program using python and numba (numpy, scipy) and found it to be a promising approach https://github.com/Konjkov/pycasino

after the implementation of the pseudopotential, I will try to implement the Hartree-Foсk method.

numba, despite the lack of a stable version, allows you to write code that works 2-5 times slower than Fortran.

Best, Vladimir

[rpmuller]

Thanks. As you probably gather from my commit history, I'm not actively developing pyquante right now, but I'm very interested in the utility of packages like Numba to make the package faster. You're positive experience in this area is exciting. Thanks for posting.