search

votca / xtp

GW-BSE for excited state Quantum Chemistry in a Gaussian Orbital basis, electronic spectroscopy with QM/MM, charge and energy dynamics in complex molecular systems
29 stars 15 forks source link

Libint on Ubuntu only supports L=4 #608

Closed JensWehner closed 3 years ago

JensWehner commented 3 years ago

Whereas fedora, opensuse and gentoo packages support up to L=6

So updating the ubuntu package would be nice.

junghans commented 3 years ago

@mbanck is that something we could add in the debian package?

junghans commented 3 years ago

ping @mbanck

mbanck commented 3 years ago

Do you have patch, or the configure options that are required?

JensWehner commented 3 years ago

We build our local version with --disable-shared --with-pic --enable-eri2 --enable-eri3 --with-max-am=6

junghans commented 3 years ago

Here are fedora’s configure options: https://src.fedoraproject.org/rpms/libint2/blob/master/f/generate-sources.sh#_45

mbanck commented 3 years ago

Hrm, so --enable-eri2 is the same as --enable-eri2=0 from looking at configure.ac, or what would it correspond to, seeing that Fedora seems to do --enable-eri2=2 (and lot of others)?

JensWehner commented 3 years ago

the number should indicate which derivatives are provided for that type of integral

mbanck commented 3 years ago

So what's the real-world benefit in going from 4 to 6, seeing how it blows up the build time (and likely library size) by a lot? What kind of computations/simulations will be possible that are currently not?

JensWehner commented 3 years ago

@mbanck So in quantum chemistry basisfunctions are labelled according to their angular momentum. Typically basisfunctions for light elements go up to g(l=4) functions, for heavier elements (3rd row+) higher functions are used. Also often auxiliary basissets are used which routinely go up to i(l=6). Also the derivatives of basisfunctions are used a lot for geometry optimisations of molecules. So the current debian/ubuntu config is basically only useful for unittesting and toy systems. The fedora config can be used in most production settings. The disadvantage obviously is, that the build times go up quite dramatically.

So I would argue the fedora config coves 90% of all usecases, whereas the debian config is like 20-30%.

Also see https://github.com/evaleev/libint/issues/190

Thank you for your help.

mbanck commented 3 years ago

I've uploaded libint2 with l=5 for now, let's see whether the autobuilders cope, I'll bump it to 6 if there's no issues.

mbanck commented 3 years ago

I've bumped it to l=6 now, and the autobuilders have been building it for 24 hours and counting.

JensWehner commented 3 years ago

I am starting to see your point.