Intermediate Matrices

[ekwan]

Hi! Is it possible to gain access to any of the intermediate matrices of the calculation like the density matrix, overlap matrix, etc.? Thanks.

[awvwgk]

Not in the current infrastructure. What are you planning to do?

[sarmaier]

Hi! I wondered if the infrastructure has been updated to print these matrices? Thanks!

[awvwgk]

It's an open source project, contributing a patch might make this feature available sooner.

[ekwan]

Well there are all kinds of interesting structural analyses that can be done with that information, which is why many programs have some way to print it. Would you be able to suggest a way to get at this information, even if it means digging into the internals?

[awvwgk]

main/property.F90 might be a good place to start (it is always called after the last singlepoint and reevaluates the integrals):

xtb/src/main/property.F90

Lines 135 to 145 in e5c5bef

	#ifdef XTB_GPU
	call sdqint_gpu(xtbData%nShell,xtbData%hamiltonian%angShell,mol%n,mol%at, &
	& basis%nbf,basis%nao,mol%xyz,trans,intcut, &
	& basis%caoshell,basis%saoshell,basis%nprim,basis%primcount, &
	& basis%alp,basis%cont,S,dpint,qpint)
	#else
	call sdqint(xtbData%nShell,xtbData%hamiltonian%angShell,mol%n,mol%at, &
	& basis%nbf,basis%nao,mol%xyz,intcut, &
	& basis%caoshell,basis%saoshell,basis%nprim,basis%primcount, &
	& basis%alp,basis%cont,S,dpint,qpint)
	#endif

wfx contains the density matrix (wfx%P) and the coefficients (wfx%C), as well as eigenvalues and occupation numbers and all the other important information. So there should be everything available you are looking for, you just have to decide on a format to print it. The ordering of the d orbitals is a bit strange (2, 0, -2, 1, -1; if I remember correctly).

[kvkarandashev]

Thank you very much! I think it does cover all my needs; should I have any questions or problems I will file them in tblite's "Issues" section.

[sarmaier]

Thanks for the awesome solution!

[awvwgk]

You're welcome.

The new library is still in early development, but I'm confident that we can put it to use in xtb soon as it is already used in DFTB+ and NWChem. Contributions to tblite are very welcome, the main library is well-documented, modern Fortran, but I'm also looking for help with improving the functionality of the Python bindings.

[linminhtoo]

hi, sorry to bump this after many months. is there ongoing work to merge this awesome contribution into xtb itself? (on top of having a separate project tblite) i would like to use xtb to print those intermediate matrices but i've never coded in Fortran so i'm not sure how difficult it is to manually add some print statements

[linminhtoo]

update: i managed to install tblite and get it working. it is a fantastic wrapper over xtb and i managed to output the intermediate matrices.

however i am also in need of the final fock matrix F which tblite doesn't seem to output in the results dictionary (created an Issue in that repo as well: tblite/tblite#101)

of course, there might also be a way to mathematically derive F based on the other matrices... but from my naive understanding we'd need additional info, like the energy eigenvalue matrix epsilon (from F.C = S.C.\epsilon)or the 2-electron electron repulsion integral tensor I.

any help on this would be highly highly appreciated! thank you