fbischoff
commented
5 months ago Hi all,
the line in question I have just copy-pasted, without knowing what it actually does, esp. as the periodic code seemed disfunctional.
The idea behind the generalization of the operator is to handle all operators of the following forms:
Coulomb: 1/r12 BSH: exp(-r12)/r12 Slater: exp(-r12) F12: 1-exp(-r12) F12^2: (1-exp(-r12))^2 = 1-2 exp(-r12) + exp(-2 r12) FG: F12 *Coulomb
by using the same expansion exponents (they depend on eps only, but not on the exponents of the operator), and add their coefficients (they depend on the actual form of the operator). That way all these operators are simply represented as a single operator.
So if I understand correctly, the parameter that breaks your code either does nothing or truncates the expansion, based on the assumption of decreasing exponents of the expansion. That should still be the case with the exception of F12 (and related operators), where I represent the “1” as a single term in the operator expansion with a tiny exponent: see gfit.h:660
Hope this helps.
Best, Florian
Florian Bischoff Humboldt-Universitaet zu Berlin, Institut fuer Chemie Unter den Linden 6, 10099 Berlin phone: +49-30-2093-82708 office: Brook-Taylor-Str 2, room 3'322 @.*** https://tinyurl.com/bischofflab
On May 22, 2024, at 2:42 PM, Jonathon Misiewicz @.***> wrote:
@evaleev and I are working on periodic code, and I find that values of the electric potential changed after rebasing onto a more recent version of MADNESS. The current results appear to have an additional "contribution" that is pure garbage (e.g., not even monotonic as you get away from the center of the sum-of-Gaussian charge distribution). The problem is this line of code added during #498. The CoulombOperator required the last argument to truncate_periodic_expansion to be true, not false, prior to the PR. Manually changing that argument reproduces our electric potential prior to rebasing onto #498, but doing that would break many other operators. @fbischoff, I noticed that you generalized some operators in operator.h to use your new function but left others the same. Is there an underlying reason? I want to make sure that my fixes to the CoulombOperator don't interfere with anything you were trying to do. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
JonathonMisiewicz
gifornl
@evaleev and I are working on periodic code, and I find that values of the electric potential changed after rebasing onto a more recent version of MADNESS. The current results appear to have an additional "contribution" that is pure garbage (e.g., not even monotonic as you get away from the center of the sum-of-Gaussian charge distribution).
The problem is this line of code added during #498. The
CoulombOperatorrequired the last argument totruncate_periodic_expansionto betrue, notfalse, prior to the PR. Manually changing that argument reproduces our electric potential prior to rebasing onto #498, but doing that would break many other operators.@fbischoff, I noticed that you generalized some operators in
operator.hto use your new function but left others the same. Is there an underlying reason? I want to make sure that my fixes to theCoulombOperatordon't interfere with anything you were trying to do.