Closed loriab closed 1 year ago
I'm not sure about the h2lpb case, but the nh3-lpb is for internal consistency, which is indeed broken because of the update, so that case is fine to fail. As I said over in psi4/psi4#2940, I'll take a look over the weekend.
Better discussed in https://github.com/psi4/psi4/pull/2952
Hi, over at Psi4, Michael kindly added the LPB feature and upgraded our interface from v0.1.3 to v0.3 (https://github.com/psi4/psi4/pull/2918/files). All our ddx tests pass nicely at v0.3.0. #140 here, though, looks like an important fix implemented for v0.4.1. When I run psi's suite against v0.4.1 (below), only the LPB tests fail (failure sizes further below).
Since the #140 fix involves a scaling factor, I'm wondering if the interface needs changing on the Psi4 side? Or perhaps just our reference values need changing? We're doing a release next week, so if you've any advice I'd appreciate it. Easiest thing on our side is to just pin to v0.3, but I'd hate to do that if v0.4.1 is right and our v0.3 ref values are wrong. We also could temporarily disable the LPB if this gets too tangled. Thanks for any advice!
assert 1.0019017425103895e-05 < 5e-06
where False = compare_values(-56.19598597466339, -56.198804366565064, 9, 'Total SCF energy with DDX versus reference data')