Since density-fitting is the default, it would make sense to compile libint to support it, not just for energies of the lighter elements, but for energies, gradients, and second derivatives across the entire periodic table.
Running a default gradient calculation on a metal-containing molecule produces the following error:
Basis Set: (6-31G* AUX)
Blend: CC-PVDZ-JKFIT + DEF2-QZVPP-JKFIT
Number of shells: 106
Number of basis function: 539
Number of Cartesian functions: 539
Spherical Harmonics?: false
Max angular momentum: 6
ERROR: ERI - Libint cannot handle angular momentum this high (6) for first derivatives.
Rebuild Libint with MAX_AM_ERI at least 7.
Is there some risk or other to compiling libint with a higher value of MAX_AM_ERI for the official releases? If not, I’d like to request that the official releases of psi4 have full support in libint for DF gradients and second derivatives for all supported elements.
Bumps the Libint download from 50 --> 260 MB (AM6 --> 8) is the short answer. And some ppl really care about download size. You’re right that this is an aggravating situation. My hoped-for compromise of offering higher-AM compilations fell through b/c (1) array-size-known-when issues mean you can’t swap Libint in a compiled Psi4 and (2) it’s hard to offer/select variations among packages in conda. Latter situation is improving. Former isn’t fixable.
Probably always doing AM8 is the widest acceptable situation. There’ll be a user survey posted here later today. Please consider filling it out and mentioning this block to your work.
Still haven’t seen the survey, but in any case I agree that supporting AM8 is preferable to a smaller libint download size. The number of people agonizing over a 260Mb download is surely less than the number of people agonizing about having to download the entire code base and learning how to compile it properly, just to get support for what is already a default method (i.e., density fitting).