Is there an option to do "mixed-contracted" MRPT with some classes contracted, other uncontracted?

[bogdanoff]

I think there is always a way to compute only some classes for the contracted MRPT.
I wonder if there is also a way to run uncontracted MRPT calculations only for specific classes, that would require higher-order RDMs in contracted calculation?
Similar to how it was done for https://github.com/pyscf/icmpspt,

[hczhai]

I think if you use StackBlock as the fcisolver (rather than block2main), pyblock2.icmr.icnevpt2 will automatically switch to the algorithm using mixed ic/uc approach. See https://github.com/block-hczhai/block2-preview/blob/p0.5.4rc7/pyblock2/icmr/eri_helper.py#L178-L180 and https://github.com/block-hczhai/block2-preview/blob/p0.5.4rc7/pyblock2/icmr/icnevpt2.py#L283-L290.

[bogdanoff]

Oh, I didn't know one can use StackBlock as the fcisolver here. But is it also possible with block2? I mean just to do aaac and aaav classes uncontracted, not the whole nevpt2 or rept2?

[hczhai]

With block2main input file, you can use nevpt2-i / mrrept2-i (for aaac) and nevpt2-r / mrrept2-r (for aaav) as the (undocumented) uncontracted dynamic correlation keywords (see https://block2.readthedocs.io/en/latest/user/keywords.html#uncontracted-dynamic-correlation) to get the uncontracted energies for those classes. As these are experimental features, they are not interfaced with pyblock2.icmr.

[bogdanoff]

Thanks, I will give it a try!