Remove built-in ADC module

[maxscheurer]

Description

This PR removes Psi's internal ADC module, so that all ADC calculations are run through adcc from now on.
Closes #1033.

User API & Changelog headlines

• remove built-in ADC module (does not remove any ADC features), closing Problems with ADC(2) in Psi4? #1033.

Dev notes & details

• remove built-in ADC code
• update proc.py
• update docs

Questions

• Question1

Checklist

• Tests added for any new features
• All or relevant fraction of full tests run

Status

• Ready for review
• Ready for merge

[loriab]

For methods (or methods in certain circumstances, say reference or conv/df) only available through an external add-on, do we want those opt-in? That is, certainly the external must be (1) installed and detectable. But do we also want to (2) require the user to set qc_module=mrcc|adcc|chemps2 ? CheMPS2 has a long history of not requiring (2). ADCC has a shorter history of being the preferred backend and automatic choice, if present. I just switched MRCC syntax in #2731 to yes require (2). That was in keeping with the user opt-ing in via energy("mrccsd").

I can go either way, and I guess I'm now leaning toward not requiring (2) and adjusting MRCC accordingly. But it seems like something to discuss and settle on a consistent treatment.

[maxscheurer]

I think in the case of ADC, it makes little sense to require an explicit qc_module keyword as of now. In my opinion, it would just create another hurdle for users, not having an actual choice in that case. I cannot say much about a policy for the other modules though 😄

[maxscheurer]

@loriab I removed the MEMORY option from the ADC section, but now some tests using fnocc are failing, because apparently, the option was only provided by the ADC part in read_options.cc... What should I do about this? adcc does not take any input on available memory, so where should the MEMORY option live from now on?

[loriab]

I removed the MEMORY option from the ADC section, but now some tests using fnocc are failing, because apparently, the option was only provided by the ADC part in read_options.cc... What should I do about this? adcc does not take any input on available memory, so where should the MEMORY option live from now on?

My best guess at why fnocc is using MEMORY is so that it can get more memory under default conditions (1gb instead of 0.5gb). Better handling would probably be to have a FNOCC_TRIPLES_FACTOR to manipulate input memory. I think comment out those MEMORY blocks in triples and lowmemory_triples, and we'll put up an issue. Those aren't running by default, and fnocc doesn't own MEMORY (as you discovered).

[JonathonMisiewicz]

Please mark this PR as closing #1033.

[JonathonMisiewicz]

Will adc be moved to the attic?

[maxscheurer]

Will adc be moved to the attic?

I can do that, sure 👍🏻

[JonathonMisiewicz]

Hooray!

Remember to move the ADC code (including the equation PDF!) to the attic. @loriab will need to file that fnocc issue.

[hokru]

lgtm