Molcas Forum

vkj

Member

Registered: 2019-04-10

Posts: 29

The SADREF keyword confusion

Dear developers,

I have confusions regarding keywords to use for analytic grad/NAC at caspt2.

1. I am not clear when to use the SADREF keyword for analytical gradients at CASPT2. Do I have always put it whether its SS-CASPT2 /(M,X,R)-multistate variants. I put the constraint that my underlying CASSCF/RASSCF is always state-averaged over multiple roots.

1. For the NAC computations, do I need just the NAC keyword and no SADREF/GRDT in all cases ?

EDITS to original question
Dear developers,

1. I have additional issues in using new analytical NAC. On the compiled master downloaded on 16 Feb 2023, the analytical NAC seems to only work inside the SLAPAF module. Outside it the program crashes complaining it can't do numerical NAC. On a related note, inside the SLAPAF module the program automagically switches to analytical grads without using SADREF/GRDT,   but in a single point computation these keywords are necessary to avoid NUMERICAL_GRADS.

2. In addition, in the ALASKA section, the  "Total derivative coupling" is identical to "CI derivative coupling". So, the division by energy-difference is not being performed. For same struct/basis-set the values are identical to what is obtained in BAGEL by using the "noweight" keyword which doesn't weigh the terms by energy-difference

VKJ

Last edited by vkj (2023-02-17 12:59:35)

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

Re: The SADREF keyword confusion

Bump.

I have an additional question about SADRef.

I tried it in a CASPT2 calculation on top of SA-RASSCF (RAS1+RAS2), with multistate = 1; 1
Got this error:

 ###    ERROR: Analytic gradients available only if all CASSCF               ###
 ###    roots are included in the CASPT2 calculation or with                 ###
 ###    the SADRef keyword. 

I had both SADRef and GRDT keywords specified. So, is it "Analytic gradients available only if all CASSCF  roots are included in the CASPT2 calculation AND with the SADRef keyword." maybe?

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

Re: The SADREF keyword confusion

I'd like to add to the SADRef question. I was not able to get it to work:
- with SA-RASSCF and single-state CASPT2, the error is shown above;
- with SA-RASSCF and multi-state CASPT2, ALASKA enters numerical gradient mode;

So, I assume, no analytic Hessian for MS-CASPT2?

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: The SADREF keyword confusion

From a communication with the developer: "SADREF has been used only for a development purpose in the initial implementation, but it is not well tested any more, so there may be some bugs. In addition, it is (strongly) discouraged to activate the option for actual studies in particular with XMS variants, because the character of these XMS variants are lost (RMS with SADREF is equivalent to XMS)."

So I think the bottom line is: Do not use SADREF.