Molcas Forum

cuplmu

Member

Registered: 2023-06-16

Posts: 3

Performance of XMS-CASPT2 MECI optimisation

Hi to everyone,
I understand that there is right now the need to call the CASPT2 module three times to optimize a MECI. Once for every derivative property (grad,grad,nac).
Each call computes the "normal" CASPT2-equations before moving to a second run computing additional terms for the computation of the Lagrangemultipliers. The "normal" CASPT2-equations should be needed only once.
A very crude inspection of the source code (currently using v23.06) and written temporary files showed, that the following MCLR module uses the additional files originating from the second run.

 PT2_Lag, GAMMA, A_PT2, $Project.caspt2.molden.#states $Project.Pt2Orb.#states 

Is it possible to compute all Lagrangemultipliers for all needed properties in one call of CASPT2? In my very amateur view this would need only small modifications in the code and would drastically speed up the optimisation of a MECI. This would just need a backing up of already computed files in the established fassion of "file#1#" or even better "file#grad1","file#grad2" or "file#nac12".
Since ALASKA or to be more precise MCLR doesn't check for the files beeing anything reasonable - eg. computing nacs from gradient lagrangians - anyway it would be in the users hand to move these files accordingly when using ALASKA and MCLR afterwards.
If I am completely off here, pleas let me know and I will just continue the intended way.

An Additional remark:
Since

 $Project.caspt2.molden.#states $Project.Pt2Orb.#states 

are presumably written in the second run - as they are different depending on which property you are calculating - this may cause errors in following modules, that expect the "true" Orbital files of the first "normal" CASPT2 run.