Molcas Forum

MHHsieh

Member

Registered: 2018-02-26

Posts: 10

Optimization using gradient from other level

Hi,

I am going to optimize my system with SA-CASSCF method combined with RICD keyword.
To my knowledge, molcas can only calculate the numerical gradient with SA-CASSCF and RICD in parallel. (i.e., the analytical gradient can be got only from serial)
but numerical gradient costs so high for me.

So I am wondering if I can calculate the gradient in another level, let's say KS-DFT or one-state CASSCF, then using this gradient to employ in every optimizing point?

Thanks a lot.

Best,
M.H.

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Optimization using gradient from other level

Well. In practice this amounts to a method-1//method-2 type of calculation, where you optimize the geometry at method-2 level and then compute single-point properties at method-1 level, since the optimization, for most purposes, cares only about the gradient and not the energy. This is certainly possible:

&GATEWAY

> Do While
&SEWARD
&SCF
&SLAPAF
> End Do

&RASSCF

You may also want to try adding constraints to your system, such that computing numerical gradients becomes an option. In that case you can also combine gradients from different methods, see https://doi.org/10.1002/jcc.23987

MHHsieh

Member

Registered: 2018-02-26

Posts: 10

Re: Optimization using gradient from other level

Oops!
You are right, it is equal to method 1//method 2 calculation.

Thanks for your suggestion in constrained optimization, and I will study your paper.

Best,
M.H.