Molcas Forum

lwang

Member

Registered: 2024-10-14

Posts: 1

Adenine in aqueous solvents RASSCF optimization does not converge

I want to optimize the ground state structure of adenine in the (12,12) active space, and the input is as follows:

&GATEWAY
coord =15
Angstrom
 C                 -1.29197700    1.69850700    0.00001200
 C                 -1.23021700   -0.61050500   -0.00001000
 H                 -1.92702000    2.58482100    0.00002000
 N                 -1.87053900   -1.79444900   -0.00006200
 H                 -1.34926700   -2.65734200    0.00018300
 H                 -2.87877100   -1.81900300    0.00020500
 N                 -1.94742100    0.52572600    0.00000500
 N                  0.01910000    1.92043400    0.00000100
 C                  2.28671300   -0.76978900    0.00001100
 C                  0.71228800    0.76994900   -0.00000600
 H                  3.29021200   -1.17985200    0.00001400
 C                  0.17875000   -0.51801700   -0.00001700
 N                  1.17904200   -1.47226300    0.00000900
 N                  2.07010400    0.58416300   -0.00000500
 H                  2.77950500    1.30522800   -0.00000400
basis =Aug-cc-pvdz 
group = C1
  RF-input
     PCM-model; solvent= water
  End of RF-input
>>> EXPORT MOLCAS_MAXITER = 500
>>> DO WHILE
&SEWARD
&RASSCF
   Charge = 0
   spin=1
   nActEl=12 0 0
   Ras2=12
   ciroot =5 5 1
   RFRoot=1
   Rlxroot=1
 &mclr
    iter =500
 &SLAPAF
 >>> END DO

The results show that there is no convergence, and in this example, it is not possible to set the number of turns to 500 when calculating the SCF electron integral (i.e. MCLR)

          494      3.963908937E+20  3.766697051E+09  8.224176458E+08  3.415873239E+18  3.850967073E+17
          495      3.943035238E+20  3.756766363E+09  8.202493874E+08  3.397885462E+18  3.830688119E+17
          496      3.917879840E+20  3.744763661E+09  8.176287268E+08  3.376207959E+18  3.806249462E+17
          497      3.878408321E+20  3.725852196E+09  8.134996127E+08  3.342193629E+18  3.767902587E+17
          498      3.853053108E+20  3.713653290E+09  8.108361131E+08  3.320343936E+18  3.743269809E+17
          499      3.812283381E+20  3.693953682E+09  8.065349163E+08  3.285210884E+18  3.703661715E+17
      No convergence for perturbation no:    1. Increase Iter.

# Convergence problem #.

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

Re: Adenine in aqueous solvents RASSCF optimization does not converge

You are basically doing CASSCF.
From my (rather very limited) experience CASSCF is generally bad: it does all possible excitations from a very limited set of orbitals.
I.e. some orbitals get overly important.

A more physical approach, and correct me if anyone knows better, is to use fewer excitations from larger amount of orbitals.
Basically, pick 2 or 3 occupied orbitals and 2 or 3 empty orbitals for RAS2, add some 10 orbitals for RAS1 with up to 2 holes, and maybe a few more empty orbitals (RAS3) with up to 2 electrons. I.e. :

RAS1=10
RAS2=6
RAS3=4
nactel = 26 2 2

Or, just keep any amount of empty orbitals in RAS2, RAS3=0, and use any amount of orbitals as RAS1, but allow no more than 2 holes - should be good enough.

RAS1=X
RAS2=Y
RAS3=0
nactel = 2*X 2 0
(this is a scheme, Molcas does not do math in keywords like that)

The above is just wild wild random guess on my side, every case is different.

With that, try different sets of orbitals and see how fast a single RASSCF converges. Long convergence (100-200 steps) might mean the space is suboptimal. Not a "scientific" metric, but I do not see any better.

After you've had enough of playing around with different spaces - use pegamoid (my personal favorite code for orbitals ever) and rasscf.h5 files and see what actually ends up in the active space, make sure that is what you need. And then do the optimization.

I guess this is why people call multiconfigurational calculations "hard" and "requiring experience".