Molcas Forum

cymantren

Member

Registered: 2017-09-18

Posts: 37

Optimization improvements

Hello everybody,

I have two questions:

A)
I want to improve the results of an( RAS-)SCF optimization which did'nt converge.
How can one read in the last orbitals or what else possibilities are there, to "help" the system to converge ?!

What I already tried was:
start the calculation once again
>> COPY FORCE QuartetOrbitals.RasOrb INPORB
Use the former calculated geometry

B)
AND by the way, It is more convenient to perform an optimization as Zmatrix. (Manual p.235ff)
Is it possible to use variables e.g. ?

---------------

$GATEWAY(?!)

ZMAT
H          
H  d1     1
H  d2 a   1 2  

d1=..
d2=..
a=..


end of zmatrix

----------------

Thank you in advance !

Best Greetings

Alfred
P.S.:

**     MOLCAS.input generated by molcas_input.py Version 2.0

&GATEWAY
COORD
3

H1    -2.02637314     0.00000000     1.17120667
H2     2.76626035     0.00000000     1.16950401
H3     0.37134679     0.00000000    -2.98228168
GROUP = nosym
TITLE = Molcas-Opt
BASIS = cc-pVDZ


**     ================ Optimization ================

>> COPY FORCE QuartetOrbitals.RasOrb INPORB
>>> DO WHILE

&SEWARD

&RASSCF
SPIN   = 4
NACTEL = 3,0,0
INACT  = 0
RAS2   = 4
CIROOT = 1,1,1
LUMORB

&SLAPAF
>>> ENDDO

**     ================ Frequencies ================

&MCKINLEY

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Optimization improvements

You can just take the latest geometry (from the log file, or from the .geo.molden file, or I believe the .Opt.xyz will also hold the latest geometry) and the latest orbitals. The possible ways to "help" the optimization depend on the reason why it didn't converge, you could start by adding the "Cartesian" keyword in SLAPAF.

As for doing the optimization in z-matrix, you can specify the variables manually in SLAPAF, see https://molcas.gitlab.io/OpenMolcas/sph … onstraints, the example with HO+H2O is probably similar to what you want.

cymantren

Member

Registered: 2017-09-18

Posts: 37

Re: Optimization improvements

Hello Ignacio,

thank you for your reply !

Regarding to A)
I already have tried as you have suggested-use MOLCAS.opt.xyz etc. -, with NO further singnificant improvements. It is possible to read in the former calculated MO's ?!
An of course you are right, by pointing to the fact, that the program can only find a minimum, if it really exits. In the case of the H3-System I am not quite sure. The dublet system seems to react like a H+H2 System, after some optimizations. However the program didn't calculates the normal vibrations, because of the missing minimum.
E.G.
-------------

...
 50     -1.65251483 -0.00000004 0.000010  0.000006 nrc001   0.005870  nrc001       -1.65251485 RS-RFO  BFGS      0    

       +----------------------------------+----------------------------------+
       +    Cartesian Displacements       +    Gradient in internals         +
       +  Value      Threshold Converged? +  Value      Threshold Converged? +
 +-----+----------------------------------+----------------------------------+
 + RMS + 1.9657E-03  1.2000E-03     No    + 1.0147E-05  3.0000E-04     Yes   +
 +-----+----------------------------------+----------------------------------+
 + Max + 2.7799E-03  1.8000E-03     No    + 5.8710E-06  4.5000E-04     Yes   +
 +-----+----------------------------------+----------------------------------+

 Convergence not reached yet!

and later
---------

....
50     -1.65251598 -0.00000002 0.000007  0.000004 nrc001   0.003846  nrc001       -1.65251599 RS-RFO  BFGS      0    

       +----------------------------------+----------------------------------+
       +    Cartesian Displacements       +    Gradient in internals         +
       +  Value      Threshold Converged? +  Value      Threshold Converged? +
 +-----+----------------------------------+----------------------------------+
 + RMS + 1.2878E-03  1.2000E-03     No    + 6.6484E-06  3.0000E-04     Yes   +
 +-----+----------------------------------+----------------------------------+
 + Max + 1.8212E-03  1.8000E-03     No    + 3.8466E-06  4.5000E-04     Yes   +
 +-----+----------------------------------+----------------------------------+

 Convergence not reached yet!

.............
Finally:

...
50     -1.65251598 -0.00000002 0.000007  0.000004 nrc001   0.003846  nrc001       -1.65251599 RS-RFO  BFGS      0    

       +----------------------------------+----------------------------------+
       +    Cartesian Displacements       +    Gradient in internals         +
       +  Value      Threshold Converged? +  Value      Threshold Converged? +
 +-----+----------------------------------+----------------------------------+
 + RMS + 1.2878E-03  1.2000E-03     No    + 6.6484E-06  3.0000E-04     Yes   +
 +-----+----------------------------------+----------------------------------+
 + Max + 1.8212E-03  1.8000E-03     No    + 3.8465E-06  4.5000E-04     Yes   +
 +-----+----------------------------------+----------------------------------+

 Convergence not reached yet!

-----------------------
*********************************************************
 * Nuclear coordinates for the next iteration / Angstrom *
 *********************************************************
  ATOM              X               Y               Z     
  H1              -1.066042        0.000000       -0.598239
  H2               2.858125        0.000000        1.221583
  H3              -0.680850        0.000000       -1.264914
-----------------------

----------------------------------------------------------------------------

This happens after 3 optimization procedures and I don't see a way to improve the results further.
I'm also not sure if there exits conic-intersections,which I suppose,  but despite this fact, a minimum should be found ?!

To B)
Yes, your suggestion is similar to what I have in mind. I read in the manual, that it is impossible to use variables in the zmatrix -format directly, but numbers. This is however possible in the SLAPAF-module.

Best greetings

Alfred

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Optimization improvements

Just by taking the last .RasOrb file you are reading the previous MOs.

The optimization is converged as far as the gradient is concerned, it's only the displacement that's slightly above the threshold. However, the displacement in internal coordinates is very small, so this is probably due to an ill-behaved automatic definition of internal coordinates or a sub-optimal guessed Hessian. I'd consider the optimization converged. If you want to see it in the output, just specify "THRS = 1.0e-6 3.0e-4".

Maybe specifying the coordinates manually would help too. You cannot, indeed, use variables in the geometry input in GATEWAY, but in SLAPAF you can define which are the degrees of freedom you want to vary or fix, this is independent of how you write the initial geometry in GATEWAY. (And the degrees of freedom to vary need not be what you would use in a z-matrix, they could be e.g. the three H-H distances, or two angles and the average distance, or whatever three linearly independent combinations of angles and distances.)