Firefly and PC GAMESS-related discussion club


 
Learn how to ask questions correctly  
 
 
We are NATO-free zone
 



Re: MCSCF does not converge with high precision

Alex Granovsky
gran@classic.chem.msu.su


Dear Dawid,

sorry for delayed reply.

I'd suggest you the following modifications of your original input:


1. 6-31G* is not spherical basis set. Set D5 option to .false.


2. Use $MP2 cutoff=1d-20 $END


3. Use $TRANS cuttrf=1.0d-20 $END

4. Relax acurcy in MCSCF to 1.0d-8 or maybe even to 3.0d-8.
acurcy=1d-9 is too tight and often cannot be reached due to finite
precision of floating point operations, so no convergence in MCSCF.

Hope this helps.

Kind regards,
Alex Granovsky




On Fri Jan 6 '17 11:56am, Dawid wrote
-------------------------------------
>Dear All,

>I would like to do XMC-QDPT2 geometry optimization of the ground-state
>using single-state CASSCF reference.

>The issue I am encountering is that the MCSCF calculations (in course
>of the geometry optimization) do not converge when high precision is
>enforced even though the output looks like this:
> ITER    TOTAL ENERGY          DEL(E)    LAG.ASYMM.  SQCDF  MICIT   DAMP
>     ----------START APPROXIMATE SECOND ORDER MCSCF----------
>   1    -439.972125989    -439.972125989  0.000560 2.088E-06  1   0.0000
>   2    -439.972130903      -0.000004914  0.000203 1.353E-06  1   0.0000
>   3    -439.972131634      -0.000000731  0.000060 3.849E-07  1   0.0000
>   4    -439.972131722      -0.000000088  0.000021 3.551E-07  1   0.0000
>   5    -439.972131755      -0.000000032  0.000009 4.861E-08  1   0.0000
>   6    -439.972131760      -0.000000005  0.000008 4.466E-08  1   0.0000
>   7    -439.972131763      -0.000000003  0.000004 1.206E-08  1   0.0000
>   8    -439.972131764      -0.000000001  0.000002 5.590E-09  1   0.0000
>   9    -439.972131764       0.000000000  0.000001 7.682E-10  1   0.0000
>  10    -439.972131764       0.000000000  0.000001 1.915E-10  1   0.0000
>  11    -439.972131764       0.000000000  0.000000 7.585E-11  1   0.0000
>.
>.
>.
>  95    -439.972131764       0.000000000  0.000000 1.789E-18  1   0.0000
>  96    -439.972131764       0.000000000  0.000000 3.993E-18  1   0.0000
>  97    -439.972131764       0.000000000  0.000000 1.440E-18  1   0.0000
>  98    -439.972131764       0.000000000  0.000000 6.709E-19  1   0.0000
>  99    -439.972131764       0.000000000  0.000000 5.794E-11  1   0.0000
> 100    -439.972131764       0.000000000  0.000011 5.792E-11  1   0.0000
> EXCESSIVE NUMBER OF ITERATIONS...

>Nevertheless couple of geometry searches are performed until

>.
>.
>.

>  82    -439.920263839       0.002669544  0.161276 5.195E-01  1   0.0000
>       SOSCF IS SCALING ROTATION ANGLE MATRIX, SQCDF=   84.966198
>       SOSCF IS SCALING ROTATION ANGLE MATRIX, DXMAX=    2.229427
>  83    -439.917588524       0.002675315  0.169614 5.195E-01  1   0.0000
>       SOSCF IS SCALING ROTATION ANGLE MATRIX, SQCDF=    3.359397
>       SOSCF IS SCALING ROTATION ANGLE MATRIX, DXMAX=    2.229428
>  84    -439.914768467       0.002820057  0.174227 5.195E-01  1   0.0000
> SOSCF ENCOUNTERS A SERIOUS PROBLEM IN -SONEWT-
> THE ROTATION ANGLE VECTOR HAS A HUGE NORM, SQCDF=   7.058E+03
> REEXAMINE STARTING VECTORS, APPROPRIATENESS OF YOUR SCFTYP, ETC...
>
>
>On the other hand, if I use default convergence criteria:

> ITER    TOTAL ENERGY          DEL(E)    LAG.ASYMM.  SQCDF  MICIT   DAMP
>     ----------START APPROXIMATE SECOND ORDER MCSCF----------
>   1    -439.971787958    -439.971787958  0.000714 5.267E-06  1   0.0000
>   2    -439.971803845      -0.000015887  0.000195 1.130E-06  1   0.0000
>   3    -439.971805101      -0.000001256  0.000103 1.072E-07  1   0.0000
>   4    -439.971805179      -0.000000078  0.000037 4.774E-08  1   0.0000
>   5    -439.971805197      -0.000000018  0.000013 8.471E-09  1   0.0000
>   6    -439.971805199      -0.000000002  0.000003 8.942E-10  1   0.0000

>the MCSCF calculations converge correctly.

>The active space is fine through the calculations (its size and
>orbitals that create it).

>So what is the possible reason for that issue? I attach input
>file for calculations that do not converge correctly.

>


[ Previous ] [ Next ] [ Index ]           Tue Jan 10 '17 4:51pm
[ Reply ] [ Edit ] [ Delete ]           This message read 484 times