Firefly and PC GAMESS-related discussion club



Learn how to ask questions correctly


Re^4: Problematic CASSCF convergence with higher states

Thomas
thomaspijper@hotmail.com


Dear Alex,

Thank you for your suggestions.

I've included states that lie close to the states of interest in the state-averaging. Unfortunately, this did not improve the convergence. Again, the geometry optimization runs fine for a few steps, but then encounters a point where multiple states get remapped by the state-tracking code after which the SCF won't converge anymore.

An additional thing that I've tried is to decrease the values of TRMAX and DXMAX. The reason for this is that I've noticed that the first steps in the geometry optimization proceed with pretty big changes in the molecular geometry (especially for aromatic rings which in the first steps alternate between increasing and decreasing in size in a oscillatory fashion). In an attempt to "dampen" this effect, I decreased these values. However, this turned out to be of no help.

I've attached the input and output files to this message. Please let me know if you have any additional suggestions on how to achieve better convergence. If not, then I guess it is probably best to choose a different active space. The current active space is good for describing the reactivity, but a XMCQDPT2 calculation shows a big reordering of states -- I guess that it should be possible to choose the active space in such a way that (almost) no reordering takes place.

Again, thank you for your help.


Kind regards,
Thom




On Mon Mar 28 '11 3:08am, Alex Granovsky wrote
----------------------------------------------
>Dear Thom,

>Sorry for the delayed reply. Quite often, geometry optimization
>of higher roots of SA-CASSCF can be the real headache.
>While it is difficult to provide the 100% working recipe, I'd
>recommend you the following.

>First, starting with your SA-CASSCF orbitals for three lowest roots
>in averaging, perform a single-point SA-CASSCF calculation with
>wstate set to

>

WSTATE(1)=1,1,0,0,0,0,1,1

>and activated state tracking. If you are lucky, SA-CASSCF will
>converge to different orbitals better describing initial
>seventh and eighth states. These states will be lowered by
>energy and most likely the final order of states will be different.

>After getting new orbitals, perform CASCI calculations and find the
>states that are lower or close by energy to the states of interest.
>It is usually good idea to include them into averaging as well,
>especially if one is exploring different molecular geometries.
>Indeed, the lack of convergence you have encountered is most
>likely due to CASSCF state passing region of avoided crossing.
>Hence, you need to include the states involved into this crossing
>into SA-CASSCF procedure, all with equal weights. Ideally, new
>averaging would include all relevant SA-CASSCF states, with orbitals
>still allowing rather good description of the excited states of
>interests.

>Kind regards,
>Alex Granovsky
>
>
>On Wed Mar 23 '11 1:15am, Thomas wrote
>--------------------------------------
>>Dear Alex,

>>For the geometry optimization of the ground state structure, I used SA-CASSCF with averaging over the three lowest CASSCF states.

>>As for the follow-up XMCQDPT2 calculation, I tried to follow the guidelines on XMCQDPT2 that you've posted earlier.

>http://classic.chem.msu.su/cgi-bin/ceilidh.exe/gran/gamess/forum/?C35e9ea936bHW-7650-1219+00.htmhttp://classic.chem.msu.su/cgi-bin/ceilidh.exe/gran/gamess/forum/?C35e9ea936bHW-7650-1219+00.htm

>>I started with averaging (both for CASSCF as well as for XMCQDPT2) over the five lowest states. The settings that I ended up using in order to include all contributions larger than 0.1 are as follows:

>>

$MCSCF CISTEP=GUGA MAXIT=400 NTRACK=16 $END                                    
$DRT GROUP=C1 FORS=.T. NMCC=104 NDOC=5 NVAL=5 $END                             
$GUGDIA NSTATE=18 ITERMX=400 $END                                              
$GUGDM2 WSTATE(1)=1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,1 $END                         
                                                                               
$XMCQDPT NSTATE=20 EDSHFT=0.02 THRGEN=1D-9                                     
WSTATE(1)=1,1,1,0,0,1,1,1,1,0,0,0,0,0,0,1,-0                                   
AVECOE(1)=1,1,1,0,0,1,1,1,1,0,0,0,0,0,0,1,-0 $END

>>Thank you for your help.
>>
>>
>>Kind regards,
>>Thom
>>
>>
>>
>>On Tue Mar 22 '11 5:47pm, Alex Granovsky wrote
>>----------------------------------------------
>>>Dear Thom,

>>>how many states were included into your initial averaging?

>>>>

EIGENVECTORS OF THE EFFECTIVE HAMILTONIAN

                      1          2          3   

               *********************************

    1            -0.996647   0.000661  -0.014560
    2            -0.046349   0.056924   0.286798
    3            -0.050143  -0.079306   0.127224
    4             0.006949   0.001413   0.017246
    5            -0.012656   0.043812  -0.068093
    6             0.001191   0.175646  -0.044638
    7            -0.000448   0.932143   0.006876
    8            -0.003086   0.016572  -0.929936

>>>Kind regards,
>>>Alex Granovsky
>>>
>>>
>>>
>>>On Fri Mar 18 '11 2:02pm, Thomas wrote
>>>--------------------------------------
>>>>Dear fellow Fireflyers,

>>>>I'm currently studying a photochemical electrocyclization reaction, using the CASSCF and XMCQDPT methods. Geometry optimizations and a relaxed PES scan of the ground state have already successfully given me the geometries of the reactant and the product (as well as the ground state transition state), so Iím now focusing on the first and second excited states.

>>>>A XMCQDPT calculation of the reactant geometry shows a big reordering of the states: the second and third XMCQDPT state are mainly formed by the seventh and eight CASSCF state.
>>>>
>>>>
>>>>

EIGENVECTORS OF THE EFFECTIVE HAMILTONIAN

                      1          2          3   

               *********************************

    1            -0.996647   0.000661  -0.014560
    2            -0.046349   0.056924   0.286798
    3            -0.050143  -0.079306   0.127224
    4             0.006949   0.001413   0.017246
    5            -0.012656   0.043812  -0.068093
    6             0.001191   0.175646  -0.044638
    7            -0.000448   0.932143   0.006876
    8            -0.003086   0.016572  -0.929936

>>>>
>>>>
>>>>Consequently, I would like to investigate the PES of these roots. However, during geometry optimizations Iíve found the CASSCF convergence to be very problematic. There is a lot of reordering of states and after a 2-3 optimization steps the SCF will not converge anymore. These are the keywords that I use:
>>>>
>>>>
>>>>
 $CONTRL SCFTYP=MCSCF RUNTYP=OPTIMIZE INTTYP=HONDO UNITS=ANGS
 FSTINT=.T. GENCON=.T. $END 
 
 $SYSTEM TIMLIM=10000 MWORDS=100 $END
 $SMP CALL64=.T. $END
 $P2P P2P=.T. DLB=.T. $END

 $BASIS GBASIS=N21 NGAUSS=3 $END

 $MCSCF CISTEP=GUGA MAXIT=400 ISTATE=3 NTRACK=10 $END
 $DRT GROUP=C1 FORS=.T. NMCC=104 NDOC=5 NVAL=5 $END
 $GUGDIA NSTATE=11 ITERMX=400 $END
 $GUGDM2 WSTATE(1)=1,1,0,0,0,0,1,1 $END

 $STATPT METHOD=GDIIS OPTTOL=0.00004 NSTEP=500 HESS=GUESS
 NPRT=-2 HSSEND=.F. $end
 
 $GUESS GUESS=MOREAD NORB=114 $END
 
! Speeding up the calculation
 $TRANS MPTRAN=2 DIRTRF=.T. AOINTS=DIST ALTPAR=.T. MODE=112 $end
 
! Extra accuracy
 $CONTRL icut=11 $END
 $moorth nostf=1 nozero=1 syms=1 symden=1 symvec=1 $END
 $GUGDIA cvgtol=1d-7 $END
 $trans cuttrf=1d-13 $END
 $mcscf acurcy=1d-7 ENGTOL=1.0d-12 $END

>>>>
>>>>
>>>>My question is if anyone has any suggestions for achieving better convergence on higher CASSCF states.

>>>>Thanks in advance!
>>>>
>>>>
>>>>Kind regards,
>>>>Thom

>>>>


[ Previous ] [ Next ] [ Index ]           Mon Apr 4 '11 3:00pm
[ Reply ] [ Edit ] [ Delete ]           This message read 1075 times