$DIABAT group    (relevant if DIABAT=.TRUE. in $MCSCF)
 
This group controls creation of diabatic states from
adiabatic states using complete active space-type MCSCF
wavefunctions.  Diabatization is performed at a single
geometry, during RUNTYP=ENERGY, if DIABAT=.TRUE. in $MCSCF.
Diabatization is presently programmed only for CISTEP=GUGA,
and may be performed at either the the CASSCF level or at
the MCQDPT level.
 
Diabatization creates first the Diabatic Molecular Orbitals
(DMOs), and then the diabatic states by rotation of the
adiabatic CI states obtained with the DMOs.  DMOs are a set
of active orbitals evolving smoothly from a reference
geometry (where the adiabatic states are cleanly separated,
i.e. already diabatic in nature), through a region where
several states might come close, or even undergo avoided
crossings, and beyond to another region where adiabatic
states are once again cleanly separable.  Only one side is
considered the reference geometry for DMO generation.  The
dominant CSFs in the diabatic reference states are taken
from the adiabatic states at this reference geometry,
possibly informed by user knowledge about the dominant CSFs
at the other side of the crossing region.
 
If MPLEVL=2 is specified, the diabatization program will
also produce diabatic states at the MCQDPT level, utilizing
the CAS-level DMOs during the MCQDPT diabatization.  This
is a simpler procedure than was used prior to 2013.  Note
that $MCQDPT input need not be given, as DIABAT=.TRUE. will
force the selection of the CSF-based MCQDPT program, and
will pass orbital counts and state weights from this input
group to the MCQDPT.
 
The method is described in:
H.Nakamura, D.G.Truhlar J.Chem.Phys. 115, 10353-10372(2001)
H.Nakamura, D.G.Truhlar J.Chem.Phys. 117, 5576-5593(2002)
H.Nakamura, D.G.Truhlar J.Chem.Phys. 118, 6816-6829(2003)
Z.H.Li, R.Valero, D.G.Truhlar
  Theoret.Chem.Acc. 118, 9-24(2007)
K.R.Yang, X.Xu, D.G.Truhlar
  Chem.Phys.Lett. 573, 84-89(2013)
 
 
REFMOS = a flag controlling reading of "order reference
         orbitals" at a reference geometry, from $DFMVEC.
         These are often obtained by the 3-fold way, or
         perhaps the 4-fold way at a nearby geometry.
         Default is .FALSE.
         If not set, this run will use the 3-fold way to
         prepare DMOs, and will also punch the DMOs for
         possible use as the $DFMVEC for other geometries.
         If set, this run will use the order reference
         orbitals to help align the active orbitals of the
         run to their order at the reference geometry.
         This is normally a good idea!
 
REFGRP = a flag controlling reading of groups of CSFs
         expected to dominate different diabatic states.
         Default is .FALSE.
         If not set, for a run at some chosen reference
         geometry, the dominant groups are prepared and
         punched as a $REFCSF group.  See also SLCTTH,
         which sets the threshold for "dominant".
         If set, diabatization will be performed, using the
         dominant CSF group information read from $REFCSF.
 
Note: real diabatization runs must set REFGRP=.TRUE. but in
some cases diabatization might have no ambiguity in orbital
ordering, so REFMOS might be .TRUE. or .FALSE.
 
SLCTTH = selection threshold for dominant configurations,
         when REFGRP=.FALSE.  Also used as a printing
         threshold in adiabatic and diabatic states.
         Default is 0.20; it pertains to CI coefficients.
 
 
    * * * Keywords related to state selection * * *
 
The defaults are quite reasonable, so most runs might omit
all of these!  Let NWEIGHT be the number of states up to
and including the highest weighted state in the MCSCF
orbital optimization (according to WSTATE in $GUGDM2).
NWEIGHT includes any states with zero weight below the
highest weighted one.  The defaults use NWEIGHT states
during the DMO generation, all with equal weight in the DMO
generation, and then diabatize NWEIGHT states.
 
NGRST  = number of low lying states to be excluded from the
         final diabatization.  Default = 0.
 
NDIAST = number of states above the first NGRST which are
         included in the final diabatization.
         Default = NWEIGHT - NGRST.
 
NEXST  = number of excited states above the states being
         dealt with.  One might set NSTATE in $GUGDIA
         fairly high, to monitor the position of states to
         ensure they don't come close in energy to the
         interesting states, which stop at NWEIGHT.
         Such extra states are ignored during generation of
         DMOs and during diabatization.
         Default = NSTATE - NGRST - NDIAST.
 
WBLOCK = array of three (3) weights for the GR/DIA/EX
         blocks of states, used during DMO generation.
         The default is to give all NGRST and NDIAST states
         equal weight in the DMO process, with no weight
         for the NEXST states, namely
           NGRST/(NGRST+NDIAST), NDIAST/(NGRST+NDIAST), 0.0
 
Suppose the system of interest has one state lying at very
low energy, two excited states of interest that are close
to each other, and the user monitors three states higher
than these.  The MCSCF might very well average only the two
states that come close together,
 $GUGDIA NSTATE=6 $END
 $GUGDM2 WSTATE(1)=0,1,1,0,0,0 $END
The situation is thus NWEIGHT=3, so if no keywords are
chosen here, both DMO generation and diabatization involve
the first three states.  The most reasonable non-default
choice is NGRST=1, to omit the low-lying ground state from
the diabatization, but keep it during the DMO generation.
In case one also wishes to have the DMO step ignore the low
lying state, enter the keyword WBLOCK(1)=0.0,1.0,0.0 which
weights only the 2nd and 3rd states.
 
          * * * three-fold way parameters * * *
 
ALPHAN = weight of the state-averaged natural orbital term,
         default = 2.0
 
ALPHAR = weight of the state-specific occupation number
         term, default= 1.0
 
ALPHAT = weight of the transition density matrix term,
         default = 0.5
 
 
          * * * four-fold way parameters * * *
 
The four-fold way is used if NMLAP and/or NDLAP are given.
 
NDLAP  = number of "resolution orbitals".  Resolution DMOs
         are introduced to determine some of DMOs in the
         '2' (or DOC) block, as defined by MOSLAB below.
         Complex multi-arrangment reactions may use this
         resolution to avoid scrambling certain DMOs with
         others of smaller occupancy.  A typical use is for
         systems with nearly filled lone pair orbitals.
         If given, $DPSVEC orbitals must be given, and if
         ORIENT is chosen, also $LCLDC input.
         Default = 0.
 
NMLAP  = number of "reference orbitals" used by the
         maximum overlap reference MO (MORMO) criterion of
         the "4-fold way" method for DMO determination.
         Reference DMOs are obtained at some reference
         geometry, and advice about their selection can
         be gained from looking at successful applications
         in the literature.
         MORMO is used to the determine some of the DMOs in
         the '1' (or VAL) orbitals, see MOSLAB below.
         If given, $DIAVEC orbitals must be given, and if
         ORIENT is chosen, also $LCLVL input.
         Default = 0.
 
ORIENT = logical flag to rotate the "reference orbitals"
         and/or "resolution orbitals" from the reference
         geometry to the present coordinates.
         Default = .FALSE.
 
MOSLAB = array containing a character assessment of the
         active orbitals:
           DOC ('2') orbitals should have occupancies close
                     to two in all electronic states,
                     and be filled in all reference CSFs;
           VAL ('1') orbitals have variable occupancies
                     in different electronic state,
                     not necessarily close to 1.0, with
                     variable occupancy in reference CSFs;
           VIR ('0') orbitals should be only weakly
                     occupied in all electronic states,
                     and be empty in all reference CSFs.
         Only orbitals marked '2' are candidates for the
         "resolution orbital" step (see NDLAP), and only
         orbitals marked '1' are candidates for the 4-fold
         MORMO step (see NMLAP).  The default treats the
         entire active space as the VAL block:
             MOSLAB(1)=1,1,1,...,1,1
         Typically the number of '2' or '1' orbitals would
         exceed NDLAP and NMLAP inputs.
         See also THDOC and THVIR to let the program
         choose the 2's, 1's, 0's based on the active
         orbitals occupation number.
 
THDOC  = threshold on the state-averaged occupation numbers
         to identify DOC orbitals.  Default=1.8 electrons.
 
THVIR  = threshold on the state-averaged occupation numbers
         to identify VIR orbitals.  Default=0.2 electrons.
 
 
      * * * three-fold and four-fold iterations * * *
 
MAXIT  = number of D3 or MORMO Jacobi cycle iterations.
         The 3-fold way iterations seem very robust,
         but the MORMO 4-fold way iterations are less
         well convergent and more numerous (default=200)
 
CONVTH = threshold for convergence of D3 and MORMO.
         Default = 1.0E-6
 
 
                       * * *
 
DMOSYM = flag to allow lowering of symmetry during the DMO
         process.  Default = .TRUE., preserving symmetry.
 
Note: it is probably safer to enter GROUP=C1 in $DRT, and
NOSYM=1 in $CONTRL, and thus have symmetry off during the
entire run, than to choose this keyword.
 
 
                       * * *
 
Diabatization runs can read additional input groups which
are not well described here.  The $REFCSF and $DFMVEC input
groups are often given.  They are usually prepared by a
three-fold DIABAT=.TRUE. run at some reference geometry,
using both REFGRP=.FALSE. and REFMOS=.FALSE.
 
$REFCSF: list of dominant configurations at the reference
geometry, which is created by a REFGRP=.FALSE. run.  Note
that the SLCTTH threshold assists in deciding how many CSFs
are placed in the reference groups.  The $REFCSF data is
then read by all REFGRP=.TRUE. runs at the various other
geometries, in the same format as it is generated.
 
$DFMVEC: a set of temporary DMOs for ordering the DMOs at
the current geometry.  Usually these are DMOs prepared by a
3-fold way calculation at one reference geometry, but could
be from a 4-fold way calculation at a geometry very close
to the current one.  This contains only active MOS, namely
NDOC+NALP+NAOS+NBOS+NVAL from $DRT.
 
$DPSVEC contains NDLAP resolution DMOs.
 
$DIAVEC contains NMLAP reference DMOs.
 
Note:  $DFMVEC, $DPSVEC, $DFMVEC are typical $VEC type
inputs.  Each is read ignoring the MO index, so you might
prepare the order reference MOs from converged natural
orbitals by simply deleting all doubly occupied orbitals,
and keeping all active orbitals.  Similarly, reference or
resolution orbitals may be plucked from any desired orbital
set: natural, canonical, localized...
 
At present, there is no way for GAMESS to generate the
orientation information, although this can be read in.
Therefore their contents are not well described:
 
$LCLDC and $LCLVL are orientation data for DPSVEC and
DIAVEC, respectively, required if ORIENT=.TRUE.