$INTGRL group                               (optional)
 
    This group controls AO integral formats.  Probably the
only values that should ever be selected are QFMM or
NINTIC, as the program picks sensible values otherwise.
 
 QFMM   = a flag to use the quantum fast multipole method
          for linear scaling Fock matrix builds.  This is
          available for RHF, UHF, and ROHF wavefunctions,
          and for DFT, but not with any other correlation
          treatment.  You must select DIRSCF=.TRUE. in
          $SCF if you use this option.  The RHF and closed
          shell DFT gradients also uses QFMM techniques.
          The Optimal Parameter FMM code will run at a
          comparable speed to a ordinary run doing all
          integrals for molecules about 15 Angstroms in
          size, and should run faster for 20 Angstroms or
          more.  See also the $FMM input.
          (default=.FALSE.)
 
 SCHWRZ = a flag to activate use of the Schwarz inequality
          to predetermine small integrals.  There is no
          loss of accuracy when choosing this option, and
          there are appreciable time savings for bigger
          molecules.  Default=.TRUE. for over 5 atoms, or
          for direct SCF, and is .FALSE. otherwise.
 
 NINTMX = Maximum no. of integrals in a record block.
          (default=15000 for J or P file, =10000 for PK)
 
 NINTIC = Controls storage of integrals in memory, with
          any remaining integrals will be stored on disk.
          Caution: memory set aside for this parameter is
          unavailable to the quantum chemistry methods.
          Positive NINTIC indicate the number of integrals,
          negative the amount of memory used for integrals
          and labels (in words).
          At present NINTIC works robustly for RHF, ROHF,
          or UHF, is thought to work for GVB or MCSCF and
          mostly works for sequential MP2 as well.  Direct
          SCF does not use this option!  (default=0).
 
 
   Various antiquated or antediluvian parameters follow:
 
 NOPK   = 0 PK integral option on, which is permissible
            for RHF, UHF, ROHF, GVB energy/gradient runs.
        = 1 PK option off (default for all jobs).
            Must be off for anything with a transformation.
 
 NORDER = 0 (default)
        = 1 Sort integrals into canonical order.  There
            is little point in selecting this option, as
            no part of GAMESS requires ordered integrals.
            See also NSQUAR through NOMEM.
 
 NSQUAR = 0 Sorted integrals will be in triangular
            canonical order (default)
        = 1 instead sort to square canonical order.
 NDAR   = Number of direct access logical records to be
          used for the integral sort (default=2000)
 LDAR   = Length of direct access records (site dependent)
 NBOXMX =  200   Maximum number of bins.
 NWORD  =    0   Memory to be used (default=all of it).
 NOMEM  =    0   If non-zero, force external sort.
 
   The following parameters control integral restarts.
      IST=JST=KST=LST=1   NREC=1    INTLOC=1
Values shown are defaults, and mean not restarting.
 
***
Following options are only valid
if you compiled GAMESS with OpenMP support
(`setenv GMS_OPENMP  true` in the install.info)
***
 
INTOMP = 0  Exchange and electron repulsion integrals
            (as well as their first derivatives) will
            be computed using legacy (MPI-only) code.
         1  (default) electron repulsion integrals will
            be computed using hybrid MPI/OpenMP code.
            J and K shell loops will be parallelized using
            OpenMP, whereas I loop will be parallelised
            using MPI. Use this scheme if the number of
            cores per node is greater than number of MPI
            ranks.
         2  electron repulsion integrals will
            be computed using hybrid MPI/OpenMP code.
            K and L shell loops will be parallelized using
            OpenMP, I and J shell loops - using MPI.
            Use this scheme if the number of cores per
            node is less than number of MPI ranks.
 
SHFOCK = .FALSE. (default) Fock matrix will be
                 replicated among OpenMP threads.
                 This algorithm is faster but consumes
                 a lot of memory on higly parallel
                 CPU architectures.
         .TRUE.  Fock matrix will be shared between
                 OpenMP threads. Can be only applied
                 if INTOMP=2.
 
  Take a note, that not all RUNTYPs are allowed with
  OpenMP two-electron integral code. In the case of
  unsupported RUNTYP the algorithm will fall back to
  pure MPI code.
  Poing group symmetry is also not supported in OpenMP
  code for now: the calculation will run as if C1 group
  is specified in $DATA.
 
==========================================================
 
 
==========================================================