$GUGDIA group     (relevant for GUGA -CI- or -MCSCF-)                           
     This group provides control over the Davidson method                       
diagonalization step.                                                           
NSTATE = Number of CI states to be found, including the                         
         ground state. (default=1, ground state only.)                          
         You can solve for any number of states, but only                       
         100 can be saved for subsequent sections, such                         
         as state averaging.  See IROOT in $GUGDM/$GUGDM2.                      
PRTTOL = Printout tolerance for CI coefficients                                 
         (default = 0.05)                                                       
MXXPAN = Maximum no. of expansion basis vectors used                            
         before the expansion basis is truncated.                               
ITERMX = Maximum number of iterations (default=50)                              
CVGTOL = Convergence criterion for Davidson eigenvector                         
         routine.  This value is proportional to the                            
         accuracy of the coefficients of the eigenvector(s)                     
         found.  The energy accuracy is proportional to                         
         its square.  (default=1.0d-5, but 1E-6 if                              
         gradients, MPLEVL, CITYP, or FMO selected).                            
NWORD  = Number of words of fast memory to use in this                          
         step.  A value of zero results in the use of all                       
         available memory.  (default = 0)                                       
MAXHAM = specifies dimension of Hamiltonian to try to                           
         store in memory.  The default is to use all                            
         remaining memory to store this matrix in memory,                       
         if it fits, to reduce disk I/O to a minimum.                           
MAXDIA = maximum dimension of Hamiltonian to send to an                         
         incore diagonalization.  If the number of CSFs                         
         is bigger than MAXDIA, an iterative Davidson                           
         procedure is invoked.  Default=100                                     
NIMPRV = Maximum no. of eigenvectors to be improved every                       
         iteration. (default = nstate)                                          
NSELCT = Determines initial guess to eigenvectors.                              
         = 0 ->  Unit vectors corresponding to the NSTATE                       
                 lowest diagonal elements and any diagonal                      
                 elements within SELTHR of them. (default)                      
         < 0 ->  First abs(NSELCT) unit vectors.                                
         > 0 ->  use NSELCT unit vectors corresponding to                       
                  the NSELCT lowest diagonal elements.                          
SELTHR = Guess selection threshold when NSELCT=0.                               
NEXTRA = Number of extra expansion basis vectors to be                          
         included on the first iteration.  NEXTRA is                            
         decremented by one each iteration.  This may be                        
         useful in "capturing" vectors for higher states.                       
         On AXP processors, enter as 0 to avoid core dumps.                     
KPRINT = Print flag bit vector used when                                        
         NPFLG(4)=1 in $CIINP.       (default=8)                                
         value  1 bit 0 print final eigenvalues                                 
         value  2 bit 1 print final tolerances                                  
         value  4 bit 2 print eigenvalues and tolerances                        
                        at each truncation                                      
         value  8 bit 3 print eigenvalues every iteration                       
         value 16 bit 4 print tolerances every iteration                        
Inputs for a multireference Davidson correction, in case                        
the orbitals are from a MCSCF.                                                  
NREF   = number of CSFs in the MCSCF (full CI) job.                             
EREF   = the energy of the MCSCF reference.                                     

generated on 7/7/2017