MODIFICATIONS  CODE ANTOINE/NOCORE         22-01-01
 
     In usual diagonalization, Diagonal Energy is precalculated and
     stored on a file  ( tmp/Zdiago ). For matrices of the NEXT generation
     this file can become very (too) large .
     It is no so evident that the computing time is shorter to read
     than to recalculate this vector .
 
***************
   CAS=13,14   Lanczos calculations
***************
 
13                    a1)
 40 42 0 0 0          a2)
 2 3    6 4 0 0
     90 1 10. 1     c)
   ........
   ........
 
  c)  FLNUC, COUL, ZCM, IDIAG
 
   IDIAG= 0    diagonal energy precalculated and stored .
   IDIAG= 1    we precalculate only pp, nn diagonal energies and 2-body
               decoupled pn matrix elements .
               Diagonal energy is recalculated at each Lanczos iteration.
    tests of time shows that in Be8 it is better to take IDIAG=1 but
    opposite in He8 .
    However the time is always negligible .(In the print of cpu time
    diagonal energy is the first one)
    conclusion : at the present time do what you want but when you
    will reach 10**9 , take IDIAG=1
 
        ****************************************************
 
 Get many states with the same spin is very difficult :
 The rounding errors produce a loss of rotational invariance.
Staes with different J (with low energy)  appears .
 In standart ANTOINE it is possible to project each new Lanczos
 iteration on J**2 (and T**2)but only for matrices with dimension not
 too large  ( all the Lanczos vectors for the J**2 operator must be
 both in ram memory).
    Possibility :
 
  start with a random pivot which has only a god time reversal symetry
   ( select sfor A even nuclei states with J even or odd) and do
   a lot of Lanczos iterations to get a maximum of converged or nearly
   converged states :
 
11
40 0 1
 4 5  4 4 0 0
 111111  1 0
 
13
41 40 0 0 0
 4 5  4 4 0 0
    91 1 10.   0
   1 20
 200  0.0005 1 0            a)
 
here we do 200 iterations to converge 20 states.
a)  we have taken ORTH=1 :  it means that each new Lanczos iteration
  is projected on good time reversal symetry .
  This is a NEW possiblity of the code .
  Notice that if a state is not fully converged it is possible to project it
  on J**2 and statr lanczos with this state but with orthogonaliqation
   at all the precedent converged states.
 
        ****************************************************
 
 REMEBER THAT IT IS POSSIBLE TO RESTART A LANCZOS CALCULATION
 
13
41 40 0 0 0
 4 5  4 4 0 0
    91 1 10.   0
   1 20
 200  0.0005 1 100          a)
 
  HERE WE HAVE NLANZ=100  MEANS THAT 100 ITERATIONS OF LANCZOS HAVE BEEN
  PRECEDENTLY DONE.