--------------------------------------------------------------------------------
 Fermonic PEPS using parity conserving tensors
--------------------------------------------------------------------------------

 - bsarray and parray classes
 - einsum 
 - create repo
 - general parray and einsum
 - initialization of peps,
   in particular merge & transpose of parray
 - operations: mps_dot & mpo_apply_to_mps 
 - peps_dot and peps contraction
 - autograd optimization
 - Hessenberg hamiltonian - fix bug for 3*3 case [fixed for Sp*Sm]
 - autograd tested against finite difference
 - save & load PEPS => callback
 - SVD-mps/mpo compression for larger systems
 - fix bugs in <A|H|B> 
 - Monte-Carlo - seems to be very suitable for long-range Hamiltoinan;
   we only needs to read out the coefficients <n|PEPS> for given |n>.
   Note: <n|H|PEPS> is required to avoid additional RI.
 - fermionic operations
   Starting from 2-by-2: 
    Q1: how operators are represented and act on PEPS ? 
    Q2: how <PEPS|H|PEPS> can be converted to bosonic TNS diagrams?
    Q3: particle number projector? any advantage 

--------------------------------------------------------------------------------
 TODOs
--------------------------------------------------------------------------------

 - memory issue? => larger systems

 - optimize the contraction scheme for <PEPS|H|PEPS>;
   reusing intermediates or using PEPO

 - regularize the PEPS for better numerical stability! 
   divide each by max element

 - eigenvalue optimization for small problem

--------------------------------------------------------------------------------
 To run it:
--------------------------------------------------------------------------------

 - Requirement: autograd
 
 - Download and change name for directory from fPEPS to fpeps2017.
   Go to fermions and run, e.g., peps2by2.py.