2.2.0 relative to 2.0.6
=======================
- Parallelization using MPI. The diagonalization can now be spread
  over several calculation nodes. In combination with the SMP
  parallelization in LAPACK routines, even higher performance can now
  be achieved.
- Portability improvements (for example, the code now compiles under 
  Mac OS X without any tweaking).
- Calculation of the temperature-dependent conductance G(T), and 
  of the temperature-dependent moments (cf. specgt, speci1t, speci2t).
- Optimized routines for spectral function calculation, contributed
  by Markus Greger.
- Enhanced support for "global" operators. A number of such operators
  are defined for a number of different symmetry types. These can
  be used to calculate the temperature-dependent susceptibility
  for the general case (specchit).
- New symmetry types: SU(2)_iso x SU(2)_iso x U(1)_spin,
  SU(2)_iso x SU(2)_iso, U(1)_ch x U(1)_ch x U(1)_ch (for three-channel
  problems), U(1)_spin x Z_2
- Improvements in output from the code, diagnostics, etc.
- Calculation of an estimate of the Kondo temperature for the Kondo
  and Anderson models (TKW output line).
- Output of real and imaginary parts of spectral weights (reim).
- Extrapolation of thermodynamic calculations to high temperatures
  (tdht).
- Tools for smoothing of spectral functions (intavg) and for smoothing
  the thermodynamical quantities (tdavg).
- Adaptable broadening tool (bw), implementing the method introduced
  by A. Freyn and S. Florens.
- Old docs, tests and examples were removed. The documentation will
  be distributed separately.
- Some defaults for parameters have been changed, taking into account
  recent algorithmic developments and the accumulated experience with
  running the NRG calculations.
- Some bugs fixed (for instance, spectral function calculation
  for symtype=NONE).
- Huge library of automatic regression tests (distributed separately,
  big file!).
- "Decoupling tools" (matrix, diag, unitary in tools/dec). 
  It is now possible to run NRG Ljubljana without Mathematica by using 
  perl scripts to generate the input 'data' file from a template file 
  (generated using option=TEMPLATE).

2.2.1 relative to 2.2.0
=======================
- Output of complex expectation values (for non-Hermitian operators)

2.2.2 relative to 2.2.1
=======================
- Bug fix in CFS code
- smooth=newsc broadening scheme for problems with a superconducting gap
- Code cleanup, some fixes in the README file
- Schur decomposition is now the default way of diagonalizing the
  Hamiltonians. The old behavior can be restored using the NOSCHUR option.

2.2.3 relative to 2.2.2
=======================
- Bug fix in CFS code for complex problems
