CHANGES OF VERSION 5.1.6

1.  BUGS

2. INPUT RELATED CHANGES

3. DEFAULTS in input and throughout the code

4. OUTPUT related changes

5. MINOR or relatively insignificant changes

6. ALGORITHMS: additions or changes  to existing functional ability

1. BUGS

· fixed bug in inputread.f90. NLPP_rcut was bcast as integer. This didn't

show up in on the t3e as real and int have same memory size.

· Fixed a bug in the bandstructure output when no scf loop was

performed, and another bug in the naming of the data explorer '.dx'

files. - DJR

· Fixed a bug in paratec/input.pm wrt ordering of basisvectors

function. -DJR

2. INPUT related changes

· changed inpmat to read in symmetries in the same format that they

are output

· put in energy convergence_criterion into cgemin_gcg and cgemin_gcg_metal

can't put into cgemin as energy isn't calculated at every step

Also put into SC method. Convergence is achieved if criterion met

for 2 consecutive cycles.

· added keyword smearing_method to choose between methods.

also added smearing_energy as alternative to gaussian_smearing

· relax_method is new input for new atomic relaxation method

3. DEFAULTS in input and throughout the code

· changed default max_iter_diag to 3

· changed diagonalization_method default to Grassmann_metal for SC method

4. OUTPUT related changes

· added an option to not write CD and potenital to disk during SCFLOOP . This can

be expensive and was optionally removed for timing purposes. io_scf (default = false) is the

keyword to remove it. They will still be written out at convergence and for maxiter.

· made some refinements to the timing in SCFLOOP

· added option of plot wavefunctions into bandstructure job. -Seung-Hoon Jhi

5. MINOR or relatively insignificant changes

· made ews only allocated if NL_rspace in diag_ham_gcg,cgemin_metal, cgemin, and cgemin_gcg

· moved several routines to being optimized in src/para/Makefile

· added subroutine forcevectors into PW_LOG.pm donr by DJR

· changed cration of random numbers in randomize_startguess to be

done in a single array - Mickael Profeta

6. ALGORITHMS: additions or changes  to existing functional ability

· incorporate an initialization of random wavefunctions for the direct

minimization. no subtrix_diag is the keyword. The algorithm does some

non-selfconsistent minimization before updating the potential.

· altered flevel to do fermilevel calc. by bisection and added new

occupation schemes. Also moved calc. of energs%esmear from charge.f90 to

flevel. This was needed for FD smearing.

· reworked the convergence criteria for Grassmann and Grassmann_metal

method. A minimum of rho/# of bands = 1d-3 is required or a maximum of

5+max_iter_diag is done.

· direct Grassmann method ensures rho/# of bands = 1d-1 before updating

potential

· finalized optimize_metal method. Performs worse than SC method. Not

suggested.

· changed line minimization check for atomic relaxation to do refinement

if energy does not decrease on inital guess of lambda = one.

· added new method for relaxing stress and forces. It is a lbfgs method

It generally seems more stable and faster than the original BFGS method.

default of relax_fast when used. relax_method is new input for new method