d3q Code

Added little trik to stop deleting my own project

Restart of tk improved, now it works as before but it is more consistent, also CG tk creates a diferent output file for each iteration

CG tk with random grid seems to be not variationalm, this would require some investigation, but I think it is the usual detailed balance theorem that is not respected, so I'm disabling it for the moment

Example updated

Examples updated

Example updated

Removed twfcollect and lkpoint_dir

New checkpoint because I rolled back to the wrong version

Making a check point before rolling back to previous version, because there is something not working properly

Also note that everything can be done 2x faster, but using much more RAM if we use the fact that A_in is symmetric (implemented here, but not in parallel)

Looks like my brilliant idea was not that brillians, fixing previous commit, now two matrices are used again (but not for SMA) also looks stable but I have to test

New form for variational tk (as in PRB paper) seems to work much better than the previous one (that was from the thermalk code), it is also twice as fast as the same matrix can be reused, but some more testing is needed. In particular, I have some doubt about the index to use in freqm1