Now the new finite-temperature Lanczos can be executed in parallel.
Moreover, I also implemented the smart symmetrization that prevents the storage of the full replica of the ensemble (they are computed on the fly).
Thanks to the sparseness of the symmetries in the polarization basis, the computation of the replica are very fast (scales as N, while the average calculation scales as N^2)
Now the new finite-temperature Lanczos can be executed in parallel. Moreover, I also implemented the smart symmetrization that prevents the storage of the full replica of the ensemble (they are computed on the fly). Thanks to the sparseness of the symmetries in the polarization basis, the computation of the replica are very fast (scales as N, while the average calculation scales as N^2)