memory in SCPH/phonon

[caperott]

Dear Dr. Tadano,
I hope this find you well. If I have interpreted this correctly, the minimum amount of memory required to allocate the 3-dimensional array used for diagonalizing the dynamical matrices in SCPH/phonon ANPHON calculations is on the order of 16 x nq x (3N)^4, where nq is the number of k-points and N the number of atoms in the primitive cell. The amount of memory thus grows very rapidly with the number of atoms in the primitive cell. Is there any way we may reduce the amount of memory needed in these calculations?

[ttadano]

Yes, the SCPH calculation becomes more memory-intensive as the number of atoms in the unit cell increases. The current implementation does not utilize distributed memory parallelization (MPI); therefore, implementing it will likely improve memory efficiency.

[caperott]

Thanks!