griffin-group / darkmagic Goto Github PK

Reusing the same MaterialParameters object for multiple calculations mutates it, causing an assertion error.

Traceback (most recent call last):
  File "/pscratch/sd/a/ash/He/hcp/pres_sweep/calc_darkmagic.py", line 44, in <module>
    material = PhononMaterial(material_name, params, yaml_file)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/global/u1/a/ash/code/DarkMAGIC/darkmagic/material.py", line 271, in __init__
    properties.validate_for_phonons(n_atoms)
  File "/global/u1/a/ash/code/DarkMAGIC/darkmagic/material.py", line 90, in validate_for_phonons
    assert any(np.any(v) for v in d.values())

The current rate attributes of the Calculator class are named and designed to maintain backward compatibility with PhonoDark. Implementing an attribute where the binned rate is a (velocity, mass, mode, bin) array should be easy. Should also implement an easy way to distinguish Umklapp processes.

A little bit of this was done in 47fb949, but the code needs cleanup and generalization.

My first thought was to use @jitclass, which I did for the numerics module in a4e2e21. However, this approach would be a long and painful road since every single class would have to be a @jitclass since they are all used in Calculator, and thus every single method in the entire package will have to be @njit compatible. This is simply not practical (nor desired, nor necessary).

The best option is a heavy refactoring of the code so that I'm computing the binned rate as a (n_velocities, n_masses, n_modes, n_bins) array, i.e., all the Single[...]Reach classes need to be heavily reworks so that I'm not looping over them and calculating the rate for each (v_e, m_chi) pair individually. Those classes should do all the setup, then the entire rate calculation can be taken outside the classes into an njitted method that does nothing but array operations.

pymatgen is a very large package and we only use a minute fraction of its functionality (a few features of the Structure class), which can easily be extracted into its own class.

https://Griffin-Group.github.io/DarkMAGIC/ should redirect to https://Griffin-Group.github.io/DarkMAGIC/latest but it 404s instead. Also changing versions in the docs seems to be bugged, it redirects to the home page instead of the same page.

this likely has a fix.