cj-wright / pyiid Goto Github PK

#Python Infinite Improbability Drive

Is designed for the Monte Carlo modeling of nanomaterials using atomic pair distributuion functions, other experimental data, and ab-initio structural calculations.

Areas that need improvement:

1. GPU based gradient and PDF potential energy functions: otherwise everything is very slow.
   1. Gradient: particularly expensive on single threaded systems, which makes Hamiltonian Monte Carlo glacially slow
   2. Generating PDFs: currently using diffpy.srreal for PDF generation, which is not GPU optimized
   3. PDF potential energy: this is quantified using the RW value which could also be put onto the GPU
2. Ab-initio calculation support:while most of this is handled by ASE issues include
   1. Scaling and Units for PDF comparison: we need a way to effectively tune the relationship between the PDF and the ab-initio, otherwise one will dominate the refinement and dynamics
   2. Support for non-ASE calculators