[manuscript] partycls performance
Closed this issue · 2 comments
By testing partycls with my own heterogeneous polymeric system here within Notebook 2 and without any graphical backend, a simple gr analysis took extensive amounts of time (in a i7 equiped laptop). BTW, the configuration I shared is a melt of cold and hot particles, have you experienced a similar behaviour?
Moreover, could you add something about performance within the Ms.? (perhaps as a new section or at the end of the examples section)
Thank you for attaching your configuration.
This is a large system (1e6 particles), so I'm not surprised that it took a while. Currently, the code works well with small- and medium-sized systems (~1e4). We ran the code on systems of 20.000 particles with no difficulty in the context of our own research, but it's true that the current implementation of the native descriptors is not well suited for very large systems (especially gr, because it needs to compute correlations at large distances). We are considering several kinds of optimizations for future versions, such as the subdivision of the system into multiple neighboring cells.
A word in the manuscript (and the README) about the system size seems like a good idea, anyway.
In the end, we simply added a note in the README (see commit 581b544). Given that this system size limitation will disappear soon with future optimizations, we thought it would be best not to leave a trace of it in the manuscript. 🙂