Instability of omega kernel for large molecules

[ivlyubim]

When calculating omega for large molecule, results become corrupted especially at low wavevectors .
The problem appears only when number of beads per CPU is much greater than 50, and the error grows with number of frames to average over. For small molecules (less than 50 beads) or when running the code on many CPUs, this numerical error seems to be negligible for trajectories up to 10000 frames.

[martintb]

I am fairly sure this is due to numerical overflow issues in the code, although I’ve never seen it happen for varying number of cpus. I’ve run into similar issues doing calculations for very large simulations. Try pulling down and compiling the dev branch of the repository and see if the issues go away.

In the dev branch, I've implemented a simple type system where users can select the level of precision or integer size they need for a given executable. The tradeoff is that the higher precision and larger integer types generally slow down the overall calculation.

See https://github.com/martintb/correlate/blob/dev/src/core/types.hpp

[ivlyubim]

Compiling dev branch did solve the issue. Thanks!
I also don't see the slow down of the calculations. Instead, I see 2 to 5 times faster performance when using dev version of the code compared to main. Calculated with dev code omegas don't depend on number of cpus, as it should be for purely deterministic calculation. This is not the case when omegas are calculated with the main version of the code.