Mismatch in the fraction of accepted translation move with two different Hoomd Versions

[hrithiknandy]

Dear Community,

I am trying to write a code to perform NVT simulation for a system of hard spheres (particle_number=2197) with packing fraction 60% using Hoomd-4.8.1. I performed the same thing with identical system using Hoomd-2.8.1. But now I am getting very different results in the count of fraction of accepted translation move when I am comparing it with the results from Hoomd2.8.1, although I used exactly same specifications like the maximum size of displacement trial move, seed etc.

The key difference between the results is the following, while in Hoomd-2.8.1 the fraction of accepted translation moves fluctuates around 27-28% in hoomd 4.8.1 this quantity keeps on increasing and could achive roughly 20% after the same simulation run.

I am attaching the code and the simulation log file(version-4.8.1). Can someone kindly point out the mistake?

Thanks
NVT.txt
output.txt

[joaander]

You should not expect software with different major revisions to behave the same: https://semver.org/

In this case, the relevant change is that HOOMD-blue >=3.0 (released more than 4 years ago) reports the total number of trial moves from the start of the call to run.

Without knowing how you generated those log files, I can only guess that your 4.8.1 results are plotting the average acceptance from step 0 to step T as a function of T. You should instead log the total number of translation moves and post-process that log to compute windowed averages of the acceptance with your preferred window width. For example: (total_accepted(T) - total_accepted(T-width)) / (total_moves(T) - total_moves(T-width)).

HOOMD 4.8.1 is already quite old. 5.0.1 is currently the latest version, and it includes a large number of new features and bug fixes compared to 4.8.1.