[BUG] Relaxation of DNA structures behaves differently between recent oxDNA versions and the old one on oxServe

[ErikPoppleton]

Describe the bug
When pulling structures into desired starting configurations using mutual traps, the version of oxDNA running on the oxServe plugin of oxView is significantly more stable than the current version here (or this fork by @Jonahowns). I have not been able to successfully compile the old version used by oxServe for local testing (which is using code from this repository. I don't know the specific checkout, maybe @zoombya can double check?). This implies that one or more of the following is different between this older version of oxDNA and the current (post move to GitHub) implementations:

• Mutual traps or external forces in general
• FENE potentials
• The way max_backbone_force and max_backbone_far work

To Reproduce
pull_test.zip
The attached zip file contains a simulation where this behavior occurs. This structure is stable with the default oxServe input file, but the very similar input file in the example will explode within the first 10000 steps if run on the current oxDNA. The structure can be made stable by decreasing the dt 10x and the max_backbone_force 100x.

Expected behavior
It would be nice to know what changed between the two versions. It would also be nice if the older oxDNA_anm version of relaxation turns out to be physically reasonable and the current behavior could be updated to match because it is much more successful at producing starting structures for simulations.

Desktop (please complete the following information):

• OS: SUSE Linux Enterprise Server 15 SP4, MacOS 13.4.1
• Tested with the latest version of the oat_dev branch (using new topology) and the latest version of oxpy_anm. I haven't been able to get it to compile on the server, but I think the latest version of anm-oxdna is close to what oxServe is running.
• cmake/make/compiler versions - probably not relevant because the behavior is the same whether I use a GPU-compiled version on the server or the CPU-compiled one on my laptop.

[lorenzo-rovigatti]

I actually cannot reproduce this issue, since the latest oxDNA version and the one compiled from here give me very similar energies and similiar stretched configurations such as this one:

[lorenzo-rovigatti]

Erik, any news on this one?

[lorenzo-rovigatti]

@ErikPoppleton has this ever been understood, worked on or fixed?

[ErikPoppleton]

No, I haven't made any progress on understanding this one...

[zoombya]

in case we can allocate someone to this, we have to search for differences between the Nanobase ANM version and the recent github one.

[lorenzo-rovigatti]

I find this difference a bit unsettling, so let me know if you find some time to craft a minimal reproducible example that shows that there are striking differences between two oxDNA versions!