Does LINCS really allow unphysical dynamics that will fail in SHAKE?

User discussions
1

GROMACS version: 2023
GROMACS modification: No

Hello,

I came across this paper (On the Effect of the Various Assumptions and Approximations used in Molecular Simulations on the Properties of Bio-Molecular Systems: Overview and Perspective on Issues - PubMed) that warns of possible issues with LINCS allowing simulations to continue (with errors) while SHAKE would fail. The relevant text is:

“Thus, a SHAKE failure can be used to detect an error in the simulation, specifically the presence and location of unphysically large forces. Other algorithms, such as LINCS, use an overall convergence criterion that depends on the number of constraints. This means that LINCS is capable of recovering from situations where SHAKE would fail. However, because the error is not required to be evenly distributed across a set of constraints, certain constraints can be repeatedly violated. This will lead to unphysical dynamics. For long side chains such as lysine in large proteins this may lead to fast rotation of NH3 groups, reaching temperatures of thousands of Kelvin.”

Does anyone have any idea if this is in fact true? I don’t see citations for this fast NH3 rotation in this particular paper.

Also, could this be why LINCS warns on rotations that are too large?

Thank you,
Chris.

2

I suppose that theoretically this is possible. But if that would happen, than I would expect that the user gets LINCS warnings notifying that there is an accuracy issue.

No reference or evidence is given in the linked paper. We observed issues with fast rotating CH3 and NH3 groups in the early days of MD. But I think that was caused by incorrect cut-off treatments: no PME, abrupt cut-offs and a constant pair list of a number of MD steps. I don’t recall exactly which of these issues caused it. But since GROMACS has a proper Verlet pair list with buffering, I have never heard of issues with fast rotating CH3 and NH3 groups.

So I think this, certainly now, is a non issue. If you have fast rotating CH3/NH3 groups, there is an already a problem that causes this. And even if you would have such a case, LINCS would warn you about it (instead of failing like SHAKE).

3

Thank you so much for the info Berk! I really appreciate your time.