I am not. At 1K the atoms are essentially “frozen” and if they are in a meta-stable state, they may remain there. With more kinetic energy in the systems, atoms can escape local potential minima and thus change conformation. As already suggested by @srtee, it is always a good idea to visualize your trajectory and have a close look.
As you are already suspecting, to make bonds rigid you do not need to also constrain angles. You only want to eliminate the fastest motion in the system, so you can use a larger timestep and that are the bond stretches involving hydrogen atoms (since hydrogen atoms are so much lighter than carbon). Angle bending in comparison is quite slow.
To better understand the error, you need to know that an angle constraint is internally converted to constraining three bonds that make a triangle formed by the constituent atoms of the angle. Now if that angle is close to 180 degrees, the triangle is very “flat” and thus it is no longer possible to come up with a suitable force constant for the two smaller bonds to “unflatten” the triangle, since those are nearly parallel to the third bond.
Ok thanks a lot. I will try again by avoiding to specify H-C-C angles types. I currently use OVITO to visualize the behavior of the molecule and I will do it now. Thanks srtee.
Hello. Following up on a statement that hydrogen bonding “has to be constrained in OPLS-AA forcefield”.
I cannot find this information in the corresponding paper. What ressources can I use to check for myself? Or is this just a standard when using a timestep of 1.0 and constraining H-X bonds due to the “large” timestep in comparison to that timescale?
Checking the original paper is a good idea, but note that OPLS was originally developed using monte-carlo simulations which do not need particular constraints with well chosen sets of movements since non-physical behaviour is excluded by design. You can check other papers using OPLS in MD such as this other paper. The constrained on the H bond is mentioned and the link with the relatively high integration step is made clear.
This is the numerical reason yes. Explicit covalent bonds with hydrogen having a higher frequency compared to other covalent bonds by at least an order of magnitude, most explicit models including them use constraint algorithms such as shake.
or you need to use some multi-timestepping method like r-RESPA to update the bond forces more often than the non-bonded forces.
Another option is to switch to a united atom force field where the hydrogen atoms at parameterized together with their carbon atoms as a single side. For OPLS there is the OPLS-AA (“all-atom”) variant and the OPLS-UA (“united atom”) variant.