Dear Tini Chu

I am not an expert in this, so take my explanation as 'my two cents' on this question, also because sometimes people use constraint/restraint interchangeably.

What you said about the protein being restrained is correct. Restraints are the general term used when you specify, for example, that you are applying an energy penalty to some type of movement. Then, in the protein sim, you can restrain the heavy atoms of the protein so that most of the position changes are experienced by water atoms/ions/hydrogen atoms, and you don't risk any weird non-physiological unfolding. The utility is basically to go from a starting configuration that can contain high forces weird packing etc to an equilibrated one, and want to go there without losing any geometrical properties while you remove these 'generation' artifacts. The same is done for example with lipid bilayers, where you restrain lipids in their position so that the bilayer does not lose integrity during the equilibrium phases.

In general restraints can also be applied for other reasons, such as energy sampling (umbrella sampling, AWH etc) or to drive some changes such as dragging an ion across a channel. Basically what you do is integrating the equations of motion by considering this additional energy penalty, which for example could take the form of a harmonic potential centered in the position of interest so that the atom is 'trapped' where you want it to be.

Constraints usually do refer to some type of constraint that is put on the bonds, such as setting a bond length or an angle among some atoms in a molecule. You constrain a bond so that it does not oscillate. Oscillation of bonds can be very fast, and therefore would need a small integration time step to be properly picked up and described. If your oscillations have a typical period of, say, 0.1 fs, and your dt is 1fs, then you wouldn't be able to properly describe what is going on in the system. In a way, you can see bond constraints as a way to remove these oscillations, and make larger integration steps achievable. For example

https://manual.gromacs.org/current/reference-manual/special/remove-fast-dgf.html

What you heard about NVT constraints could be true, depending on the context, but it is not a general rule. Bond constraints are a serious thing, and strongly depend on the integration time step and the forcefield you are using, and therefore you must follow what is suggested for that forcefield, depending also on what you are trying to achieve and investigate. Differently from restraints, a constraint is usually applied using some specific algorithms (LINCS, SHAKE, etc) that corrects the position of the atoms after the dynamical step has been taken, that is after an unconstrained step. The implementation is quite different from constraints and it could also depend on the specific algorithm and code you are using for MD (LAMMPS, CHARMM, GROMACS, etc).

Hope this helps,

Nicola

Dear Nicola Piasentin,

Thank you so so much for spending time answering!! I also learnt that constraints were for limiting motions, so I was confused at the moment. Now it's clear and it makes sense for me. I really appreciate your help! By the way, may I know if there's any recommended learning materials on constraints/ restraints from you?

Regards

Haseeb Javed

Dear Tini Chu

I don't know much material explicitly written for learning only about this specific topic. I use to read online as much as I can on forums like these, plus I really recommend reading the manual on the software that you are using to run MD. Best option is probably to look into books about molecular modelling, such as Molecular Modelling by Leach. A must are also the original papers reporting the method you are using, such as

https://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=4ACE851098F73C4323B916B4D73F7055?doi=10.1.1.48.2727&rep=rep1&type=pdf

Best,

Nicola

Thank you for the support!!