Greetings Michele Vanini.

I do not know if I have a misunderstand of your system. If you have a polymer and an amount of CO2 molecules under an NPT ensemble, then there's an expectation of this "mean" behavior of the molecules interacting around your polymer (I guess is some kind of membrane or surface). That's why there's a velocity distribution according to Maxwell function of probability. If you want to create a "flux" of molecules against your surface, then you have nonequilibrium molecular dynamics since you want to biased the behavior toward certain energy landscape region.

I am not an expert but what about using an umbrella sampling coupled with a nonequilibrium approximation such as pulling force among a coordinate reaction (steered molecular dynamics) to sample the free energy landscape between your CO2 and the polymer? You can get more thermodynamic insights about the interaction and there are some tutorials about using these techniques in GROMACS.

Thanks very much for you suggestion! I will try to look into detail at the umbrella sampling tutorial.

Best regards.

What are you trying to study ? If you have trying to do free energy calculations, then there are several options , like said above. But it is then unlikely to be several CO2 molecules. If you are trying to simulate a shock, then best way is to output the gro file of the system with velocities and then write a script to change vx values to vx+v, v being your desired velocity in x direction. Then use this gro file as the initial configuration file.