 So if you just follow these instructions you throw a protein in the computer, apply periodic boundary conditions, hit the play button. In all likelihood the computer will scream loud, there is going to be something strange and the computer will stop simulating your protein. What has happened? Well that's because real proteins are not perfect. If you pick a molecule from the protein data bank, in all likelihood there will be at least one place where two atoms are overlapping a bit, a bond might be a bit too long or too short compared to the ideal values we have in our force field. Maybe two atoms are too far away from each other, so they wouldn't be attracting each other a bit and having a strong force initially. You also have to add water. My algorithms are not perfect. Maybe I happen to place water oxygen too close to another oxygen so they will repeal each other or maybe I put the water too far from the system so it will actually end up attracting the water quite strongly. But the simulation would fix that right? So here's now my protein in water and you hit the play button and I have a large force and then you try to integrate that and this happens because the force is so large that in one pic of second they would move 10 nanometers. This is probably even larger than your box but what's worse now we have two parts of your molecule and they're likely both overlapping with water. That would make each of these molecules undergo the similar type of explosion because now they're on top of water molecules and everything you can probably guess. It's all going to be downhill from here. It will keep screwing up the entire system. There will be overlap everywhere. The temperature will be 5 million Kelvin before you think about it and the computer will give up. All this happened because you started with a system that was very very far away from equilibrium. Don't do that. So how do we not do that? It's going to be a very simple method called energy minimization. In principle I'm not really interested in finding a minimum of the energy but I need to avoid these nasty peaks or in particular the very strong derivatives here right? Because it's going to be the derivatives that are the slopes that are the forces and if I am at a very steep slope I'm going to have a very large force and that's what I need to avoid but the algorithms that do that in practice is energy minimization. The most common one we're going to use is actually called steepest descent and what that means is that if my atom here starts I calculate the gradient of the potential which is the same as the force and then I check what is the direction in which I'm going downhill quickly and then I take a step in that direction for every single atom of the 100,000 or so in the system and then I repeat the process. Now that I've gone slightly downhill I look around again and see what direction is most downhill right now and I take another step and another step and another step and within a few hundred steps already we're going to be in a region that starts to get close to a local minimum. Not a minimum, local minimum, not global minimum but I don't care because my point here was not to find the global minimum my point was to avoid very large derivatives in the forces. Let me show you that for a real protein but the motions here are going to be much smaller than you think which is also an indication that there are tons of local minima here so when I hit the button when I come to three you're going to start the minimization. One, two, three and there we go super exciting. You might be able to see some atoms moving that blue sheet popped a little bit there it was mostly the program choosing to draw it in a different way the atoms didn't really move that much. Just to give you a feeling for this I'm going to jump back to the starting state now and then you can see it again. So the complete energy minimization here did not even move the atoms more than a tiny fraction of an angstrom. We do that at the start of each simulation not to get an energy minimum but again to get rid of the bad derivatives. So if there's one take home is we need to call this energy minimization otherwise nobody will know what you're talking about but it's really energy maximum avoidance or really high force gradient avoidance that's going to be difficult to say and nobody else will have any idea what you're talking about so energy minimization it is. You need to do that at the start of a simulation or things will go bad.