This post covers work from 08/26 - 08/30

Fixing Titan Sims

From last week, I’ve figured out that 25 cells / micron is not a fine enough resolution due to the electron energy suddenly jumping in an unphysical way.

This is an example of what happens to the fields at around 300-400 fs we start seeing this dramatic increase in field energy

and electron energy.

I decided to change to 50 cells / micron. And in doing so, the ionization doesn’t jump all that quickly to max ionization.

Running at 50 cells/micron

I ran a sim for 1 ps and all looked well up to that point. I then continued the sim by restarting at that snapshot, and this is what the energy looks like

but from just looking at the charge density, I wouldn’t be able to tell that anything weird had happened

but from the fields, I can definitely see something weird happening at 1400fs

I will refer to this as the ball of death and it seems that everytime the energy goes haywire, I see this ball emerge from the front side of the target. Another thing to note is that if I check the slurm output files, the ETA remaining around 1380fs starts decreasing to increasing. So, this instability is inflating the time to train as well.

Maybe the weird blip at t = 1000fs is due to when I restarted it. But also, it looks likd there was some other numerical instability at t = 1400fs. So maybe, this resolution is still not high enough. I will try again with 100 cells/micron with 20 ppc and also 50 cells/micron with 50 ppc and see what takes less time and is more accurate.

Hyperparameter Optimization

I have been working on adjusting the hyperparameters for all the models used in paper 2. What interests me is that the polynomial regression prefers not to have a regularization term and alpha just as small as possible. Additionally, the polynomial regression seems to have a higher validation score than the other methods.

This is interesting to me, because we know that the polynomial regression ultimately doesn’t perform that well in the end. Also, I noticed that the total proton energy has a much higher error than the maximum energy. I think there is some division by zero stuff going on. Jack put in a value of 1e-2 to avoid this issue, but I’m not sure if this is the smartest choice. I’m going to take a deeper look at the fuchs function and decide what to do with these values that get wiped out to zero. For example, if the scales of the energies are in MeV, I feel like the lowest max and average proton energies should be around a keV or 1e-3, while the total proton energy should have a smallest value of much more than a MeV, because all the energies are being added up. Maybe the smarter thing is to get a heuristic on the lower bound and use that.

Target Figure

Here is the current draft of the target figure that I made which used the minimum intensity instead of the maximum intensity due to having a smaller effective thickness.