Optimization Algorithm

[ramess101]

@mrshirts @jpotoff @msoroush

I thought it would be helpful to include the recommended optimization algorithm, namely, starting with the TraPPE force field, obtaining pseudo-optimals for each lambda, simulating at these points, refining the optimization. In two simple small batches of simulations we go from TraPPE to MiPPE.

I put the algorithm at the end of Section 3.3 so that it is right before the Case Study. Here is what I have so far. Please provide feedback on how to make this more clear:

[ramess101]

@mrshirts @jpotoff @msoroush

I can think of one potential flaw in the stopping criterion. If the true optimal does not have 50 effective samples, it could get trapped in a different pseudo-optima. In practice, this shouldn't happen because the pseudo-optimal will be close enough that there should be at least 50 samples in neighboring parameter sets that are closer to the optimum.

[ramess101]

@mrshirts @jpotoff @msoroush

Another subtle issue is whether we are looking at the average number of effective snapshots (in the liquid phase) or the minimum number of effective snapshots. For example, this plot shows the average for liquid phase with our last iteration of lam = 16:

However, the lowest temperatures always have fewer snapshots, so if we plot the minimum of the liquid phase we get:

Although this plot has some anomalies, I feel like our stopping criterion should be based off the minimum Keff in the liquid phase. What do you all think?

[ramess101]

@mrshirts @jpotoff @msoroush

Here is a revised version that discussed using the minimum Keff as the stopping criterion:

[jpotoff]

@ramess101 The explanation of the algorithm looks good.