qmcpack DMC densities have incorrect weight (CPU code)

Per the report in #925 , the weight in the first block of the DMC AoS code is incorrect. This is a definite bug and needs to be fixed in AoS and SoA. A similar fix may be needed in the CUDA code, but an additional bug is likely present there.

Jul 18 '18 20:07 prckent

I wonder if this is related to the remaining first step weights issue in #789. See the last "Trace of trial energy" plot there. The spikes right at the beginning of each DMC could relate to improper handling of the weights as is seen in the density estimator.

Aug 18 '18 10:08 jtkrogel

In regards to what might be special about the first DMC block, I noticed the following during some data analysis.

The number of walkers fluctuates little in the first block, and much more in other blocks: nwalkers_trace

The weight varies even less (almost none) than the walker count in the first block (look for the short horizontal line at the beginning of each series): weight_trace

The trial energy is absolutely bonkers for this block, likely reflecting the extreme population control needed to constrain the weights as above: trial_energy_trace

Conclusion: extreme population control in the first block to blame for the incorrect weights in that block? Thoughts @ye-luo?

Mar 12 '19 17:03 jtkrogel