Results with new neural network that incudes raster x and y input neurons. Have not yet incuded a z-position reconstruction using the tracker. This introduces a tracker geometry dependence and I still do not understand why I have to move the tracker close to get consistent results.

It is worth noting that the delta-Y target does not use the tracker. It only uses bigcal and a track-propagator, however, it seems to be able to see a positron peak. The red curv in the "positron_id2" plots is a cut on the tracker delta-Y while the blue is a cut on the target delta-Y (which does not use tracker).

The positron_id6 animatied gifs show the improvement in the tracker delta-Y when the detector is moved forward. The best position is somewhere around 49.5 cm. At this position the peaks in delta_Y tracker are aligned and the peak moves to the left with increasing energy.

In the tracker cut there remains a problem with rastered beam in that the spread at the tracker becomes large.

	Name	Last modified	Size	Description
	Parent Directory		-
	positron_id2_72994.png	2015-01-09 17:56	31K
	positron_id2_72994_5..>	2015-01-09 17:56	31K
	positron_id2_72999.png	2015-01-09 17:56	31K
	positron_id6_c2_7299..>	2015-01-09 17:56	882K
	positron_id6_c2_7299..>	2015-01-09 17:56	910K
	positron_id6_c2_7299..>	2015-01-09 17:56	943K

To do:

• Compare 72994 (carbon without raster) to another carbon run with raster
• Train NN with enlarged raster/target Z size.
• Train new NN with reconstructed Z.