One of the ways to produce embedding faster is to apply early filtering to remove events that won't pass analysis cuts. That way those events won't need to be processed via slowest parts of embedding (mixing, track reconstruction). It was proposed [https://drupal.star.bnl.gov/STAR/system/files/MZ190708_JMEmbedding.pdf] to remove some unneeded events by placing a tight cut on the 2to2 process pseudorapidities: -1.5 < \eta_{3,4} < 1.5.

This post will try to investigate the viability of this approach. First looking at the macro, it sets:

ckin (13)=[ETAmin]
ckin (14)=[ETAmax]

1. The eta distribution is not symmetric. This is because the cuts shouldn't be set using CKIN(13) and CKIN(14). Those are cuts on the values of max(\eta_3, \eta_4) [https://arxiv.org/abs/hep-ph/0603175]. The -1.5 < \eta_{3,4} < 1.5 cut can be implemented as:

   ckin (13)=[ETAmin]
   ckin (16)=[ETAmax]

2. Interesting thing: the CKIN(13) and CKIN(14) cuts do not have any effect in unbiased mode when CKIN(3) = 0, CKIN(4) = -1
3. I was able to reproduce the total nondifractive cross section for pt bin CKIN(3) = 2, CKIN(4) = 3. It is ~9mb for [-40,40] or [-10,10] and ~5mb for [-1.5,1.5]. This agrees with the values found in Maria's presentation.
4. Since the cross section for default [-40,40] eta cut and the [-10,10] cut doesn't differ, setting the latter gives no positive effect.
5. It is strange that we don't see these very drastic effects in the comparisons from the presentation.