kinks

[fewagner]

resolve kinks in likelihood limits

[fewagner]

looks much better since ed789a8 but not resolved yet

what we can try: take one of the problematic datasets and dm masses - try to reproduce the problem - check if its because of the best fit already (plot), or because of the exclusion fit

[fewagner]

kinks look like this
thats data/limit/limit_all_0.txt

[fewagner]

I did another run with brenth, and more accepted tries for increasing the search interval (V3). Here I also tracked the parameters, and figured out that the kinks are appearing when the global fit prefers a high DM component over high background components. I try to fix it with using a Nelder Mead in V4.

[fewagner]

resolved with dad688d

the problem was that the basinhopping optimizer sometimes prefers to push up the DM component instead of the background

using Nelder Mead everywhere fixed the issue