STAR HardProbes PWG

[Ziyue Zhang <zzhan70 AT uic.edu>]

Hello conveners and HF enthusiasts,

I'll continue the discussion on the "bin-by-bin correlated systematic uncertainty" topic here.

In general there are three independent categories:

1) luminosity

Run09 and Run15 quoted 8% while Run12 used 8.1%.
However, if I trace back in the reference of Run09 and Run12 paper, they are pointing at the same paper. (also same as the dimuon paper's reference)
Therefore there's a good reason to believe this should just be a rounding difference. 

2) Trigger Bias Factor estimation

Run12: MB (bin 0,1,2) 7.04%, HT0 and HT2 (bin 3-14) 3.61%

Run15: HT2 (bin 8-13) 3%

Run09: lack of estimation in the original paper. In the combination, the algorithm follows this principle: Run09 will not influence this uncertainty entry when it is combined with Run12 and Run15's already combined result. Or in other words, it would be copied from the Run12 and Run15 combined result.

3) Embedding model difference

Run09: 7.5% (constant fit over pt)

Run15: Taken into account in another way (by pt dependent scaling correction, and the scaling factor's statistical uncertainty has made its way in the total statistic uncertainty)

Therefore here it should be 0

Run12: lack of estimation in the original paper; in Run15 one can estimate this number as 3.4%; therefore the average = 0.5*(7.5%+3.4%) = 5.4% is assigned to Run12 across the pT

What happens after combination:

1) luminosity: I can quote 8% in the caption. In terms of calculation, I can either use 8% everywhere, or use 8.1% for Run12 to preserve a little authenticity of the original paper, not a big deal.

2) TB estimation: see screenshot (in fraction of the mean, e.g. ipt=0 has ~7% uncertainty)

ipt 0-2: directly copy from Run12 since that's the only data

ipt 3-7 and 14: Run09 is copied from Run12 only, therefore the final result equals to Run12 estimation

ipt 8-13: "mixture" of Run12 and Run15

3) Model Difference: see another screenshot (in fraction of the mean)

ipt < 4 : just the average value of Run09 and Run15 estimation assigned to Run12

the rest: mixture of different runs therefore varies.

2+3) Quadrature sum of 2) and 3): varies among pt bins, no luck here

The luminosity contribution can be removed from the plot, while the rest should still somehow make its way onto the canvas.

However if we already have to put a third set of markers in, the luminosity contribution may also be included under it.

Any comment/experience is welcomed.

Best,

Ziyue

On Tue, Feb 27, 2024 at 1:40 PM Ma, Rongrong <marr AT bnl.gov> wrote:

Hello Ziyue

Thanks for your response. Sorry, I misread your slide. It is chi2/ndf ~ 3.5, not 3.5 sigma. 2.2 sigma, which really isn't rare, further supports approach-1, I think. 

Best

Rongrong

On Feb 27, 2024, at 2:15 PM, Ziyue Zhang <zzhan70 AT uic.edu> wrote:

To Rongrong

I don't think there's a specific reason to rule out this Run09 data point, other than if the chi2 test is not satisfactory.

Run09 has HT0 and HT3 data: HT0 2-10 GeV/c, HT3 7-14 GeV/c, with 7-10 GeV/c being the overlap.

For a good reason, the paper used the HT3 result for the overlap pt range.

In my combination, the problematic pt bin is 6-7 GeV/c, which is the last HT0 made into the final result, but in the middle of HT0 covered pt range.

I was under the impression that this was the first HT3 bin and maybe the trigger efficiency estimation was biased, but as you can see this is untrue. So no reason to drop it.

PS: I don't quite understand why you quote 3.5 sigma here. I'm naively thinking the number of sigma is the solution to Erf(x/sqrt(2))=1-p, where p is the p value from the chi2 test 0.029. (x ~ 2.2)