STAR Flow, Chirality and Vorticity PWG

[jagbir <jagbir AT rcf.rhic.bnl.gov>]

Dear Fuqiang,

   Please find my replies below:-

1.   I understand your motivation doing that but I don't agree this is 
the right approach
   (I think it causes biases).

   Please let me know why this is not the right approach and what kind  
of biases you
   meant.


2.   you have only one point left. Your f_cme is basically the
     (Delta gamma of those events in that Dbmax bin)
     - (Delta gamma of a random collection of events in the same 
centrality bin which happen
     to have the same Dbmax_shuffle bin)
     - (Delta gamma of those same random events calculated after the 
charges are shuffled)
     Do I understand it correctly?

   If only one point is left as you wrote, please see the explanation 
below:

   Here, there is nothing like random collection of events in the same 
centrality bin.
   We select events depending on Db+-max i.e., depending on the 
back-to-back charge
   separation fDbCS = Db+-max-1. As you wrote only one point, in that 
case it is the
   top 10% Db+-max events corresponding to maximum back-to-back charge 
separation events.
   It is similar to selecting events in particular collision centrality 
depending on
   the impact parameter or event multiplicity. As we have selected in 
the data top 10%
   Db+-max corresponding to maximum back-to-back charge separation 
events, in  same way
   we select events from charge reshuffle in the same collision 
centrality for top 10%
   Db+-max(here Db+-max of charge reshuffle) corresponding to maximum 
back-to-back
   separation. Now we get delta_gamma_data of real data events 
corresponding to top
   10% Db+-max of data and delta_gamma_sta of charge reshuffle events 
corresponding to top
   10% Db+-max of charge reshuffle for a given collision centrality 
which gives
   us delta_gamma due to statistical fluctuations. Now for correlated 
background we
   look for real events in the data corresponding to the top 10% Db+-max 
 of charge
   reshuffle in a given centrality and get delta_gamma_cor from those 
real
   data events.

   Now the f_CME is obtained as

   f_CME = N1*(delta_gamma_data - delta_gamma_sta - 
delta_gamma_cor)/(delta_gamma_data * N)

   Where N1 is number of events in top 10% Db+-max and
   N is total number of events in a given collision centrality.

Thank you,
With regards,
Jagbir Singh

On 2020-12-14 10:32, Wang, Fuqiang wrote:
> Jagbir,
>
> Thanks for your answers.
>
> To your questions:
> > Please clarify the following
> >       Of course you've also removed the large charge-shuffle 
> > background which is basically
> >       an autocorrelation effect (sort to speak) due to the Dbmax (and 
> > Dbmax_shuffle)
> >       selection bias.
> I was just saying it in passing, referring to the fact that you're
> largely selecting on statistical fluctuations and trying to remove the
> auto-correlation effect by shuffling. It wasn't a question.
>
> >       Please explain the following:
> >       So is your finite signal really due to the difference between 
> > the average of ratios
> >       and the ratio of averages (or perhaps also due to residual 
> > effect from shuffling)?
> I was referring to the fact that if you had a single Dbmax bin (i.e.
> taking average first and then ratio) then you'd get zero signal by
> definition. You now have 10 bins and take ratios first in each bin and
> then take average of the ratios, and get a positive signal. During the
> focus meeting discussion, it was made clear that your analysis
> required multiple Dbmax bins, not taking average of all bins, but only
> those with Delta gamma > 0. So now I think I understand technically
> how you did it. I understand your motivation doing that but I don't
> agree this is the right approach (I think it causes biases).
> So let me try to understand better:
> On slide 9 of your focus meeting presentation
> https://drupal.star.bnl.gov/STAR/system/files/CME_FOCUS.pdf, you
> state:
> (1) If Delta gamma_bkg. is negative then it is taken as zero.
> (2) If gamma_SS is not negative and gamma_OS is not positive then
> delta gamma = 0.
> Now to slide 25, let's take one centrality say 40-50%, you have the
> blue points (signal) and red+green points (bkg). The 8 points to the
> right of this centrality: all of them have negative bkg and negative
> Delta gamma, so they are not counted in your calculation of CME
> fraction. Now you're left with the two leftmost points. Do both points
> satisfy (2) above? I know both points seem to have bkg>0 & Delta
> gamma>0 but it's unclear if they satisfy SS<0 & OS>0. Assume they do,
> then you're taking average of these two data points. For the sake of
> simplicity, let me say you have only one point left. Your f_cme is
> basically the
> (Delta gamma of those events in that Dbmax bin)
> - (Delta gamma of a random collection of events in the same centrality
> bin which happen to have the same Dbmax_shuffle bin)
> - (Delta gamma of those same random events calculated after the
> charges are shuffled)
> Do I understand it correctly?
>
> Best regards,
> Fuqiang
>
>
>
> > -----Original Message-----
> > From: jagbir <jagbir AT rcf.rhic.bnl.gov>
> > Sent: Sunday, December 13, 2020 10:12 AM
> > To: Wang, Fuqiang <fqwang AT purdue.edu>
> > Cc: STAR Flow, Chirality and Vorticity PWG <star-fcv-l AT lists.bnl.gov>; 
> > star-cme-
> > focusgroup-l AT lists.bnl.gov; aggarwal AT pu.ac.in
> > Subject: Re: [Star-fcv-l] FCV PWG meeting, 18/Nov/2020 (Wed) 9:30am 
> > (New
> > York time zone)
> >
> > Dear Fuqiang, all,
> >
> >    Sorry for not answering your email. Infact, I did not look this
> >    email. Please go through my replies below:
> > ------------------------
> >
> >   1. A few events do not satisfy this cut so not including in it 
> > Db+-max
> >      but in overall calculations all events are included.
> >
> >   2. Yes
> >
> >   3. We reshuffle charges in each event. We donot randomize charges
> >      according to the positive/negative charge ratio of the given 
> > event.
> >      In fact, we pick up one event and reshuffle positive/negative 
> > charges
> >      keeping theta,  phi, number of postive charges and number of 
> > negative
> >      charges as such. After this we calculate gamma correlator. This 
> > procedure
> >      is repeated for each event. The Db+-max of reshuffle  is a bit 
> > bit wider
> >      than the  real distribution which may be due to some correlations 
> > in the
> >      real data whereas reshuffle is purely randomize. Db+-max binning 
> > is
> >      done on the basis of same fractions.
> >
> >   4. Let me explain this point
> >
> >      We pick up a real data event and calculate following
> >      i)  Dbmax+- of real data event
> >      ii) reshuffle charges in an event
> >
> >      iii) again calculate Dbmax+- and termed it Db+-max of charge 
> > reshuffle
> >      iv)  calculate gamma of real data event
> >      v)   calculate gamma of reshuffle event
> >
> >      Now for a given centrality
> >      Steps i) to v) repeated for each event. Db+-max (data) and
> > Db+-max(reshuffle)
> >      sliced into ten percentile bins.
> >      Now average gamma is found in every sliced Db+-max (data) and
> > Db+-max(reshuffle)
> >      from the respective event samples. It should be noted that events 
> > in the top
> >      say 10% Db+-max(data) are not the same as in the top 10%
> > Db+-max(reshuffle) i.e,
> >      real events in the top 10% Db+-max(data) are different from those 
> > top 10%
> >      Db+-max(reshuffle). Now the correlated background is caculated 
> > from the
> > real events
> >      corresponding to the top 10% Db+-max(reshuffle) events.
> >
> >       Please clarify the following
> >
> >       Of course you've also removed the large charge-shuffle 
> > background which is
> > basically
> >       an autocorrelation effect (sort to speak) due to the Dbmax (and
> > Dbmax_shuffle)
> >       selection bias.
> >
> >   5.  Db+-max distribution is sliced in to ten percentile bins which 
> > represent
> >       different amount of charge separation in each sliced db+-max 
> > bin.
> > Let us
> >       say we have Db+-max = 2, in this case fractional dumbbell charge 
> > separation
> >       f_DbCS = Db+-max-1=1 i.e., 100% back-to-back charge separation 
> > i.e.,
> >       positive charged particles on one side of the dumbbell and 
> > negative charge
> >       particles on other side of the dumbbell. So, computing gamma in 
> > different
> >       Db+-max and calculating things is different from just making a 
> > single wide
> >       bin as you mentioned. This method is designed to get  CME-like 
> > enriched
> > sample
> >       in given collision centrality as one divides all events into 
> > different collision
> >       centralities depending on either the impact parameter or event 
> > multiplicity
> > but
> >       one does not study all events taken together without making 
> > different
> >       collision centrality classes. However, for a single wide Db+-max 
> > bin as you
> >       wrote we will get zero signal.
> >
> >       Please explain the following:
> >
> >       So is your finite signal really due to the difference between 
> > the average of
> > ratios
> >       and the ratio of averages (or perhaps also due to residual 
> > effect from
> > shuffling)?
> >
> >
> > Thank you,
> >
> > with regards,
> > Jagbir Singh
> >
> >
> >
> > On 2020-11-18 23:18, Wang, Fuqiang wrote:
> > > Hi Jagbir,
> > >
> > > Your results are quite interesting. I have a few further questions
> > > about the details of your analysis:
> > > 1. For each event you have Dbmax with the condition of |Dbasy|<0.25.
> > > You bin events of each centrality in Dbmax. You use all events in your
> > > analysis (i.e. you're not throwing away events based on Dbmax or
> > > Dbasy), right?
> > > 2. In your calculation of gamma=<...>/v2c for a particular Dbmax bin
> > > of a given centrality, the v2c is calculated using those events only,
> > > right?
> > > 3. For the charge reshuffle, you reshuffle the charges of all events,
> > > and repeat your analysis from step 1 (i.e. you treat this as a
> > > completely separate "new" data sample), right? Did you "randomize" the
> > > charges according to the positive/negative charge ratio of the given
> > > event? On s11, the Dbmax_shuffle distribution is a bit wider than the
> > > real distribution, do you understand why? How do you bin the Dbmax and
> > > Dbmax_shuffle into 10 bins, respectively (same bin edges or same
> > > fractions)?
> > > 4. Your correlated background gamma is calculated for the Dbmax bin
> > > where Dbmax is from the charge-shuffled events, but using restored
> > > charges, right? If so, then you're effectively taking gamma difference
> > > between Dbmax_i events and Dbmax_shuffle_i events (which are different
> > > events), right? Of course you've also removed the large charge-shuffle
> > > background which is basically an autocorrelation effect (sort to
> > > speak) due to the Dbmax (and Dbmax_shuffle) selection bias.
> > > 5. You divide Dbmax (and Dbmax_shuffle) into 10 bins and do your
> > > analysis in each bin separately, and then take the weighted average
> > > for your f_cme result. You could just use a single wide Dbmax (and
> > > Dbmax_shuffle) bin, then in principle you should get zero signal
> > > because the correlated "background" is your real signal since they are
> > > now identical event sample. So is your finite signal really due to the
> > > difference between the average of ratios and the ratio of averages (or
> > > perhaps also due to residual effect from shuffling)?
> > >
> > > This is a complicated analysis. It would be really good to have more
> > > discussions so the details can flesh out better.
> > >
> > > Thanks,
> > > Fuqiang
> > >
> > >
> > >
> > >> -----Original Message-----
> > >> From: Star-fcv-l <star-fcv-l-bounces AT lists.bnl.gov> On Behalf Of
> > >> jagbir via Star- fcv-l
> > >> Sent: Tuesday, November 17, 2020 10:49 AM
> > >> To: ShinIchi Esumi <esumi.shinichi.gn AT u.tsukuba.ac.jp>; STAR Flow,
> > >> Chirality and Vorticity PWG <star-fcv-l AT lists.bnl.gov>
> > >> Subject: Re: [Star-fcv-l] FCV PWG meeting, 18/Nov/2020 (Wed) 9:30am
> > >> (New York time zone)
> > >>
> > >> Dear ShinIchi, Prithwish and Jiangyong,
> > >>
> > >> I would like to give "Update on event by event charge separation in
> > >> Au+Au collisions at 200GeV with STAR detector"
> > >>
> > >> Please add me to agenda.
> > >> I will post my slides later.
> > >>
> > >> Thankyou,
> > >> Jagbir Singh
> > >>
> > >> On 2020-11-16 15:57, ShinIchi Esumi via Star-fcv-l wrote:
> > >> > Dear FCV PWG colleagues
> > >> > We will have our weekly FCV PWG meeting on coming Wednesday
> > >> > 18/Nov/2020
> > >> > 9:30AM (in BNL) at our usual time and place. So if you have
> > >> > anything to present, please let us know and please post your slide by
> > Tuesday.
> > >> > We'll talk about the "HLT express productions" in the beginning of
> > >> > the meeting as you see in the agenda page. Jiangyong, please send a
> > >> > link to your slide from last week.
> > >> >
> > >> > The zoom room link, ID and password are in our usual drupal agenda
> > >> > page below.
> > >> > Please also keep in mind that all the preliminary plots should have
> > >> > already been there in the summary area below.
> > >> > Best regards, Jiangyong, Prithwish and ShinIchi
> > >> >
> > >> > Meeting agenda page with zoom link :
> > >> > https://drupal.star.bnl.gov/STAR/blog/jjiastar/bulkcorr
> > >> >
> > >> > Preliminary page :
> > >> > https://drupal.star.bnl.gov/STAR/pwg/bulk-correlations/bulkcorr-pre
> > >> > lim inary-summary
> > _______________________________________________
> > >> > Star-fcv-l mailing list
> > >> > Star-fcv-l AT lists.bnl.gov
> > >> > https://lists.bnl.gov/mailman/listinfo/star-fcv-l
> > >> _______________________________________________
> > >> Star-fcv-l mailing list
> > >> Star-fcv-l AT lists.bnl.gov
> > >> https://lists.bnl.gov/mailman/listinfo/star-fcv-l