[Frost] [EXTERNAL] Re: Follow up of last FROST meeting

Sun Dec 29 01:41:10 EST 2019

Thank you for your comments!! I will make the mmsq comparison plot of the
first half and second half of the butanol target.

Yes, there is a lot of ambiguity in ML approach and until I find a way to
probe into how the training is actually done in neural nets, it is still a
black box that I should not trust.
My first aim is to correctly produce the results following the simplest
approaches without any techniques that has unknown errors. Only reason why
I used machine learning is because my dissertation proposal includes the
use of machine learning (or at least try) and my graduate committee
approved on it.

I think there are many obstacles in regards to ML even before trying to
determine whether results from ML is usable or not. First I need to be able
to know how the training is done and the reasoning behind the
classification if I am to use the result.
To do so, I need to have a visualization of how weights are being changed
as they are being passed on to each neural layers and how they are being
optimized. Secondly, I need to find a way to compute the uncertainty in
training. I was trying to build a kind of neural net, called bayesian
neural network which passes probability density distribution instead of
scale weights as shown in below figure. Then use the final uncertainties.

Bottom line is, I will not focus on ML until my standard procedures are
approved and my results on asymmetry E are reasonable.

[image: BayesianNN.png]

Thank you,
Chan

On Sat, Dec 28, 2019 at 10:43 PM Michael Dugger <dugger at jlab.org> wrote:

>
> Hi,
>
> Sounds like a lot of work for little gain. Arguing that the shapes are
> correct will require more effort than it will be worth, and comparing
> front to back is not as straight-forward as it might seem (differing
> amounts of eloss correction and acceptance at low momentum and low angle).
>
> If that is how Chan wants to use his time, I will not stand in his way.
>
> Take care,
> Michael
>
> > Then in that case it should show up in different dilution factor for the
> > first half of the target as compared to the second half. This should be
> > rather straightforward to show.
> >
> > If one can argue that the shapes are correct (vertex resolution) one can
> > use these to tailor the z vertex cuts on a bin by bin basis.
> >
> > Sent from my iPhone
> >
> >> On 29 Dec 2019, at 02:47, Michael Dugger <dugger at jlab.org> wrote:
> >>
> >> Hi,
> >>
> >> The same problem applies for the separation of carbon from butanol.
> >> There
> >> is no way to distinguish the two types of bound content. The neural net
> >> is
> >> making some sort of distinction that, on the face of it, looks
> >> reasonably
> >> good. It looks like there are two gaussian distributions that overlap.
> >> The
> >> problem is that there is no reason to believe that the separation is in
> >> any way faithful. Part of the carbon tail in the overlap region is
> >> actually some combination of carbon and butanol and part of the butanol
> >> tail in the overlap region is actually some combination of carbon and
> >> butanol. We just do not know the relative amounts. Perhaps the shapes
> >> are
> >> somewhat correct and we could make a good guess as to the proportion of
> >> each type of events in the overlap?
> >>
> >> In short, the separation looks nice but can not be used as representing
> >> pure samples of carbon (or butanol) in the overlap region.
> >>
> >> Take care,
> >> Michael
> >>
> >>> I agree that the classification of free vs bound protons is not
> >>> possible.
> >>> What the plot he sent shows (maybe thats just me misreading it) is the
> >>> classification of events based on target: i.e. butanol vs carbon. If
> >>> the
> >>> ML can correctly classify butanol vs carbon events then a vertex cut is
> >>> not really essential. If he can show that the dilution (ratio of bound
> >>> to
> >>> total) is the same between events classified as butanol and events with
> >>> tight z vertex cuts, that is a first indication that the classification
> >>> is
> >>> adequate.
> >>>
> >>> I might not be following the discussion or have miss-interpreted
> >>> things:
> >>> its 2am here afterall :)
> >>>
> >>> Sent from my iPhone
> >>>
> >>>>> On 29 Dec 2019, at 00:55, Michael Dugger <dugger at jlab.org> wrote:
> >>>>
> >>>> Hi,
> >>>>
> >>>> My impression of the machine learning implemented by Chan was that it
> >>>> was
> >>>> not able to separate the ice from the carbon.
> >>>>
> >>>> The problem is that there is no way to take an event originating from
> >>>> a
> >>>> free proton (like that in ice) and distinguish that event from one
> >>>> that
> >>>> originates from a bound proton (like that in carbon). Because of this,
> >>>> the
> >>>> two types of events are NOT linearly separable.
> >>>>
> >>>> Look at it this way: If Chan could remove the signal coming from ice
> >>>> from
> >>>> that of carbon, why not just classify the free proton directly from
> >>>> the
> >>>> bound content within butanol?
> >>>>
> >>>> If machine learning could distinguish free proton from bound, there
> >>>> would
> >>>> be no need for a carbon subtraction we would just use machine learning
> >>>> to
> >>>> do the separation.
> >>>>
> >>>> In the MM-Sq region that is physical, the protons from bound and
> >>>> unbound
> >>>> have angles and momentum that overlap and can not be distinguished
> >>>> from
> >>>> one another.
> >>>>
> >>>> Take care,
> >>>> Michael
> >>>>
> >>>>> Hi Chan,
> >>>>>
> >>>>> I dont know the details of your classification, but my first
> >>>>> impression
> >>>>> is
> >>>>> yes. If you have a reliable classification then you can (in
> >>>>> principle)
> >>>>> not
> >>>>> apply z vertex cuts. You can just use your butanol events as chosen
> >>>>> by
> >>>>> the
> >>>>> classification And your carbon events for establishing the bound
> >>>>> nucleon
> >>>>> contributions. This would of course need to be studied in a
> >>>>> systematic
> >>>>> way
> >>>>> and a classification uncertainty needs to be established. It looks
> >>>>> like
> >>>>> you were also able to remove the ice build up downstream the carbon
> >>>>> target. Might be good to see the missing masses originating from the
> >>>>> two
> >>>>> targets. A quick check for your classification would be to plot the
> >>>>> missing mass of events coming with z vertex that selects the first
> >>>>> half
> >>>>> of
> >>>>> your target (any events away from the carbon target ). Then scale
> >>>>> this
> >>>>> and
> >>>>> see if you can get the missing mass distribution to match the one
> >>>>> from
> >>>>> your classified butanol events. If it does, it means you are
> >>>>> correctly
> >>>>> removing any contributions from the carbon target which would only
> >>>>> contribute to background. If you fail to match the free proton peak
> >>>>> and
> >>>>> the bound nucleon contributions it means that the classification
> >>>>> might
> >>>>> need some work.
> >>>>>
> >>>>> Sent from my iPhone
> >>>>>
> >>>>>>> On 28 Dec 2019, at 21:26, Chan Kim <kimchanwook at gwmail.gwu.edu>
> >>>>>>> wrote:
> >>>>>>
> >>>>>> 
> >>>>>> Thank you for your replies.
> >>>>>>
> >>>>>> @Michael,
> >>>>>> For the z-vertex plot, only p<300MeV particles are removed. I will
> >>>>>> change the momentum selection region and see how z-vertex resolution
> >>>>>> changes as soon as I have other steps finished. I kept them for now,
> >>>>>> so
> >>>>>> I can compare my asymmetry with and without particles with p=[300,
> >>>>>> 400]
> >>>>>> MeV.
> >>>>>>
> >>>>>> @Nick,
> >>>>>> In regards to free polarized proton contributions in carbon vertex
> >>>>>> (left
> >>>>>> tail of the carbon distribution), would it be ok to use
> >>>>>> classification
> >>>>>> results from machine learning? Below is a plot of separation of
> >>>>>> butanol
> >>>>>> and carbon in small angles.
> >>>>>>
> https://clasweb.jlab.org/rungroups/g9/wiki/images/0/06/Zvrt_1bin_angle.png
> >>>>>>
> >>>>>> Thank you,
> >>>>>> Chan
> >>>>>>
> >>>>>>
> >>>>>>> On Sat, Dec 28, 2019 at 12:42 PM Nicholas Zachariou
> >>>>>>> <nickzachariou at gmail.com> wrote:
> >>>>>>> Dear Chan,
> >>>>>>>
> >>>>>>> I have done several studies on this that should be documented in my
> >>>>>>> analysis note. As Michael said, the difficulty lies in the
> >>>>>>> determination of the dilution factor. If you are using carbon data
> >>>>>>> to
> >>>>>>> determine this, then you need to make sure you account correctly
> >>>>>>> for
> >>>>>>> free polarised proton contributions within your carbon vertex cut
> >>>>>>> that
> >>>>>>> actually comes from butanol events (this is different than the ice
> >>>>>>> built up). If you are using parametrization of the bound nucleon
> >>>>>>> contributions then the dilution is more straightforward, but you
> >>>>>>> would
> >>>>>>> need to think/study free unpolarised proton contributions to
> >>>>>>> butanol
> >>>>>>> events from the ice built up downstream your carbon target. This
> >>>>>>> can
> >>>>>>> been done by varying the butanol z vertex cut andcomparing your
> >>>>>>> determined observable. There are several ways you can do this and
> >>>>>>> we
> >>>>>>> can chat more if needed.
> >>>>>>>
> >>>>>>> Best regard,
> >>>>>>> Nick
> >>>>>>>
> >>>>>>> Sent from my iPhone
> >>>>>>>
> >>>>>>>> On 28 Dec 2019, at 16:23, Michael Dugger <dugger at jlab.org> wrote:
> >>>>>>>>
> >>>>>>>> Chan,
> >>>>>>>>
> >>>>>>>> The variables that are important are polar angle and momentum of
> >>>>>>>> the
> >>>>>>>> identified particle. Did you keep the low momentum protons?
> >>>>>>>>
> >>>>>>>> When it comes to choosing the range of vertex selection for each
> >>>>>>> target,
> >>>>>>>> the method you use for removing the bound nucleon content might be
> >>>>>>>> a
> >>>>>>>> factor. From the plots you created, you can see the difficulty
> >>>>>>>> that
> >>>>>>> you
> >>>>>>>> face at low angle.
> >>>>>>>>
> >>>>>>>> At low polar angle you have the case that there very well could be
> >>>>>>> some
> >>>>>>>> leakage of carbon vertices in your butanol definition. This could
> >>>>>>> cause
> >>>>>>>> dilution factors to be messed up.
> >>>>>>>>
> >>>>>>>> I would cautiously move forward concentrating on the bins that
> >>>>>>>> look
> >>>>>>> good.
> >>>>>>>> Once everything was running well for the "good" stuff, I would
> >>>>>>>> start
> >>>>>>>> trying to fix the more difficult bins.
> >>>>>>>>
> >>>>>>>> Take care,
> >>>>>>>> Michael
> >>>>>>>>
> >>>>>>>>> Hello Michael,
> >>>>>>>>>
> >>>>>>>>> The pi0 mass shifting from 125MeV (slide 2) to 134MeV (slide 6)
> >>>>>>> happened
> >>>>>>>>> simply due to a change in number of bins. On slide 6, I tried to
> >>>>>>> have the
> >>>>>>>>> same binning for both ELOSS and ELOSS+momentum correction
> >>>>>>>>> distributions,causing inaccurate peak locations. After momentum
> >>>>>>>>> correction,
> >>>>>>>>> the mmsq distribution has much sharper peaks. So, 125MeV is the
> >>>>>>> correct
> >>>>>>>>> value of mmsq after ELOSS distribution, but for ELOSS + momentum
> >>>>>>>>> correction, I will redo the calculation. Sorry for the confusion.
> >>>>>>>>>
> >>>>>>>>> I have another question regarding the z-vertex resolution in
> >>>>>>>>> angles.
> >>>>>>> Below
> >>>>>>>>> is a plot that Eugene suggested to make (Z-vertex position in
> >>>>>>>>> bins
> >>>>>>> of
> >>>>>>>>> photon energies and angle bins).
> >>>>>>>>>
> https://clasweb.jlab.org/rungroups/g9/wiki/images/a/ac/Zvrt_select_lab.png
> >>>>>>>>>
> >>>>>>>>> As you can see, the z-vertex resolution is very poor in small
> >>>>>>> scattering
> >>>>>>>>> angles, unable to clearly distinguish the butanol and carbon
> >>>>>>>>> region.
> >>>>>>> do
> >>>>>>>>> you
> >>>>>>>>> have any suggestions in choosing the selection ranges for small
> >>>>>>> angles..?
> >>>>>>>>>
> >>>>>>>>> Thank you,
> >>>>>>>>> Chan
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>> On Thu, Dec 26, 2019 at 4:53 PM Michael Dugger <dugger at jlab.org
> >
> >>>>>>> wrote:
> >>>>>>>>>>
> >>>>>>>>>> Chan,
> >>>>>>>>>>
> >>>>>>>>>> It is probably a good idea to try and keep the low momentum
> >>>>>>>>>> events,
> >>>>>>> but
> >>>>>>>>>> I
> >>>>>>>>>> would start out by removing them. Once you feel like you have
> >>>>>>> everything
> >>>>>>>>>> else under control, then you add in the p < 400 MeV/c events.
> >>>>>>>>>>
> >>>>>>>>>> You can see if the momentum cut makes the pi0 mass closer to the
> >>>>>>> nominal
> >>>>>>>>>> value. Essentially, you just recreate slides 2 and 6 from your
> >>>>>>> December
> >>>>>>>>>> 18
> >>>>>>>>>> update:
> >>>>>>>>>>
> >>>>>>>>>>
> https://clasweb.jlab.org/rungroups/g9/wiki/images/9/94/FROST_2019_12_18.pdf
> >>>>>>>>>>
> >>>>>>>>>> Note: On slide 2 you have pre-ELOSS mass of pi0 = 196 MeV and
> >>>>>>> post-ELOSS
> >>>>>>>>>> mass of pi0 = 125 MeV. BUT on slide 6 you have post-ELOSS mass
> >>>>>>>>>> of
> >>>>>>> pi0 =
> >>>>>>>>>> 134 MeV!! How did that happen? Then after ELOSS + momentum
> >>>>>>> correction
> >>>>>>>>>> the
> >>>>>>>>>> pi0 mass = 89 MeV. You want to get that all straightened out
> >>>>>>>>>> with
> >>>>>>> the p
> >>>>>>>>>>>
> >>>>>>>>>> 400 MeV/c events. You mass of the pi0 after the ELOSS and
> >>>>>>>>>> momentum
> >>>>>>>>>> corrections should be the correct value. The momentum correction
> >>>>>>>>>> is
> >>>>>>>>>> designed to bring the pi0 mass to the correct value.
> >>>>>>>>>>
> >>>>>>>>>> Take care,
> >>>>>>>>>> Michael
> >>>>>>>>>>
> >>>>>>>>>>> Hello Michael,
> >>>>>>>>>>>
> >>>>>>>>>>> Yes, I now understand what you are saying about the momentum
> >>>>>>>>>> acceptance
> >>>>>>>>>>> varying rapidly between 300 and 400 MeV around scattering angle
> >>>>>>>>>>> of
> >>>>>>>>>> 35deg.
> >>>>>>>>>>>
> >>>>>>>>>>> Would it be a bad idea to apply different low momentum cuts for
> >>>>>>>>>> particles
> >>>>>>>>>>> under 35deg and above 35deg? So, 400MeV for particles with
> >>>>>>>>>>> angle
> >>>>>>>>>>> <
> >>>>>>>>>> 35deg
> >>>>>>>>>>> and 300MeV for particles with angle >35deg.
> >>>>>>>>>>> Because for higher energy photon events (photon energy > 1
> >>>>>>>>>>> GeV),
> >>>>>>> it
> >>>>>>>>>> looks
> >>>>>>>>>>> like there are lots of events in p = [350, 400] MeV region.
> >>>>>>>>>>>
> >>>>>>>>>>> Thank you,
> >>>>>>>>>>> Chan
> >>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>> On Tue, Dec 24, 2019 at 3:50 AM Michael Dugger
> >>>>>>>>>>> <dugger at jlab.org>
> >>>>>>>>>> wrote:
> >>>>>>>>>>>
> >>>>>>>>>>>> Chan,
> >>>>>>>>>>>>
> >>>>>>>>>>>> For these sorts of plots it is interesting to make an
> >>>>>>>>>>>> additional
> >>>>>>> MM^2
> >>>>>>>>>>>> cut.
> >>>>>>>>>>>> If you make a missing mass cut near the mass of the pion, you
> >>>>>>>>>> restrict
> >>>>>>>>>>>> the
> >>>>>>>>>>>> events to have the kinematics of
> >>>>>>>>>>>>
> >>>>>>>>>>>> gamma p -> p pi0 .
> >>>>>>>>>>>>
> >>>>>>>>>>>> As it currently stands, you are showing the theta versus
> >>>>>>>>>>>> momentum
> >>>>>>>>>>>> distribution for all possible reactions. This is not a bad
> >>>>>>>>>>>> thing
> >>>>>>> and
> >>>>>>>>>> the
> >>>>>>>>>>>> plots have interesting features.
> >>>>>>>>>>>>
> >>>>>>>>>>>> The depletion stripes are probably bad time-of-flight paddles.
> >>>>>>> You
> >>>>>>>>>> can
> >>>>>>>>>>>> see
> >>>>>>>>>>>> from the depletion stripes how the magnetic field is bending
> >>>>>>>>>>>> the
> >>>>>>> path
> >>>>>>>>>> of
> >>>>>>>>>>>> the charged particles as a function of momentum.
> >>>>>>>>>>>>
> >>>>>>>>>>>> The enhancement stripes are probably from the reaction gamma p
> >>>>>>>>>>>> ->
> >>>>>>> p
> >>>>>>>>>> pi0
> >>>>>>>>>>>> (compare to
> >>>>>>>>>> https://userweb.jlab.org/~dugger/pi0PphaseSpacePlotNew.png
> >>>>>>>>>>>> ).
> >>>>>>>>>>>>
> >>>>>>>>>>>> You can clearly see the acceptance in polar angle and
> >>>>>>>>>>>> momentum.
> >>>>>>> For
> >>>>>>>>>>>> angle
> >>>>>>>>>>>> above 35 degrees, it looks like the momentum acceptance is
> >>>>>>> rapidly
> >>>>>>>>>>>> changing between 300 and 400 MeV/c. This is probably the
> >>>>>>>>>>>> clearest
> >>>>>>>>>>>> evidence
> >>>>>>>>>>>> that a cut at 400 MeV/c is appropriate. For angles below ~35
> >>>>>>> degrees
> >>>>>>>>>> the
> >>>>>>>>>>>> momentum acceptance issue is being caused by a bad
> >>>>>>>>>>>> time-of-flight
> >>>>>>>>>>>> paddle.
> >>>>>>>>>>>>
> >>>>>>>>>>>> Take care,
> >>>>>>>>>>>> Michael
> >>>>>>>>>>>>
> >>>>>>>>>>>>> Hello All,
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> Here is a plot of momentum vs lab angle in bins of photon
> >>>>>>> energies.
> >>>>>>>>>>>>>
> >>>>>>>>>>>>
> >>>>>>>>>>
> https://clasweb.jlab.org/rungroups/g9/wiki/images/2/2e/Low_mome_select_p_abs_theta.png
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> Thank you,
> >>>>>>>>>>>>> Chan
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> On Mon, Dec 23, 2019 at 3:47 PM Chan Kim
> >>>>>>>>>> <kimchanwook at gwmail.gwu.edu>
> >>>>>>>>>>>>> wrote:
> >>>>>>>>>>>>>
> >>>>>>>>>>>>>> Hello Michael,
> >>>>>>>>>>>>>>
> >>>>>>>>>>>>>> Thank you for your suggestion! I should have been more clear
> >>>>>>> about
> >>>>>>>>>>>> the
> >>>>>>>>>>>>>> plots. The mmsq distribution I sent out yesterday was prior
> >>>>>>>>>>>>>> to
> >>>>>>> any
> >>>>>>>>>>>> event
> >>>>>>>>>>>>>> selections.
> >>>>>>>>>>>>>>
> >>>>>>>>>>>>>> I have two quick questions..
> >>>>>>>>>>>>>> 1. For momentum ranges of [0.31, 0.39] GeV, the mmsq
> >>>>>>> distribution
> >>>>>>>>>>>> (ones
> >>>>>>>>>>>>>> I
> >>>>>>>>>>>>>> sent yesterday) look like a gaussian distribution centered
> >>>>>>>>>>>>>> near
> >>>>>>> 0,
> >>>>>>>>>>>> plus
> >>>>>>>>>>>>>> a
> >>>>>>>>>>>>>> bump at 1GeV. Are these events bad because of these bumps at
> >>>>>>> 1GeV?
> >>>>>>>>>>>>>>
> >>>>>>>>>>>>>> 2. Are events with backward scattering angles not good
> >>>>>>>>>>>>>> because
> >>>>>>>>>> they
> >>>>>>>>>>>>>> don't
> >>>>>>>>>>>>>> have information from DC?? since the drift chamber only
> >>>>>>>>>>>>>> covers
> >>>>>>>>>> from
> >>>>>>>>>>>> 8deg
> >>>>>>>>>>>>>> to
> >>>>>>>>>>>>>> 142deg???
> >>>>>>>>>>>>>>
> >>>>>>>>>>>>>> Regards,
> >>>>>>>>>>>>>> Chan
> >>>>>>>>>>>>>>
> >>>>>>>>>>>>>> On Mon, Dec 23, 2019 at 3:33 PM Michael Dugger
> >>>>>>> <dugger at jlab.org>
> >>>>>>>>>>>> wrote:
> >>>>>>>>>>>>>>
> >>>>>>>>>>>>>>> Hi,
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>> The last plot was for pion lab-momentum and lab-angles and
> >>>>>>>>>>>>>>> the
> >>>>>>>>>>>>>>> center-of-mass angle definitions were also messed up :(
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>> I have a new plot at
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>> https://userweb.jlab.org/~dugger/pi0PphaseSpacePlotNew.png
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>> The above plot shows that the momentum values below 400 MeV
> >>>>>>> ARE
> >>>>>>>>>>>>>>> important
> >>>>>>>>>>>>>>> for MANY kinematic bins.
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>> However, even with the knowledge that we would be killing a
> >>>>>>> bunch
> >>>>>>>>>> of
> >>>>>>>>>>>>>>> bins,
> >>>>>>>>>>>>>>> we may have to remove events with momentum below 400 MeV
> >>>>>>>>>>>>>>> due
> >>>>>>> to
> >>>>>>>>>> our
> >>>>>>>>>>>>>>> possible inability to reconstruct the missing pi0 at low
> >>>>>>>>>> momentum.
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>> Sorry about any confusion my previous plot may have caused.
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>> Take care,
> >>>>>>>>>>>>>>> Michael
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>> Eugene,
> >>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>> Thanks for catching that. I think I have pion angles
> >>>>>>>>>>>>>>>> instead
> >>>>>>> of
> >>>>>>>>>>>>>>> proton.
> >>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>> I am going to back over the code to fix this.
> >>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>> Take care,
> >>>>>>>>>>>>>>>> Michael
> >>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>> Mike,
> >>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>> The vertical axis can't be right. The proton can't go
> >>>>>>>>>> backwards
> >>>>>>>>>>>> in
> >>>>>>>>>>>>>>> the
> >>>>>>>>>>>>>>>>> lab
> >>>>>>>>>>>>>>>>> system
> >>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>> -Eugene
> >>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> -----Original Message-----
> >>>>>>>>>>>>>>>>>> From: Frost <frost-bounces at jlab.org> On Behalf Of
> >>>>>>>>>>>>>>>>>> Michael
> >>>>>>>>>> Dugger
> >>>>>>>>>>>>>>>>>> Sent: Monday, December 23, 2019 14:07
> >>>>>>>>>>>>>>>>>> To: Stuart Fegan <s.fegan.glasgow at gmail.com>
> >>>>>>>>>>>>>>>>>> Cc: frost at jlab.org
> >>>>>>>>>>>>>>>>>> Subject: Re: [Frost] [EXTERNAL] Re: Follow up of last
> >>>>>>>>>>>>>>>>>> FROST
> >>>>>>>>>>>> meeting
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> Hi,
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> Chan is just trying to answer questions raised about a
> >>>>>>>>>> possible
> >>>>>>>>>>>>>>>>>> momentum
> >>>>>>>>>>>>>>>>>> cut. I suggested that he look at where his events are in
> >>>>>>>>>> terms
> >>>>>>>>>>>> of
> >>>>>>>>>>>>>>>>>> kinematic
> >>>>>>>>>>>>>>>>>> bins he will report on. My idea was to see if pushing up
> >>>>>>> the
> >>>>>>>>>>>>>>> momentum
> >>>>>>>>>>>>>>>>>> cut
> >>>>>>>>>>>>>>>>>> to perhaps 400 MeV would cause any issues within his
> >>>>>>>>>> kinematic
> >>>>>>>>>>>>>>> binning.
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> I just made a plot that can be found at
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> https://userweb.jlab.org/~dugger/pi0PphaseSpacePlot.png
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> that shows the lab angle versus lab momentum for protons
> >>>>>>>>>> coming
> >>>>>>>>>>>>>>> from
> >>>>>>>>>>>>>>>>>> the
> >>>>>>>>>>>>>>>>>> reaction gamma p -> p pi0. The black curves are for
> >>>>>>> constant
> >>>>>>>>>>>> photon
> >>>>>>>>>>>>>>>>>> energy
> >>>>>>>>>>>>>>>>>> and the blue curves are for constant proton
> >>>>>>>>>>>>>>>>>> center-of-mass
> >>>>>>>>>>>> cosine
> >>>>>>>>>>>>>>>>>> values.
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> The above plot would have to be verified but it looks
> >>>>>>>>>>>>>>>>>> like
> >>>>>>>>>> there
> >>>>>>>>>>>> is
> >>>>>>>>>>>>>>> no
> >>>>>>>>>>>>>>>>>> need to
> >>>>>>>>>>>>>>>>>> worry about low momentum protons.
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> The idea I had was for Chan to produce this type of
> >>>>>>>>>> information
> >>>>>>>>>>>>>>> using
> >>>>>>>>>>>>>>>>>> real
> >>>>>>>>>>>>>>>>>> data, but I did a poor job of explaining what I meant.
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> I was trying to make life easier, but perhaps did not
> >>>>>>>>>> accomplish
> >>>>>>>>>>>>>>> that
> >>>>>>>>>>>>>>>>>> :(
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> Take care,
> >>>>>>>>>>>>>>>>>> Michael
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>> Hi Chan,
> >>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>> I'm going to chuck my two cents in, and reply to the
> >>>>>>>>>>>>>>>>>>> FROST
> >>>>>>>>>>>> list,
> >>>>>>>>>>>>>>>>>>> because I missed the meeting last week.Ã‚  Given the
> >>>>>>>>>> pion
> >>>>>>>>>>>> is
> >>>>>>>>>>>>>>>>>>> reconstructed from the proton missing mass, what's the
> >>>>>>>>>>>> motivation
> >>>>>>>>>>>>>>> for
> >>>>>>>>>>>>>>>>>>> looking at proton momenta below the threshold where it
> >>>>>>>>>>>>>>>>>>> can
> >>>>>>>>>>>>>>> reliably
> >>>>>>>>>>>>>>>>>>> reconstructed in CLAS as a proton?Ã‚  Is this to
> >>>>>>>>>>>>>>>>>>> tune
> >>>>>>>>>> the
> >>>>>>>>>>>> cut,
> >>>>>>>>>>>>>>> perform
> >>>>>>>>>>>>>>>>>>> systematic studies, or is there a physics motivation
> >>>>>>>>>>>>>>>>>>> here
> >>>>>>>>>> that
> >>>>>>>>>>>>>>> I'm
> >>>>>>>>>>>>>>>>>> missing?
> >>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>> Cheers,
> >>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>> Stuart
> >>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>> On 23/12/2019 11:47, Michael Dugger wrote:
> >>>>>>>>>>>>>>>>>>>> Chan,
> >>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>> It is a bit of a data dump.
> >>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>> What is your binning going to by for the analysis? Are
> >>>>>>> you
> >>>>>>>>>>>>>>> really
> >>>>>>>>>>>>>>>>>>>> going to report values for E_gamma near 400 MeV?
> >>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>> On slide 3 you show MM^2 and state that MM^2 for p <
> >>>>>>>>>>>>>>>>>>>> 280
> >>>>>>>>>>>> MeV/c
> >>>>>>>>>>>>>>> don't
> >>>>>>>>>>>>>>>>>>>> look like the others. I'm not convinced that you can
> >>>>>>>>>>>>>>>>>>>> say
> >>>>>>>>>> much
> >>>>>>>>>>>>>>> about
> >>>>>>>>>>>>>>>>>>>> the MM^2 shown above p = 280 MeV/c.
> >>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>> For your previous presentation:
> >>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>
> https://clasweb.jlab.org/rungroups/g9/wiki/images/9/94/FROST_2019_12_
> >>>>>>>>>>>>>>>>>>>> 18.pdf
> >>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>> on slide 2 you had a nice fit to the MM^2 distribution
> >>>>>>>>>> where
> >>>>>>>>>>>> you
> >>>>>>>>>>>>>>>>>>>> pulled off a pi0 mass. Are you able to do that for the
> >>>>>>> low
> >>>>>>>>>>>>>>> momentum?
> >>>>>>>>>>>>>>>>>>>> Is it possible that you can not pull out any pi0 from
> >>>>>>>>>>>>>>>>>>>> the
> >>>>>>>>>> low
> >>>>>>>>>>>>>>>>>>>> momentum data? I just do not see any pi0. Am I missing
> >>>>>>>>>>>>>>> something?
> >>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>> Take care,
> >>>>>>>>>>>>>>>>>>>> Michael
> >>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>> Dear FROST run group,
> >>>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>> Hello, below is a link to my slides for follow up of
> >>>>>>> last
> >>>>>>>>>>>> FROST
> >>>>>>>>>>>>>>>>>>>>> meeting(12/19):
> >>>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>
> https://clasweb.jlab.org/rungroups/g9/wiki/images/e/ed/FROST_2019_12
> >>>>>>>>>>>>>>>>>>>>> _22.pdf
> >>>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>> 1. Distributions of kinematics (MMSQ, dt, d\beta) for
> >>>>>>>>>>>> particles
> >>>>>>>>>>>>>>> in
> >>>>>>>>>>>>>>>>>>>>> lower momentum ranges are plotted to see whether
> >>>>>>>>>>>>>>>>>>>>> lower
> >>>>>>>>>>>> momentum
> >>>>>>>>>>>>>>>>>>>>> particles are of any use for my asymmetry
> >>>>>>>>>>>>>>>>>>>>> calculation.
> >>>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>> 2. Proton selection, using beta difference, was
> >>>>>>>>>>>>>>>>>>>>> revised
> >>>>>>>>>> to a
> >>>>>>>>>>>>>>>>>> simpler
> >>>>>>>>>>>>>>>>>>>>> version where static cuts on beta diff are applied at
> >>>>>>> +/-
> >>>>>>>>>>>> 0.06
> >>>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>> Thank you,
> >>>>>>>>>>>>>>>>>>>>> Chan
> >>>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>> _______________________________________________
> >>>>>>>>>>>>>>>>>>>>> Frost mailing list
> >>>>>>>>>>>>>>>>>>>>> Frost at jlab.org
> >>>>>>>>>>>>>>>>>>>>> https://mailman.jlab.org/mailman/listinfo/frost
> >>>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>>> _______________________________________________
> >>>>>>>>>>>>>>>>>>>> Frost mailing list
> >>>>>>>>>>>>>>>>>>>> Frost at jlab.org
> >>>>>>>>>>>>>>>>>>>> https://mailman.jlab.org/mailman/listinfo/frost
> >>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>> --
> >>>>>>>>>>>>>>>>>>> Dr Stuart Fegan
> >>>>>>>>>>>>>>>>>>> Honorary Research Associate
> >>>>>>>>>>>>>>>>>>> Nuclear Physics Group
> >>>>>>>>>>>>>>>>>>> University of Glasgow
> >>>>>>>>>>>>>>>>>>> (Currently at the University of York)
> >>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>> E-mail: s.fegan.glasgow at gmail.com
> >>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>> _______________________________________________
> >>>>>>>>>>>>>>>>>>> Frost mailing list
> >>>>>>>>>>>>>>>>>>> Frost at jlab.org
> >>>>>>>>>>>>>>>>>>> https://mailman.jlab.org/mailman/listinfo/frost
> >>>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>>> _______________________________________________
> >>>>>>>>>>>>>>>>>> Frost mailing list
> >>>>>>>>>>>>>>>>>> Frost at jlab.org
> >>>>>>>>>>>>>>>>>> https://mailman.jlab.org/mailman/listinfo/frost
> >>>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>> _______________________________________________
> >>>>>>>>>>>>>>> Frost mailing list
> >>>>>>>>>>>>>>> Frost at jlab.org
> >>>>>>>>>>>>>>> https://mailman.jlab.org/mailman/listinfo/frost
> >>>>>>>>>>>>>>>
> >>>>>>>>>>>>>>
> >>>>>>>>>>>>>
> >>>>>>>>>>>>
> >>>>>>>>>>>>
> >>>>>>>>>>>> _______________________________________________
> >>>>>>>>>>>> Frost mailing list
> >>>>>>>>>>>> Frost at jlab.org
> >>>>>>>>>>>> https://mailman.jlab.org/mailman/listinfo/frost
> >>>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>
> >>>>>>>>
> >>>>>>>>
> >>>>>>>> _______________________________________________
> >>>>>>>> Frost mailing list
> >>>>>>>> Frost at jlab.org
> >>>>>>>> https://mailman.jlab.org/mailman/listinfo/frost
> >>>>>
> >>>>
> >>>>
> >>>
> >>
> >>
> >
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/frost/attachments/20191229/2784cc02/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: BayesianNN.png
Type: image/png
Size: 85689 bytes
Desc: not available
URL: <https://mailman.jlab.org/pipermail/frost/attachments/20191229/2784cc02/attachment-0001.png>