[Eg6_analysis] Meeting tomorrow

[Stepan Stepanyan]

Hovanes and Nathan,

I agree with description that Hovanes gave. In such cases
the straight forward solution is to assume equal probability
of each hypothesis. But even with that assumption things
will be quite complicated, probability density function is
2-D, since dE/dX depends on momentum and for each
momentum one must have normalized distributions for
all spices, but with momentum and dE/dX resolutions
this will become a nightmare.

So, I think we should drop the idea of providing probability
for given track to be certain particle. Instead we should just
limit output of RTPC to 3-IDs and 3-momenta (corrected for
energy loss according to IDs). Basically what Nathan's originally
proposal was.

My two cents.

Stepan

On 8/17/12 11:46 AM, Hovanes Egiyan wrote:
> Hi Nathan,
>
> During the last eg6 meeting we briefly discussed what
> number to use to indicate our confidence in a certain ID
> of the particle based on the dE/dx measurement in RTPC. I think I said
> that we could use some probabilities that are defined for each
> individual particle type. After giving some more thought on that
> I think it should be done differently. It is more complicated than I
> thought,
> but it needs to be to be of good use. To summarize in the beginning, I
> think
> these number can be useful if someone wants to do something really quick
> without developing his own RTPC PID scheme. For detail analysis
> people would need to refine RTPC PID.
>
> If we want to specify a probability that a given track was created
> by a certain particle based on some measured quantity, for instance
> dE/dX, not only we need to know the probability density function for
> that quantity given
> the particle ID, but also how often that particular particle is created
> in the
> scattering process that we are looking for. That frequency, strictly
> speaking ,
> depends on the event topology (and kinematics), since for certain
> selection of particles detected
> in CLAS some of the types of the recoil particles will be very unlikely.
> For instance, if one measures +2 total charge in CLAS the probability of
> having a He4 in RTPC with respect to the probability of having a proton
> (or a deuteron)
> becomes smaller compared to the case when the total charge seen in CLAS
> is -1, because
> the total charge in CLAS&  RTPC&  undetected particles needs to be the
> same as
> the charge of the original target and electron.  And then if one uses
> some cuts,
> like missing mass cuts to select his reaction then these prior
> probabilities can be
> completely different.
>
> Therefore, determining the correct probability for particle ID for
> every reaction
> by just  one number will be impossible. One can do it for the total sample
> of the tracks in the RTPC by somehow figuring out the prior
> probabilities for
> the different particle ID regardless of what was detected in CLAS.
> One also could pick a particular "important" event topology in CLAS to
> define these numbers, for instance four photon topology, or electron and
> a photon.
> In the first approximation these would be  reasonable numbers, but then
> every
> reaction would need to improve them by "recalibrating" the prior
> probabilities for
> different PID for RTPC tracks and reassigning the final PID
> probabilities.  One would need to have the dE/dX information to redo
> the probability assignment, although the value for the probability
> distribution functions (f_{id}(dE/dx)) for the ionization losses
> for each PID hypothesis would suffice.  Then the calculation of the
> probability that the track was created by a particle with "id" should be:
>
> P(id) = [ a_{id} \times f_{id}( dE/dX ) ]  /  \sum_{ids} ( a_{ids}
> \times f_{ids}( dE/dX ) )
>
> where in the denominator we sum over all IDs we are considering, and
> a_{id} is
> the prior probability for sending a particle with such an particle ID
> into the RTPC. These
> a_{id}-s can also be kinematics-dependent, but I think this is something
> for each
> analysis to decide.
>
> Of course, the person doing his own detail data analysis might argue
> that he'd rather
> just look at the dE/dX distribution vs P and use some curves to pick out
> his He4 tracks,
> and in  many cases he might be right...
>
> Let me know what you think.
>
> Best,
>       Hovanes.
>
>
> On 08/15/2012 01:34 PM, Stepan Stepanyan wrote:
>> Dear all,
>>
>> We will have eg6 meeting tomorrow, Thursday, August 16 at 10 am in F227.
>>
>> Preliminary agenda is at
>> http://clasweb.jlab.org/rungroups/lowq/wiki/index.php/Meeting_on_August_16_at_10:00am
>>
>> Calling instruction can be found at -
>> http://clasweb.jlab.org/rungroups/lowq/wiki/index.php/Agendas_and_Minutes_of_the_meetings#Agendas
>>
>>
>> Regards, Stepan
>> _______________________________________________
>> Eg6_analysis mailing list
>> Eg6_analysis at jlab.org
>> https://mailman.jlab.org/mailman/listinfo/eg6_analysis
> _______________________________________________
> Eg6_analysis mailing list
> Eg6_analysis at jlab.org
> https://mailman.jlab.org/mailman/listinfo/eg6_analysis