1.- General
--------o Online exclusion Jason Nielsen2.- Analysis
o Discovery exercise, using Hqq cuts Pedro Teixeira-Dias
--------o A Few Developments in Hvv Jennifer Kile
Jason presented a proposal for a system which will be able to run
the analyses online.
The analyses used are those used to provide the 99' Moriond
conf-note.
This follows the proposal sent around by Nikos&Pascal on 5 April.
It is intended to use these analyses for Summer conf and End of Run LEPC
results; and to at least give their results for Winter Conf and
publication (together with the results of possible improved analyses).
The architecture proposed consists in four tasks.
a) create the cards from scanbook; get the integrated luminosity.
b) run the analyses, collect information about the selected
candidates
c) compute expected and observed Confidence levels
d) provide plots and other stats.
Presently the implemented version takes into account the NN
analyses used for Moriond99 results, handles hA/hZ overlap and also
the combination of analyses in the missing energy channel.
It includes the CL calculations from the likelyhood ratio estimator
using the FFT method by Hongbo.
Jason stated that the authors will include the cut analyses
and the standard Clevel calculation.discussion
- The Confidence Level is computed only with the FFT method
(presented before by Hongbo).
It is now publically available and an ALEPH note is promised emminently.
As it was stated in the previous meeting, the online analyses will
anyway provide the material for the ALEPH results presented to the
milestones (Summer conf, End of Run LEPC, Winter Confs) and will
be performed using the old clevel calculation following the
guideline which states "we need a back-up solution in anycase" .Only for clarification, because not directly related to the online
analyses, in case of winter Confs improvements can also be under
consideration if they are accepted (x-checks and approval) and the
deadline are met. The 20 November deadline was proposed and it is
still under consideration.Some discussion was raised to understand what amount of change
would be allowed for the online result w.r.t. the Moriond 99 version.
It is proposed by the convenors to limit changes to the absolute minimum
i.e. only Ecm, and in case or serious detector malfunction,
and to take systematic errors of last year for now.
The FFT method could be considered as
an improvement and will follow the above statements (decision
and approval). No such decision occurs during the meeting.
FFT was already used by their authors as an
alternative tool for different studies ( e.g. limit distribution and
elsewhere effect) where the increase of the speed of this
method presents an advantage. The FFT method is presently available
for general test and people are invited to test it as soon as
possible. Meanwhile it is proposed that
the standard CL will remain our main result.
Only pieces of discussion were exchanged for information; The
discussion is not closed and will continue next meeting.
- shapes of the discriminant variables distribution are needed.
Frederique and Pedro would accept to be in charge of the signal MC
production.
- more control plots will be included, b-tags ones for instance;
workers in each analyses are invited to indicate that they
estimate to be the most useful inputs.
It was noted that for the first part such a similar system was
elaborated last year by Marumi and was used during the last
data-taking . It was proposed to nominate for the responsibility
of the "online system" Jason and Marumi (if they accept).
Discovery exercise, using Hqq cuts
Cl curves in presence of signal lead to CL_sb^obs lower than
expected. CL_b^obs tends to be high throughout m_H but there is no
other information about the Higgs boson mass.
Pedro used three fake data samples, based on the Poisson
statistics to indicate the fraction of the various backgrounds and
signal. The real data set is the fourth data sample. In order to
know where is an excess, he constructed the significance of the
ratio data/MC per five GeV bin in m_H. The maximum value defines
the guessestimate of m_h. To cross-check these estimations of m_H
the Sandard Model Higgs cross-sections were used. An estimate of
m_His given by the Higgs boson mass which corresponds to the
equality between the number of observed data and the sum of
expected background and signal number of events. The two kind of
estimates are roughly compatible. Next step will be to combine the
fake data with similar ones for other channels.
It was suggested to plot the log(likelihood ratio) as function
of Mh as this gives a signal estimator AND a mass estimate.
A Few Developments in Hvv
Jennifer gave a brief description of the cocktail training for
an
11 variables NNet used by Wisconsin. The simulated
Higgs bosons
masses used are 85, 90 and 95 GeV. The training
and variables are
similar to the previous Wisconsin NNet except
that no
reconstructed mass was used in the training.
Alain commented that maybe one should establish
whether this is a
good thing to do and why and then stabilise
this point.
The optimisation was performed using the usual
receipe (80% qq and
WW subtraction, mass as discriminating variable).
The working point
obtained corresponds
to an efficiency of 35% and 5.6 expected events
from the
background processes (175.5pb-1). Jen came
to the comparison
between the cocktail and the Hvv combination.
Whatever one or two
discriminating variable are used, the cocktail
leads to similar
performances. The cocktail is better than
the combination only in
low Higgs boson masses while the combination
is better when a 95
GeV Higgs boson mass is assumed. With such
a hypothesis on the
Higgs boson mass, the CL is 18% for the combination
and 20.1% in
case of the cocktail. When two discriminating
variables are used
in the CL computation, a very small gain is
obtained with respect
to the only variable calculation (less than
10MeV and only for
sin2(b-a) values less than 0.1). The cocktail
training for low
masses will not help MSSM analyses. When the
hvv combination is
used, the improvement of the final limit is
around 100 MeV on the
hZ limit [sin2(b-a)=1.]