------------------------------------------------------------------------
Summary of Muon Layour working group meetings held on December 15 and 16
------------------------------------------------------------------------
(compiled by Burkhard)

1. Alison gave a short update on the matching studies she is pursuing.
The efficency for correctly matched tracks with a reasonable chi2^ cut
shows now the expected improved performance with the new layout (with
higher granularity in x in stations 2 and 3): 88% vs 93%.
Misidentification Probability: ~1.4%

2. Paul gave a very nice and detailed presentation with the goal to answer
most of the outstanding questions in the layout optimization.
a) He showed that higher granularity in R1 of stations 2+3 makes the system
more robust in case of high background (about 10-30% higher B->mu
eff. with 2-1% MB retention and backgound*4).
b) Further optimization of the FOI with high bkg. showed that
improvements are possible. However, with very high bkg (*6), the
optimal FOI for R1 in M1 would be 0, thus R1 obsolete.
Note: It became evident by the presentation of Andrei in the panel
that the bkg is lower, so we have not to consider the bkg*6 case
anymore.
c) He compared the new (default) layout with a layout without station 1
but infinite x-granularity in M2-M5. The B->mu eff. drops by 15-40%
for 2-1% MB ret.
He showed that we loose in all regions, but increasingly from R4 to
R1. Having infinite granularity, this loss is not due to reduced
Pt-resolution, but the missing FOI in station 1, which is a powerful
cut on MB events.
Using a multiple scattering cut on the comparison of the x-slopes of
station M2,M3 and M4,M5 helps to recover, in particular for 3% MB.
ret. However, when going to a more realistic layout (same
granularity in x in M4,M5 as in M2,M3 , the pt-resolution drops by
about 50% and the loss in performance even with MS-cut is still
12-32% with 2-1% MB ret.
Note: It is not at all evident that an additional MS-cut could be
applied within the allocated L0-latency. Moreover, a new DMP would
have to be designed if such a cut was required, which is very
diffcult, to say the least.
Moreover, the muon trigger looses dramatically in robustness
when dropping M1 (~50% less B->mu eff. with bkg.*4).
=> Therefore, we need certainly M1.
d) Next he looked whether M1 could be replaced by the SPD of the
calorimeter system. For this purpose, he adopted the SPD layout to
M1, assuming that the border between 4x4 blocks and 12x12 blocks is
between regions 2 and 3. The deterioration in performance, using the
default trigger algorithm is still around 16-23% for 2-1% MB. ret.
The losses are due to the reduced inner acceptance, the coarser
x-granularity (factor 1.5-4), and the lost y-projectivity.
=> Thus, in its present configuration, the layout of the SPD is not
adequate for the muon trigger. An additional discussion in the
afternoon with the calorimeter people showed that it is
essentially impossible to improve this situation.
e) Paul looked also in the y-granularity. A slightly coarser
granularity (*1.25) improves the symmetry of the whole system, which
is particularly useful in the trigger implementation. With such a
layout, e.g., the y-FOI in M4 and M5 becomes also 0.5, as in M2 and
M1, which simplifies the FIP-processor.
The robustness of such a layout against backgound is satisfactory up
to bkg*4.
f) The performance with small logical entities in region 2 has been
tested and good robustness against bkg has been found (we loose only 3%
B->mu eff. with bkg.*4).
A study with small logical entities in region 1 as well is still
ongoing. A preliminary conclusion is that we cannot have such
logical entities in M2-M5, while retaining the robustness of the
system. However, it is hoped that we can have them in stations 2 and
3, where we require now the improved x-granularity. Paul will report
on this in January.

3. Olivier compared the trigger performance he obtains with Paul's numbers
and found agreement within errors. A comparison of SICB 219 with 220
shows 10% less efficiency in 220. After the meeting the following
explanation came up, which seems to be reasonable:
MB events in v219 are of type 51, while in v220 they are of type 61,
thus with some pile-up events inside, using a different pythia
subprocesses! Apparently the definition of minimum bias changed, but
as usual it is not documented!

4. Renaud gave an update on the L0 trigger implementation. The new layout
provides good matching between sectors and logical entities, which was
not the case in the past. Detailed studies are planned in the coming
weeks.

5. In the discussion we agreed on the following (logical) granularity
obtained with pads and/or strips (all the numbers are normalized to
station M1):

M1 M2 M3 M4 M5
R1 1x2.5 1x2.5 1x2.5 2x2.5 2x2.5
R2 2x5 1x5 1x5 4x5 4x5
R3 4x10 2x10 2x10 8x10 8x10
R4 8x20 4x20 4x20 16x20 16x20

In R1 we still assume logical pads everywhere. Only when it can be shown
that we can cope with logical entities in some stations at least, e.g. 2
and 3, will we reduce the pad size there to 0.5x2.5.
In R2 we assume pads in M1 and logical entities of 12x20 in M2-M5.
In R3 we assume pads in M1 and logical entities of 48x40 in M2-M5.
In R4 we assume pads in M1 and logical entities of 96x80 in M2-M5.

From the point of view of physical layers, we agreed that a proposed detector
implementation should contain:

- one double layer (in OR) in M1,
- two ORed double layers in M2 and M3 which are combined by a logical AND
(or a 2 out of 3 majority logic in case of RPCs). The reason for the AND
in M2 is the higher background there compared to the other stations,
while in M3 we want to start the trigger algorithm with a good seed.
- one ORed double layer in M4 and M5.